1.1 Cosmology, Astrophysics, and Astronomy

Cosmology, astrophysics and astronomy all investigate the universe beyond the boundaries of the solar-system, but with different focuses. While the boundaries between the three disciplines are blurred, a rough delineation of the three will help us to get clearer on the current subject matter (see also Jacquart, Reference Jacquart2020, fn. 1).

Cosmology studies the origin and evolution of the universe at the largest scales. It describes how the universe evolved from an initial hot dense state until today, approximately 13.8 billion years later, and how, during that time, the large-scale structure in the universe arose from the formation of the lightest elements to the development of the ‘cosmic web’, a structure of galaxy clusters and filaments connecting them. As a discipline, relativistic cosmology is approximately 100 years old. Einstein (Reference Einstein1917) first applied general relativity to the universe as a whole. However, for the first fifty years or so, the discipline remained largely theoretical, with a limited empirical basis. This changed in 1965 with the discovery of the Cosmic Microwave Background radiation (CMB), a remnant of the Big Bang that provides a map of our universe when it was very young. Today, cosmologists use large-scale structure surveys, detailed data from the CMB and various other observational probes to test their models of the universe at the largest scales.

Astrophysics describes the evolution of stars, galaxies, galaxy clusters, black holes and other structures in the universe, using the laws of physics. As a discipline, it originated in the nineteenth century, when spectroscopy was first applied to the Sun (Boyd, Reference Boyd, Boyd, Baerdemaeker, Heng and Matarese2023, 15-16). This work revealed for the first time the applicability of nuclear physics to stellar interiors. Today, astrophysics is extremely broad in scope. It covers an enormous range of objects and scales – from individual molecules in the inter-stellar medium to galaxy clusters. Its primary aim is to offer causal explanations for observations of astrophysical objects. Unlike cosmology, it is not just interested in unique objects, but also in generalizations about types of objects (instead of token examples), from main-sequence stars to spiral galaxies (Anderl, Reference Anderl and Humphreys2016).

Astronomy is primarily a cartographic exercise, mapping out different structures in the sky. It does not primarily concern itself with offering causal explanations based on, for instance, what happens in stellar interiors, although it does offer the observational basis for much of cosmology and astrophysics. It is one of the oldest scientific disciplines, dating back to prehistoric times. Today, we have entered the era of multi-messenger astronomy. Astronomers are no longer confined to a single type of signal, for example, electromagnetic waves in the optical spectrum. Instead, astronomers now study a very broad range of the electromagnetic spectrum, but they also study neutrino signals, gravitational waves, and cosmic rays (Abelson, Reference Abelson2022b).

The three fields don’t just differ in subject matter, but also in terms of their philosophical investigation. While cosmology has received fairly consistent attention from philosophers over the last couple of decades, philosophy of astrophysics is a much younger field. And although it is possible that some papers in philosophy of cosmology could qualify as philosophy of astrophysics (e.g., some work in philosophy of dark matter), that is not going to be the case for all. In reality, philosophy of astrophysics and philosophy of astronomy so far, even more so than philosophy of cosmology, remain an untapped mine for philosophy of science.

1.2 Some Scientific Background

Before delving into the philosophical debates, it will be useful to provide some scientific background. The examples of novel physics discussed in Sections 2–4 are deeply connected to $\mathrm{\Lambda }$ CDM (“Lambda Cold Dark Matter”). the current concordance model of cosmology. $\mathrm{\Lambda }$ CDM uses general relativity (GR) in combination with assumptions that the universe is homogeneous (“no special points in spacetime”) and isotropic (“no preferred directions”) to describe the evolution of the universe. The combined assumptions of homogeneity and isotropy are known as the ‘Cosmological Principle’ (see Section 6.2.3).

Given the Cosmological Principle, which also implies that the stress-energy tensor is that of a perfect fluid, the universe can be described by an FLRW-metric:

$d{s}^2=-d{t}^2+a(t)d{\sigma }^2$(1)

a(t) denotes the (dimensionless) scale factor. It is related to the Hubble parameter, which measures the expansion rate of the universe:

$H(t)\equiv \frac{\dot{a}(t)}{a(t)}$(2)

$H_0$ is the current value for the Hubble parameter. The evolution of the scale factor is described by the Einstein Field Equations, which, thanks to the homogeneity and isotropy assumptions, simplify to the Friedman equations:

$\begin{array}{rl}{\left(\frac{\dot{a}}{a}\right)}^2& =\frac{8\pi G\rho }{3}-\frac{k}{a^2}+\frac{\mathrm{\Lambda }}{3}\end{array}$(3)

$\begin{array}{rl}\frac{3\ddot{a}}{a}& =-4\pi G(\rho +3p)+\mathrm{\Lambda }\end{array}$(4)

In these equations, $\mathrm{\Lambda }$ is the cosmological constant. The curvature is represented by k. $k=1$ represents a closed universe, $k=-1$ a hyperbolic universe, and $k=0$ a flat universe, which occurs when the energy density of the universe equals the critical density ${\rho }_{\mathit{\text{crit}}}=\frac{3H^2}{8\pi G}$. $\rho$ and p are the usual density and pressure.

Different contributions to the energy budget of the universe have a different effect on the evolution of the scale factor. This is reflected in their equations of state, that is, relations between pressure and density. The equation-of-state parameter w, defined in the barotropic equation of state as $p=w\rho$, can be used to track how different energy density contributions influence the evolution of the scale factor in the Friedman equations:

$\rho \propto a^{-3(1+w)}$(5)

The most important equations of state are those for radiation (neutrinos and photons; $w=1/3$), matter (baryonic and dark, assumed to be collisionless; $w=0$), and dark energy ($w=-1$).

FLRW-models offer a first-order description of the evolution of the universe at the largest scales. In order to account for large-scale structure formation, small near-scale-invariant Gaussian primordial density perturbations are introduced. These are detected in the CMB as perturbations of the order of ${10}^{-5}$, but they form the seeds of what through gravitational collapse become galaxies, clusters, and super-clusters. One of the central questions of early-universe cosmology is what formed these primordial density perturbations. The leading answer since the 1980s comes from cosmic inflation (discussed in Section 2).

Over time, it has become clear that in order to describe the evolution of large-scale structure in the universe, the universe must contain more than baryonic matter alone, that is, matter consisting of three quarks, like protons or neutrons. Most matter that humans encounter on a day-to-day basis is baryonic. However, the current energy density budget of the universe also contains approximately 27% dark matter and 69% dark energy, in addition to the 5% baryonic matter (Planck Collaboration, Aghanim, et al. Reference Collaboration, Aghanim and Akrami2020). This is reflected in the name of the current cosmological concordance model: $\mathrm{\Lambda }$ refers to dark energy, CDM to (cold) dark matter. Dark matter (see Section 3) behaves as an additional source of gravitational collapse, while dark energy (see Section 4) is a placeholder term for what causes the accelerated expansion of the universe. Current observations are compatible with the total energy density of the universe equalling the critical density, suggesting that the universe is flat.

The success of $\mathrm{\Lambda }$ CDM as a theory of large-scale structure formation is remarkable. The predictions derived through computer simulations (more on this in Section 6) agree with observations of large-scale structure. Further evidence comes from observations of element abundances corresponding to predictions based on Big Bang Nucleosynthesis, the theory of the formation of the lightest elements in the early universe. It is even more remarkable how quickly this success was achieved: the fact that the universe is larger than the Milky Way was only broadly confirmed in 1925, and the fact that it is expanding became relatively established in the late 1920s and early 1930s (De Baerdemaeker & Schneider, Reference De Baerdemaeker and Schneider2022). The first unification between cosmological and astrophysical missing mass (dark matter) did not occur until the 1970s (de Swart, Reference de Swart, Blum, Lalli and Renn2020, Reference de Swart2022; de Swart, Bertone, & van Dongen, Reference de Swart, Bertone and van Dongen2017), cosmic inflation was first proposed in the 1970s and 1980s (Guth, Reference Guth1981), and the supernova evidence for the accelerated expansion of the universe was discovered in the 1990s.

Arguably the most important source of empirical evidence for cosmology today is the CMB. The CMB provides a snapshot of the universe approximately 380,000 years after the Big Bang. It was formed at recombination, when atomic nuclei and electrons formed neutral atoms, and photons became free-streaming. Since recombination, these photons have been free-streaming throughout the universe. Although shifted to the red because of the universe’s expansion, their spectrum has remained unchanged. Penzias and Wilson discovered the CMB in 1965 as an ineliminable background noise in their radio antenna. At the time, this evidence played a crucial role in establishing Big Bang Cosmology over the rival steady-state theory (although see Ćirković & Perović Reference Ćirković and Perović2018, for nuances to the common narrative). In the 1990s, the COBE satellite allowed scientists to establish for the first time that the CMB had a perfect black-body spectrum. Follow-up satellites WMAP’s and Planck’s increased sensitivity allowed cosmologists to detect the slight anisotropies in the CMB, and to map out their full power spectrum. Today, the precise values of the six free parameters of the $\mathrm{\Lambda }$ CDM model (aside from the dark matter density and the baryonic matter density also the age of the universe, the scalar spectral index, the curvature fluctuation amplitude, and the reionization optical depth) are determined by the Planck results from the 2010s (Planck Collaboration, Aghanim, et al. Reference Collaboration, Aghanim and Akrami2020).

Given the recent nature of most of what, today, is considered ‘established knowledge in cosmology’, it should come as no surprise that there are many open questions. Some of them will return in the next sections. What is the nature of dark matter? What causes the accelerated expansion of the universe? And, the focus of the next section, whence the primordial density fluctuations revealed by the CMB?

2.1 The Early Case for Inflation: Fine-Tuning Problems

The original motivation behind inflation was to resolve various fine-tuning problems within our universe (Guth, Reference Guth1981): the flatness problem, the horizon problem, and the magnetic monopole problemFootnote ⁴. These problems refer to peculiar features of our universe, that, although not in tension with any physical theory, are so striking that they seem to demand an explanation.

As McCoy (Reference McCoy2015) highlights, while it is easy to give a compelling gloss for the flatness and horizon problems, it’s much harder to precisely pin down what the explanatory gap exactly is that inflation supposedly fills. For space reasons, I restrict the discussion here to McCoy’s intuitively compelling gloss. Readers interested in a more detailed discussion of the fine-tuning problems themselves should turn to McCoy’s paper and references therein.

The flatness problem refers to the fact that the current spatial curvature of the universe appears to be (nearly) zero. However, zero curvature is an unstable point: even the slightest deviation from flatness would lead to the universe evolving to ever-increasing deviations from flatness. Given its current value, the curvature in the early universe must have been even closer to zero. Inflation would solve this problem because the exponential expansion of space would smooth out any curvature that was present before the inflationary epoch. It would basically reset the universe to close-to-flatness.

The horizon problem refers to the remarkable uniformity of the universe as observed in the CMB. This uniformity suggests that the entire observable universe must have been in causal contact to generate this uniformity (e.g., by reaching some thermal equilibrium) before the decoupling of the CMB. Any such process would have required the different regions of the observable universe to be causally connected before recombination. Here’s the catch: this is in direct tension with the fact that certain regions of the observable universe have not been in causal contact with one another according to the regular expanding-universe models. Due to the finite speed of light, they fall outside of each other’s particle horizons. Without assuming some novel process in the early universe that allowed these regions to be in causal contact, the homogeneity of the CMB could only be explained by uniform initial conditions across the observable universe.

Inflation offers such a novel process. The exponential expansion of space during the inflationary epoch would cause nonuniform regions of space to be stretched beyond the particle horizon scales. The different regions of the CMB were therefore in causal contact before the inflationary epoch. Specifically, they were part of a single region that was in equilibrium before inflation began, and that was stretched out beyond the horizon during inflation. Different parts of that region keep re-entering the horizon after reheating, as the universe continues to expand.

Regardless of the specifics of the fine-tuning problems, one could wonder if a new physical process is necessary to account for them. Couldn’t they be solved by referring to the initial conditions of the very early universe? The explanation then is that the universe appears fine-tuned because it had peculiar initial conditions that were evolved forward. After all, many standard explanations in physics combine initial conditions with laws of nature to explain an observed physical state of a system (Earman & Mosterin, Reference Earman and Mosterin1999; McCoy, Reference McCoy2015).

In cosmology, however, referring to the initial conditions of the universe tends to be unsatisfactory. Pragmatically, initial conditions are something of a dead-end for research (the importance of this pragmatic consideration in theory choice is also highlighted by Wolf & Thébault Reference Wolf and Thébault2023). They amount to brute facts that need to be accepted at face value. But the early universe is one of the only regimes where we might find observable traces of Planck-scale physics or theories of quantum gravity (Schneider, Reference Schneider2021, Reference Schneider2023). Resorting to initial conditions thus amounts to shutting down an entire research avenue. Relatedly, there is a difference between the initial conditions of the universe and other physical systems: for other physical systems, we can usually explain where the initial conditions come from (‘the ball is at the top of the hill at rest because I put it there’). There is no analogue for this in case of the universe – so what explains those initial conditions? Earman and Mosterin (Reference Earman and Mosterin1999) therefore identify the fact that inflation allowed scientists to ask why-questions about the early universe as one of the sociological factors leading to its rapid spread. Finally, the initial-conditions route is incomplete because it does not reveal where the initial conditions are supposed to be imposed. At the Planck scale? Earlier? Later? In light of all of this, a dynamical explanation that makes the puzzling features (or the initial conditions) generic (in a way to be specified later) would be at least pragmatically preferable. Inflation purports to offer precisely such a dynamical explanation.

However, pragmatic reasons are not always accepted as grounds for epistemic confirmation. If we reject the initial-conditions explanation based on pragmatic grounds, does that mean that inflation can be epistemically confirmed based on its resolution of fine-tuning problems? There are some reasons to be doubtful.

The fine-tuning problems consist of features of the observable universe that are not prohibited by or in tension with the standard model of cosmology, but that nonetheless strike physicists as ‘unlikely’, ‘surprising’ or ‘improbable’ if their explanation is restricted to specific initial conditions. In other words, there is no strict empirical anomaly that inflation purports to solve here. Rather, the observed flatness and uniformity of the universe require very specific initial conditions, and physicists consider it unlikely that our universe would have exactly those initial conditions out of the much broader set of all possible initial conditions. Inflation supposedly solves the fine-tuning problems by offering a dynamical explanation: the dynamics of inflation are such that the universe, regardless of its initial conditions before inflation took place, would generically lead to a universe with the observed ‘surprising’ features of our universe.

While intuitively compelling, there are some issues. First, there is the open question whether inflation suffers from an initial conditions problem itself. It is not clear whether all models of inflation can get triggered regardless of the preceding initial conditions (Brandenberger, Reference Brandenberger2017). Second, McCoy (Reference McCoy2015) points out that moving beyond the intuition to pinpoint where the exact problem lies is a difficult task, but this is essential in order for inflation to gain confirmation from solving the fine-tuning problems. To properly consider the fine-tuning problems as ‘problems’, two questions need to be answered: (i) why are these initial conditions special?; and, (ii) why are such special initial conditions problematic? McCoy finds it plausible that for the flatness and horizon problems, (i) could be answered. The horizon problem could be considered ‘special’ because it requires some maximal degree of symmetry, for example, while the flatness problem requires a dynamical instability as an initial condition.

But just pointing out that the initial conditions of the universe are special is insufficient for them to constitute a problem. Sometimes systems just have peculiar initial conditions. As long as these are not in tension with the applicable theory, it’s unclear that there is a problem to be solved. What would that require? McCoy sees two possibilities. One reason would be that these special initial conditions are somehow improbable. Making this case requires defining a physically meaningful probability measure (roughly, a function that assigns probability values to all members of a set and with a total value of 1) over all initial conditions. This hasn’t been done yet: both defining the space of possibilities, and defining a meaningful probability measure over that space turn out to be nontrivial (see, e.g., McCoy Reference McCoy2017, for a critical take on one candidate measure). At most, this route leads to a promise of a problem, not a formulation of the actual problem, and the road in between is riddled with obstacles.

Another reason to consider the fine-tuning problems genuine anomalies is that the initial conditions are somehow not explanatory in themselves. For instance, Wolf and Thébault (Reference Wolf and Thébault2023) argue that the problematic nature of initial conditions can be cashed out in terms of explanatory depth. But here, McCoy (Reference McCoy2015) points out that it is not obvious that special initial conditions are by default not explanatory. As mentioned before, multiple philosophical accounts of explanation, including the deductive-nomological or various causal ones, explicitly include initial conditions in the explanans. It’s not clear why the ‘specialness’ of the initial conditions would make them any less explanatory. Special initial conditions may serve as a guide for future research, but a theory making them generic doesn’t imply confirmation.Footnote ⁵

What does this imply for the status of (classic) inflation? Not much, in the sense that inflation is still consistent with the fine-tuned features of the observable universe. In other words, it doesn’t mean that inflation is now a theory in crisis, faced with anomalies. It does mean, however, that while fine-tuning may have been a fine pragmatic motivation, inflation solving fine-tuning problems does not necessarily confer any confirmation of the theory. This is where structure formation comes in.

2.2 The Current Case for Inflation: Structure Formation

Immediately after Guth’s seminal paper, a new, much stronger argument for inflation arose. In the 1970s, cosmologists had convened on a phenomenological model for the spectrum of initial density perturbations that provided the seeds for the observed large-scale structure. The Harrison-Peebles-Zel’dovich spectrum is one of small, adiabatic, near-scale-invariant, slightly red-tilted, Gaussian perturbations. But the origin of these perturbations remained unknown. Scientists soon realized that quantum fluctuations in the inflaton field could produce a spectrum with the required properties (Guth & Pi, Reference Guth and Pi1982).

The CMB prediction is still considered the strongest argument in favor of inflation. While the power spectrum, like the horizon and flatness problems, ultimately is an initial-conditions problem, it does provide a more severe test for the predictions of inflation due to the more specific properties of the power spectrum. This empirical success has also been important for the further development of inflation as a theory: as more details about the primordial density perturbations are revealed by better CMB observations, further constraints are placed on the nature of the inflaton field and the shape of the inflaton potential (Planck Collaboration, Akrami, et al. Reference Collaboration, Akrami and Arroja2020).

Yet, despite this new case in favor of inflation, controversy remains and rivaling theories for the early universe are not counted out. One reason is that inflation’s prediction of the CMB power spectrum suffers from the so-called ‘trans-Planckian problem’, first introduced by Martin and Brandenberger (Reference Martin and Brandenberger2001). The trans-Planckian problem boils down to a failure of separation of scales, such that modes at and beyond the Planck scale become relevant for the prediction of the power spectrum.Footnote ⁶

Another reason is that there is disagreement about what requirements should be imposed on a theory of the early universe and how well inflation satisfies them. As we will see next, many of the discussions implicitly or explicitly draw on Popper’s work on the demarcation problem, according to which falsifiability is a necessary property of any scientific theory. Moreover, Popper admonished the use of ad hoc changes to scientific theories in order to accommodate recalcitrant data. It turns out, however, that the line between novel prediction and mere accommodation is difficult to draw in context of inflationary cosmology.

The disagreement about the scientific status of inflation came to a head in 2017, when in the pages of Scientific American two camps were battling about the status of inflation as a scientific hypothesis (Guth et al., Reference Guth, Kaiser and Linde2017; Ijjas, Steinhardt, & Loeb, Reference Ijjas, Steinhardt and Loeb2017).Footnote ⁷ Several charges were brought against inflation, the most scathing one that inflation is not falsifiable (or generally untestable) and therefore not scientific. Thus, despite its empirical success, it is worthwhile to examine the nature of inflation as a scientific theory in more detail. What grounds this criticism of unfalsifiability?

2.3 Multiverse Realism

To conclude the discussion of early-universe cosmology, I want to briefly turn to scientific realism and inflation. Scientific realism in general is a commitment to the (approximate) truth of our best scientific theories, and therefore also to the existence of the observables and unobservables they posit (Chakravartty, Reference Chakravartty and Edward2017). The realism question about the inflationary multiverse has not been discussed in the philosophical literature so far, but that should not be surprising: it may seem preemptive to discuss the metaphysical implications of inflation before the theory itself is developed and further confirmed. And indeed, given the preceding analysis, it seems premature to commit to belief in the existence of the inflationary multiverse, regardless of one’s tendencies elsewhere. Nonetheless, it is worthwhile to look at inflation as a peculiar test case for scientific realism.

Let us suppose that a smoking gun is found in favor of inflation and the scientific consensus grows further that inflation is indeed a successful theory of the early universe. What does this imply for the scientific realist? Is she now committed to accepting the reality of the inflationary multiverse? In other words, is she now committed to accepting that, as Guth put it, anything that can happen will happen an infinite number of times?

Let me explain why inflation presents a puzzle for the realist debate.Footnote ¹⁰ Any empirical confirmation of inflation would necessarily come from observations within our universe. Inflation generically implies eternal inflation, according to which our universe is just one in an infinite multiverse, but where we are causally closed off from every other pocket universe. This is what makes the case peculiar: the theory explicitly precludes any causal chain that bridges between other pocket universes and our own universe. That makes it different from, say, subatomic particles: while those are inaccessible to our naked eyes because of the associated length scales, there is a realist case to be made that we can track the causal effects of these subatomic particles to observed detector signals. It also makes it different from extinct organisms, which are empirically inaccessible because of the associated time scales. Based on the fossil record, it is still possible to construct a causal inference from the fossil record to the now-extinct organism that would suffice to warrant a realist commitment. In contrast, CMB observations may permit a causal inference from primordial gravitational waves to the once-existence of an inflaton field and its associated inflationary phase in the early universe. But does commitment to an inflationary phase in the evolution of our universe automatically imply a commitment to the existence of the inflationary multiverse – a different causal effect of the inflaton field, distinct from the primordial gravitational waves?

It seems to me that several versions of selective scientific realism would be hesitant about this inference. These views are only committed to realism about specific parts of our current best scientific theories, thus leaving room for theory change without this theory change affecting the realist commitment. For example, the explanationist realist along the lines of Psillos (Reference Psillos1999) is committed to the existence of those entities that are indispensable to explaining the predictive success of a scientific theory. It is not immediately clear whether the multiverse is indispensable in such a way, since it seems more like an unwanted side-effect than an essential component of inflation. Entity realists (e.g., Hacking Reference Hacking1983) are committed to realism about entities to the extent that they can manipulate them to bring about other effects. They would likely be hesitant about commitment to the multiverse because of the lack of causal access to it. These examples suggest that commitment to the multiverse is not a necessary consequence of a broader realist commitment to inflation.

3.1 The Case for Dark Matter

At galaxy scales, Rubin’s observations of flat galaxy rotation curves were ground-breaking (Rubin & Ford, Reference Rubin and Ford1970). Assuming the observed mass distribution of luminous matter, Newtonian dynamics (the applicable approximation to GR on galaxy scales) predicts that the rotational velocity of stars will fall off with increasing distance from the galactic center. But Rubin observed that the velocity of stars remained more or less constant, regardless of their distance from the center. The rotation curves, which plot rotational velocity in function of distance from the galactic center, were therefore called ‘flat’. One way of accounting for these observations is by modifying the assumed mass distribution in the galaxy: instead of only taking into account the luminous matter, one could introduce an additional, nonluminous mass source, that is, dark matter.

At cluster scales, both velocity dispersions and gravitational lensing provide evidence for the presence of additional nonluminous matter. In the 1930s, Zwicky (Reference Zwicky2009/1933) determined the dispersion of galaxy velocities in the Coma Cluster, that is, the spread of velocities of galaxies around the mean. Zwicky found that the dispersion was large, too large for the Coma Cluster to be held together by the gravitational pull of the luminous matter alone. By assuming the presence of nonluminous matter, the stability of the Coma Cluster as a gravitational system can be accounted for, despite the large velocity dispersion.

Much more recently, observations of the Bullet Cluster were touted in the paper title (!) as “a direct empirical proof” for the existence of dark matter (Clowe et al., Reference Clowe, Bradac and Gonzalez2006). The Bullet Cluster is actually a collision event between two galaxy clusters, where one resembles a bullet moving through the other – hence the name. The observers used X-ray observations to determine the distribution of luminous matter. They mapped out the gravitational mass distribution using strong gravitational lensing of background galaxies by the Bullet Cluster. The two did not line up with one another. Clowe et al. concluded that there must be some additional, nonluminous gravitating matter present to account for this misalignment.Footnote ¹¹

Finally, at cosmic scales, the success of $\mathrm{\Lambda }$ CDM as a model of large-scale structure formation crucially depends on positing the existence of cold dark matter. One line of evidence comes from the CMB. The CMB power spectrum is differently sensitive to the presence of dark matter and baryonic matter. The third peak in particular puts tight constraints on the dark matter density ${\mathrm{\Omega }}_{DM}$ (Planck Collaboration, Aghanim, et al. Reference Collaboration, Aghanim and Akrami2020). Other evidence comes from Baryonic Acoustic Oscillations (BAOs), that is, oscillations in the matter density in the universe as a result of counteracting influence of gravitational collapse and outward radiation pressure (Eisenstein et al., Reference Eisenstein, Zehavi and Hogg2005). They are detectable as a preferential length scale at which galaxies are separated from one another. Because dark matter is only subject to gravitational collapse and not to radiation pressure, it contributes to stronger galaxy formation than what would be the result of baryonic matter alone. Similar lines of evidence also come from Big Bang Nucleosynthesis and from large-scale structure formation. In all cases, explaining the observations requires more gravitating and otherwise minimally interacting matter present than what baryonic matter can account for.

The cosmological and astrophysical case for dark matter is clearly strong. Nonetheless, many candidates remain as far as dark matter constituents are concerned, ranging from primordial black holes over supersymmetric partners to axions. Many hope that dark matter will provide a window on physics Beyond the Standard Model of particle physics, but as of yet, this has not been substantiated. This is not just a scientific problem, but also a philosophical one. Epistemically, there is a question of how to make progress on dark matter research, given that it constitutes genuinely novel physics (Section 3.3). Metaphysically, it is not obvious whether the scientific realist should be committed to the existence of dark matter (Section 3.4). But first, let me turn to a different question, one that the majority of philosophy of dark matter has focused on so far: is there a genuine underdetermination between positing the existence of dark matter and modifying Newtonian dynamics?

3.2 A Peculiar Philosophical Focus

The current scientific consensus is that dark matter makes up approximately 27% of the current energy density of the universe and that there generally is about five times as much nonbaryonic cold dark matter in the universe as there is baryonic matter. Nonetheless, a significant part of the philosophical literature on dark matter centers around whether or not there is genuine underdetermination between two scientific hypotheses: dark matter and MOND – short for Modified Newtonian DynamicsFootnote ¹² (see, e.g., Jacquart Reference Jacquart2021; Martens et al. Reference Martens, Carretero Sahuquillo, Scholz, Lehmkuhl and Krämer2022; Massimi Reference Massimi2018). Indeed, some papers go as far as suggesting that MOND is ‘preferable’ over dark matter, counter to the scientific consensus (see, e.g., McGaugh Reference McGaugh2015; Merritt Reference Merritt2017, Reference Merritt2020, Reference Merritt2021; Milgrom Reference Milgrom2020). This raises an obvious question: do the philosophical arguments in favor of MOND hold? Answering this question affirmatively could have significant implications: it would imply that the strong scientific consensus is misguided.

MOND was developed in the early 1980s by Milgrom. To understand its original appeal, it’s important to note that the empirical case for dark matter was not as strong then as it is today: several of the key observations that are now rehearsed as evidence for dark matter date post-2000. In fact, the dark matter hypothesis itself only gained traction in the 1970s. de Swart, Bertone, and van Dongen (Reference de Swart, Bertone and van Dongen2017) (see also de Swart Reference de Swart2022) argue that a crucial step in the acceptance of dark matter came in 1974, when two papers were published back-to-back on large-scale structure formation (Einasto, Kaasik, and Saar Reference Einasto, Kaasik and Saar1974; Ostriker, Peebles, and Yahil Reference Ostriker, Peebles and Yahil1974). For the first time, these papers brought together observations from large-scale structure formation, velocity dispersions in clusters, and galaxy rotation curves to argue for the existence of dark matter. The observational evidence was complemented with a theoretical argument: dark matter was necessary to ‘close the universe’, and a closed universe was at the time preferred on aesthetic grounds. CMB evidence or the Bullet Cluster observations were not available in the 1970s or 1980s. This means that although there was already a good case for cold dark matter in the 1980s, when Milgrom proposed MOND, the case was not quite closed.

Milgrom proposed MOND largely in response to the anomalous flat rotation curves (Bekenstein & Milgrom, Reference Bekenstein and Milgrom1984; Milgrom, Reference Milgrom1983a, Reference Milgrom1983b, Reference Milgrom1983c). Recall that dark matter accounted for the flat rotation curves by changing the assumed mass distribution. Milgrom’s proposal was to keep the mass distribution intact, but to change the governing dynamical laws. In other words, Newtonian dynamics was modified on galactic scales – hence the name MOND.

Today, MOND has largely been discredited by the majority of scientists. There are multiple reasons. First, as discussed before, the empirical case for dark matter is significantly stronger today than it was in the 1980s. In contrast, while MOND has been successful describing phenomenological correlations on galaxy scales, it struggles to account for phenomena on cluster and cosmological scales. This is related to the second reason: there is no broadly accepted relativistic extension of MOND that can replace general relativity and its empirical successes.Footnote ¹³ Worse still, MOND requires changing tried and tested dynamical laws at one, seemingly arbitrary scale. Meanwhile, dark matter requires no modification of fundamental theories of physics and it is compatible with the presumed incompleteness of the Standard Model of particle physics.

But despite these scientific arguments, several philosophical defenses of MOND have been mounted. Richard Dawid and I have summarized and critically evaluated the three most common types of argument (De Baerdemaeker & Dawid, Reference De Baerdemaeker and Dawid2022). The rest of this section draws on that analysis.

First, defenders of MOND appeal to Popper’s demarcation criterion of falsifiability. They argue that dark matter as part of $\mathrm{\Lambda }$ CDM is unscientific because $\mathrm{\Lambda }$ CDM is unfalsifiable. This is because predictions from $\mathrm{\Lambda }$ CDM have to be derived through complex computer simulations. Especially for galaxy- or cluster-scales, these simulations tend to include a large swath of free parameters to model nongravitational interactions (at the largest scales, gravity is dominant, significantly reducing the required complexity). As a result, defenders of MOND claim that just about anything can be predicted with those simulations, or, in other words, that these simulations are unfalsifiable.

Second, defenders of MOND argue that the dark matter hypothesis itself is unfalsifiable as long as there is no specific dark matter candidate identified as the sole constituent of dark matter. Without such a specification of ‘the’ dark matter candidate, any failure to detect a given dark matter candidate with a targeted experiment can just be explained away by arguing that the searched-for candidate does not in fact constitute dark matter. For example, the lack of empirical success for supersymmetry has put pressure on the idea that the lightest supersymmetric partner is a good dark matter candidate, but the dark matter hypothesis itself has not been affected by this result.

Finally, defenders of MOND offer a positive argument in favor of MOND. This positive argument appeals to Lakatos’ view on theory change. According to Lakatos, scientific research programs are progressive as long as they successfully make novel predictions, including after modifications to account for previously recalcitrant evidence. They are degenerative if the modifications that are required to accommodate new observations do not lead to novel predictions. Defenders of MOND argue that MOND, unlike dark matter, has made many risky predictions for galaxy-scale phenomenology, for example the Baryonic Tully-Fisher relation or the Radial Acceleration Relation. These predictions have all been borne out by observations. Thus, MOND is a progressive research program: based on a small set of assumptions, it has successfully made several novel predictions.

There are questions to be raised about the scientific basis for these arguments. But even setting those aside, we show that there is a basic philosophical issue with framing any defense of MOND in terms of Lakatosian or Popperian views on theory assessment (De Baerdemaeker & Dawid, Reference De Baerdemaeker and Dawid2022, §4). Defenders of MOND aim to make an epistemic claim that MOND is confirmed by its predictive successes and furthermore, that MOND is additionally confirmed by the failure of its only rival, dark matter. This stands in clear tension with the philosophical views they employ. Popper’s claims about falsifiability were normative: scientists should not construct unfalsifiable theories. Popper found any notion of credence or confirmation of a scientific theory misguided – but this is exactly what defenders of MOND aim to establish. And Lakatos’ work should be understood as a descriptive analysis of the history of science. In other words, it is backward-looking. Lakatos gave no criteria for determining whether an ongoing research program is progressive or degenerative. Using either view for epistemic arguments is therefore a category mistake.Footnote ¹⁴

I therefore conclude that the philosophical arguments in favor of MOND are so far unconvincing, in line with the scientific consensus. This also raises a methodological concern about philosophy of dark matter so far. From the perspective of philosophy of science in practice, I find the focus on a debate that is largely absent from the scientific literature to be a missed opportunity. To be sure, there is interesting conceptual work to be done in comparing the conceptual basis of MOND and dark matter (see, e.g., Martens & Lehmkuhl Reference Martens and Lehmkuhl2020a, Reference Martens and Lehmkuhl2020b), and the debate can be a useful test case for philosophical views on modeling and theories (see, e.g., Jacquart Reference Jacquart2021; Massimi Reference Massimi2018). Explicitly arguing counter to the scientific consensus that there is genuine underdetermination or reason to accept a rival theory, however, comes with a high burden of proof – one that hasn’t been met so far. And perhaps it is worthwhile for philosophers interested in dark matter to spend their efforts elsewhere, given how much of dark matter research is as of yet unexplored. The remainder of this section can hopefully give a nudge in that direction.

3.3 Searching for Dark Matter

Perhaps the central question in contemporary dark matter research is: what constitutes dark matter? This question presents a unique challenge. Despite the strong astrophysical and cosmological evidence for dark matter as well as its gravitational behavior (a perfect collisionless fluid), there is very little guidance as to its nongravitational behavior. As Weatherall (Reference Weatherall2021) points out, that may be because, if dark matter is constituted by primordial black holes, we already know everything there is to know about it. Nevertheless, the question stands: how can we know that dark matter is constituted by primordial black holes, or axions, or any other candidate? In other words, how can scientists make progress on learning more about dark matter’s nongravitational behavior (including, in a worst-case scenario, that it only interacts gravitationally), given that there are so little constraints on that nongravitational behavior?

Given that many dark matter search experiments are underway, making such progress must not be hopeless. Some searches are so-called production experiments: they look for signatures of produced (hence the name) dark matter particles at high-energy colliders like the LHC. Others are direct detection searches with enormous detector vats. They aim to detect a scattering event where a dark matter particle from the Milky Way halo deposits some energy to a waiting atomic nucleus, similar to how solar neutrinos were first discovered. Still others are indirect detection experiments. Unlike direct detection experiments, they don’t look for signatures of dark matter scattering off a standard-model particle. Rather, they appeal to the fact that if dark matter is a particle, it is possible that in dark-matter-rich environments (e.g., in the galactic center), dark matter particles might annihilate into standard-model particles, resulting in a detectable signal. Indirect searches look for those signals.Footnote ^15, Footnote ¹⁶

Interestingly enough, many experiments searching for dark matter candidates tend to use tried-and-tested technology from high-energy physics. Particle colliders had long proven their success before they explicitly started to focus on dark matter searches. Direct detection experiments were explicitly modeled on the detection methods for solar neutrinos (De Baerdemaeker, Reference De Baerdemaeker2021, 136-139). This is quite peculiar, since one of the few things that is broadly accepted about what constitutes dark matter is that it cannot be standard model particle physics. So how can these experiments legitimately claim to be searching for dark matter?

I have argued that the structure of justification for new physics searches takes a different form than most other experiments (De Baerdemaeker, Reference De Baerdemaeker2021). Let me illustrate by contrasting the justification for dark matter experiments with a more typical example: MRI research. Scientists use MRIs to study brain structure and other soft tissues in the human body. MRIs use strong magnetic fields that interact with protons in the tissue of interest. Based on differences in alignment times with the magnetic field and in energy releases, scientists can differentiate between different types of tissues.Footnote ¹⁷ Why do scientists believe that MRI scans are an effective way to map out brain structure? Because they know that the human body consists of different types of tissue and those different types of tissue are constituted by atoms which contain protons. That’s it. That very basic knowledge about the human body is sufficient to establish that MRI scans are an appropriate method to detect different brain structures, tumors, or injuries.

The problem with setting up dark matter search experiments is that that very basic knowledge, the analogue of the fact that the human body contains protons, is largely lacking. After all, there are no established properties of dark matter that can give scientists guidance as to which methods may be able to detect the nongravitational behavior of dark matter particles. All we have is gravity, and limits on dark matter’s nongravitational behavior.

So how are these dark matter searches justified? The justification follows a different structure, where scientists assume what features dark matter would need to have in order for tried-and-tested experimental setups to be able to detect dark matter particles. Once again, consider an example. As mentioned before, a lot of dark matter direct detection searches were modeled on successful neutrino detection experiments. They are (roughly – I’m setting aside many details here, see De Baerdemaeker (Reference De Baerdemaeker2021, 136-137) for a more complete reconstruction) justified in the following way:

1. P1: Direct detection experiments like those used for neutrino detection use the weak coupling to detect extraterrestrial particles by their deposited energy in scattering events.

2. P2: If dark matter is constituted by weakly interacting particles with an appropriate mass, then direct detection experiments like those used for neutrino detection can be used to search for signatures of dark matter particles from the galactic halo.

3. P3: Given current knowledge about dark matter based on cosmology and astrophysics, it is possible that dark matter is constituted by weakly interacting particles with an appropriate mass.

   ____________________________________

4. C: Direct detection experiments like those used for neutrino detection can be used to search for signatures of dark matter particles from the galactic halo.

Notice that this kind of ‘method-driven’ justification requires that the assumptions about dark matter’s weak interaction and its mass range are physically permitted given existing knowledge about dark matter (P3), even if there is no positive evidence for it. But even when strong evidence is lacking, the epistemic situation can sometimes be improved through plausibility arguments. In the case of several direct detection experiments, for instance, that plausibility argument comes in the form of the WIMP miracle: the fact that for WIMPs (Weakly Interacting Massive Particles), a popular class of dark matter candidates that also includes supersymmetric particles, the theoretically predicted dark matter abundance precisely matches the observed dark matter abundance.

This structure of justification should raise an obvious concern: doesn’t this make the results of dark matter searches dangerously dependent on the assumed properties? Of course, the method-driven justification of dark matter experiments lies on a spectrum with the ‘usual’ justification for experiments like MRIs, where the location on the spectrum depends on how well the required assumptions for the experiments can be independently justified. It seems that in the case of dark matter searches, these assumptions create an additional hurdle to establishing the reliability of experimental results for dark matter.

A common response to such model-dependence of experimental results is to use robustness arguments: if multiple independent (but model-dependent) results all agree on a measured parameter, this is taken as a good reason to believe that the result is a reliable measurement of the parameter in question. Antoniou (Reference Antoniou2023) shows, however, that robustness arguments have limited power in dark matter searches for two reasons. First, there is very little overlap between methods that are searching for the same dark matter candidate – in other words, most dark matter searches don’t even purport to probe the same parameter. Even when they probe the same parameter, different methods often probe different regions of parameter space, thus still not warranting robustness arguments. Second, the assumptions made in different dark matter searches are mostly about highly specific dark matter candidates and their specific features. That means that results can’t be compared across searches for different candidates (as Antoniou puts it, constraining the mass of WIMPs is very different than constraining that of axions). Antoniou concludes that although robustness arguments can be used to increase confidence in results for specific dark matter candidates, they are mostly useless in building up constraints for the overarching dark matter concept.

In light of this limitation on dark matter searches, one could take a different approach and aim to improve the constraints from cosmology and astrophysics on the space of possibilities of dark matter candidates. This can happen through more detailed astrophysical and cosmological observations. The Bullet Cluster, for instance, was crucial in setting some limits on interaction strengths. Similarly, the behavior of dark matter on galactic and sub-galactic scales within the framework of $\mathrm{\Lambda }$ CDM is not yet well-understood. Making progress on how dark matter behaves on smaller scales could provide an important clue for what candidates can constitute dark matter.

But just like with dark matter candidate searches, caution is key. Deriving predictions from $\mathrm{\Lambda }$ CDM for (sub-)galactic scales has to rely on complex computer simulations, since gravity is no longer the sole dominant interaction in this regime. This makes establishing the reliability of simulation outputs all the more challenging (more on this in Section 6). In fact, there are several discrepancies between what cosmological simulations predict, and what is observed: the so-called ‘small-scale challenges’ for $\mathrm{\Lambda }$ CDM (see, e.g., Bullock & Boylan-Kolchin Reference Bullock and Boylan-Kolchin2017, for a review). While it has sometimes been argued that these small-scale challenges provide motivation to modify the CDM-hypothesis, I submit such claims are premature: the simulations contain too many idealizations and are too ill-understood to draw any final conclusions about $\mathrm{\Lambda }$ CDM (De Baerdemaeker & Boyd, Reference De Baerdemaeker and Boyd2020). The only conclusion that can be drawn at this point is that making progress on the dark matter problem will require an iterative exchange between searches for dark matter candidates, cosmological and astrophysical observations, and complex modeling efforts.

3.4 Realism about Dark Matter

Dark matter also raises new challenges for the metaphysics of science. As has become clear so far, dark matter is an extension of or complement to scientific theories with a long-standing history. In some ways, it falls outside of the scope of established physics. Worse, because it doesn’t interact electromagnetically, we can’t ‘see’ dark matter structures (Jacquart, Reference Jacquart2020; Weisberg et al., Reference Weisberg, Jacquart, Madore and Seidel2018) in the way that other astronomical objects are seen. What does all this imply for the scientific realist? (For the scientific anti-realist, dark sector physics doesn’t seem different from other unobservables). Recall from Section 2.3 that scientific realism is committed to the existence of both observables and unobservables that are posited by our best scientific theories. If one is a scientific realist, should one be committed to the existence of dark matter?

As was mentioned before, contemporary views on scientific realism are selective in various ways. They don’t imply commitment to the existence of any entity or structure postulated by our current best scientific theories. The key question is: does dark matter clear the threshold for realist commitment? In the limited philosophical literature on dark matter realism, the early consensus seemed to fall on anti-realism (Allzén, Reference Allzén2021; Martens, Reference Martens2022) (although see Dellsén Reference Dellsén2019, for an opposing view).

Martens provides the most detailed argument, building on Chakravartty’s (Reference Chakravartty and Edward2017) breakdown of realism into three different commitments: metaphysical, semantic, and epistemic. The metaphysical commitment is to a mind-independent reality. The semantic commitment is that scientific claims have truth-value. The epistemic commitment is that scientific claims are actually true.

Martens takes issue with the semantic commitment when it comes to dark matter. Following De Baerdemaeker (Reference De Baerdemaeker2021) and Martens (Reference Martens2022), let us refer to everything that is currently broadly accepted about dark matter based on cosmology and astrophysics as the ‘common core’ dark matter concept: this is the core of properties that all dark matter candidates have to satisfy to qualify as dark matter. This includes that dark matter is nonbaryonic, electromagnetically neutral, slow-moving, and that it constitutes 27% of the current energy density of the universe.

Martens argues that this common core concept of dark matter is too thin to warrant realism:

The issue, it seems, is that because this common core still permits so many dark matter candidates with such widely varying properties (one would be hard-pressed to call primordial black holes and axions similar), this core is too vague to warrant any semantic realist commitment. Simply put, we wouldn’t know what we are realists about if we were to adopt a realist stance to dark matter.

Instead, Martens adheres to an “indefinitely suspended realism” (7) about dark matter. This leaves the door open for future realism about dark matter. If future scientific work would make the dark matter concept more precise and limit the underdetermination of models, then “semantic and epistemological realism would be vindicated simultaneously” (9). For instance, if dark matter searches turn out to be successful, or astrophysical and cosmological work puts tighter constraints on the space of possibilities, realism would be back on the table. Importantly, even if full-fledged dark matter realism is warranted in the future, that would still not imply that we would have been warranted to be dark matter realists today.

Vaynberg (Reference Vaynberg2024) has recently provided a response to Martens. Vaynberg begins by arguing that the thin common core dark matter concept that emerges out of cosmology and astrophysics, while minimal, is the concept “in virtue of which [dark matter] is causally connected to the phenomena it is supposed to explain” (6). That means that the concept is kind-constitutive. Furthermore, Vaynberg argues that there is empirical evidence supporting the existence of dark matter as defined by this thin common-core concept, thanks to the observations of the Bullet Cluster.

I agree with Vaynberg. While I concur with Martens that the current dark matter concept is fairly thin and that there are a large number of models that are still compatible with that thin common core, it is also the case that there are plenty of models that are not compatible with it, like neutrinos or massive compact halo objects. This, to me, provides grounds for a modest semantic realist commitment: scientific theories including dark matter posit the existence of something that is compatible with that common core. This ‘something’ plays an important explanatory role in cosmology and astrophysics. To be sure, what I am suggesting is a realist commitment to a vague concept. But that doesn’t seem to be a prima facie problem.

4.1 The History of the Cosmological Constant

As “ $\mathrm{\Lambda }$: the constant that refuses to die”, the title of Earman’s (Reference Earman2001) review of the history of the cosmological constant suggests, $\mathrm{\Lambda }$ has never fully disappeared from relativistic cosmology since it first appeared in Reference Einstein1917. Einstein (Reference Einstein1917) introduced a nonzero cosmological constant in a model of a closed and static universe, primarily motivated by matters of philosophical taste: Einstein sought to model a homogeneous static universe (assumed to be the case for the actual universe), while also satisfying Mach’s principle and the relativity principle (Earman, Reference Earman2001, 192-193). By 1931, the former point had already become moot. The cosmology community broadly accepted the expansion of the universe based on Hubble’s observational results (Hubble, Reference Hubble1929; Hubble & Humason, Reference Hubble and Humason1931) and theoretical work by Friedman (Reference Friedman1999/1922) and Lemaître (Reference Lemaître1927). Note that this did not immediately lead to the rejection of the cosmological constant: the Eddington-Lemaître cosmological model, which was the consensus view for a brief time around 1931, modeled a closed, uniformly expanding universe starting from an (unstable) Einstein universe extending to $t=-\mathrm{\infty }$, with a nonzero cosmological constant (De Baerdemaeker & Schneider, Reference De Baerdemaeker and Schneider2022).

The first model of an expanding universe with $\mathrm{\Lambda }=0$ was published by Einstein in 1931, but it gained little attention (O’Raifeartaigh et al., Reference O’Raifeartaigh, McCann, Nahm and Mitton2014). Einstein and de Sitter published an even simpler model in Reference Einstein and de Sitter1932: theirs was a monotonically expanding universe from an initial singularity. Despite these theoretical developments, and although Einstein had started to express unease with $\mathrm{\Lambda }$ relatively soon after first introducing it in 1917 (Earman, Reference Earman2001; O’Raifeartaigh & Mitton, Reference O’Raifeartaigh and Mitton2018), $\mathrm{\Lambda }$ stuck around long after the 1930s. There were several reasons for this.

First, the aforementioned cosmological models were not very well-known nor in line with other ideas about the universe. Einstein’s 1931 model went unnoticed for a long time. And at least according to Eddington (Reference Eddington, Needham and Pagel1938, 128), neither Einstein nor de Sitter considered their expanding universe model from 1932 particularly interesting. Einstein and de Sitter’s model was also the first explicitly open-universe model, while the preference for a closed universe seemed to have permeated cosmology for several decades (de Swart, Reference de Swart, Blum, Lalli and Renn2020).Footnote ¹⁹ Second, from a theoretical perspective, introducing the cosmological constant is a natural part of general relativity. Including the cosmological constant term gives the most general version of the Einstein Field Equations, with $\mathrm{\Lambda }$ an undetermined constant. This was already pointed out to Einstein by Tolman, who argued that this implied that the value of $\mathrm{\Lambda }$ could not be decided arbitrarily (Earman, Reference Earman2001, 197). Instead, Tolman advocated for it being determined empirically (the other alternative being to provide strong theoretical arguments for a particular value). Third, and in line with Tolman’s concerns, there was a serious empirical reason in the 1930s to set $\mathrm{\Lambda }\ne 0$: the timescale problem. Based on Hubble’s velocity-distance relation, the age of the universe based on its expansion history was thought to be an order of magnitude younger than the age of the oldest stars, thus also not permitting sufficient time for structure formation. This was the main motivation for Lemaître, for example, to hold on to a nonzero cosmological constant. It allowed him to introduce a ‘coasting phase’ in the expansion of the universe, such that the universe was older than the second expansion phase (De Baerdemaeker & Schneider, Reference De Baerdemaeker and Schneider2022). And finally, some, like Eddington, had theoretical preferences for a nonzero cosmological constant (Earman, Reference Earman2001, 203).

Some of these concerns disappeared in the second half of the twentieth century. Eddington’s theoretical musings were not successful, and the Einstein-de Sitter model gained prominence as the standard model of cosmology (see for instance Weinberg’s (Reference Weinberg1972) seminal textbook). Meanwhile, thanks to Sandage’s (Reference Sandage1958) and Baade’s (Reference Baade1956) work correcting Hubble’s distance measurements, the original timescale problem disappeared. That meant that the available empirical observations could be modeled with $\mathrm{\Lambda }=0$.

Again, however, this didn’t mean that $\mathrm{\Lambda }$ disappeared from view (Earman, Reference Earman2001, §10). Tolman’s argument still stood: $\mathrm{\Lambda }$ is part of the generalized Einstein Field Equations, with an undetermined – albeit bounded by observations – value. The search for a theoretical argument for the cosmological constant’s value was on. Beginning with Zel’dovich in the 1960s, particle physicists started to show interest in cosmology, and particularly in the cosmological constant. The realization dawned that the energy density of the vacuum behaves like a cosmological constant and that it should therefore be possible to derive a prediction for its value based on quantum field theoretic (QFT) considerations (Earman, Reference Earman2001, 207).

However, it quickly became clear that the value derived for $\mathrm{\Lambda }$ from QFT was orders of magnitude too large compared to the empirically determined upper bound – so large that the expansion of the universe would have happened too rapidly to allow for structure formation, let alone the development of human life. To solve this issue, one could assume that the vacuum energy density contributes to the effective cosmological constant ${\mathrm{\Lambda }}_{\mathit{\text{eff}}}$ together with a ‘bare’ cosmological constant ${\mathrm{\Lambda }}_0$:

${\mathrm{\Lambda }}_{\mathit{\text{eff}}}={\mathrm{\Lambda }}_0+8\pi G⟨{\rho }_{\mathit{\text{vac}}}⟩$(6)

${\mathrm{\Lambda }}_0$ is a free parameter that can cancel out the vacuum energy density contribution ${\rho }_{\mathit{\text{vac}}}$. This solution is hardly satisfactory, however: the value of ${\mathrm{\Lambda }}_0$ would have to be fine-tuned to cancel out over 118 decimal places to keep ${\mathrm{\Lambda }}_{\mathit{\text{eff}}}$ in line with observational bounds. Weinberg (Reference Weinberg1989, 1) proclaimed this problem “one veritable crisis”: the cosmological constant problem (CCP). The CCP motivated decades of searches for some ‘natural’ solution to the problem, where an unknown type of symmetry would justify setting ${\mathrm{\Lambda }}_{\mathit{\text{eff}}}$ to zero despite the large zero-point energies predicted by QFT (Koberinski, Falck, and Smeenk, Reference Koberinski, Falck and Smeenk2023).

In the 1990s, cracks started to appear in the case for naturalness due to new empirical anomalies. Starting in the 1980s, a new timescale problem emerged, this time related to globular star clusters. These are tightly gravitationally bound clusters of stars, often found in the halo of spiral galaxies like the Milky Way. Astrophysicists generally assume that stars in such clusters all formed around the same time, so they all have approximately the same age. Weinberg (Reference Weinberg1989, 8) mentioned that the estimated globular cluster age was older than the estimated expansion age of the universe, leading some to propose nonzero values for $\mathrm{\Lambda }$. The age estimates for globular clusters were not sufficient to force a general acceptance of $\mathrm{\Lambda }\ne 0$, however. That would require stronger, less model-dependent empirical evidence.

In 1998, this evidence was put forward by two competing teams working on supernova cosmology: the Supernova Cosmology Project and the High-z Collaboration. Supernova cosmology uses the fact that Type Ia supernovae (SNe Ia) are highly uniform to probe the evolution of the universe. SNe Ia that are observed at higher redshift reveal the deeper past of the universe. Guralp (Reference Guralp2020) gives a detailed historical reconstruction of the preambles and studies that ultimately led to the discovery of the accelerating expansion of the universe. The rest of this subsection is based on Guralp’s account.

While supernova cosmology had its start in the 1980s, it didn’t get properly going until the 1990s.Footnote ²⁰ One of the main hurdles was to find sufficient SNe Ia to achieve meaningful results. After revising their search method, the Supernova Cosmology Project published their first measurement for cosmological parameters in 1997. The results were wrong: they measured $\mathrm{\Lambda }=0$ matter density parameter ${\mathrm{\Omega }}_M\approx 1$ (Guralp, Reference Guralp2020, §3.2-3.3). The High-z Collaboration went through a similar prehistory (Guralp, Reference Guralp2020, §3.4). However, despite these erroneous results, Guralp shows that the early papers were essential in establishing the reliability of supernova cosmology and the teams’ respective methodologies.

By 1998, both teams were prepared to unleash the full potential of their newly established measurement protocols. In the span of less than a year, the two teams announced their surprising results that the universe’s expansion was not slowing down but speeding up, and that $\mathrm{\Lambda }>0$.

Given the aforementioned history of the cosmological constant and especially the expected naturalness arguments for $\mathrm{\Lambda }=0$, the evidence claims by the two teams needed to be extraordinarily strong. Guralp identifies two ways in which robustness arguments helped strengthen the evidence.Footnote ²¹ First, each team internally ran several types of data analyses (The Supernova Cosmology Project used different subsets of the data; the High-z Collaboration, which had less supernovae in its data set, used different analysis techniques) (Guralp, Reference Guralp2020, §3.5-3.6). Second, the fact that two competing teams announced the same result around the same time, helped convince the broader community of its accuracy (Guralp, Reference Guralp2020, §4). In an oral interview conducted by Guralp, Riess (High-z) recalled:

In other words, there was a strong robustness argument on the table.

In subsequent years, more and (according to Guralp) different robustness arguments became available through consistency checks with the CMB power spectrum. The WMAP results showed the universe being close to flat, but the matter contributions to the energy density only amounting to approximately 0.3 of the critical density (Spergel et al., Reference Spergel, Verde and Peiris2003). The remaining 0.7 can be contributed by a nonzero cosmological constant, or, in other words, dark energy. (The term dark energy was coined by Turner after the publication of the supernovae results as an umbrella term for whatever causes the accelerated expansion of the universe.) Thus, the cosmological constant came back from the dead, and stronger than ever.

4.2 The Cosmological Constant Problem(s)

The reintroduction of a nonzero cosmological constant brought back the CCP in a new guise. When Weinberg first introduced the CCP as a crisis, $\mathrm{\Lambda }$ was still an undetermined constant, albeit bounded by the fact that any form of life is possibleFootnote ²² and with all observations consistent with it being effectively zero. And while the discrepancy with the zero-point energy density was an issue, there was, at least at the time, hope for a solution via a naturalness argument. All in all, the only challenge was the so-called ‘old CCP’: explaining the discrepancy between $⟨{\rho }_{\mathit{\text{vac}}}⟩$ being large and $\mathrm{\Lambda }$ zero.

The supernovae results gave an exact empirical determination of $\mathrm{\Lambda }$ for the first time. The conflict with the zero-point energy remained an issue (the observed value of $\mathrm{\Lambda }$ is still many orders of magnitude smaller than any prediction for zero-point energy), but it added the difficulty that a solution through naturalness became implausible. This is what philosophers of science often refer to as the ‘new CCP’: the discrepancy between $⟨{\rho }_{\mathit{\text{vac}}}⟩$ being large and $\mathrm{\Lambda }$ being small-but-not-zero. While a solution to Weinberg’s original problem may have seemed within reach, the new CCP seems more of a veritable crisis because it lacks an obvious route to a solution.

The supernovae results also raised a new question: what is the physics behind the cosmological constant? In other words, what drives the accelerated expansion of the universe? If not vacuum energy density, then what? Following a recent review (Foundational Aspects of Dark Energy (FADE) Collaboration et al., Reference Collaboration, Bernardo and Bose2023) I will refer to this as the ‘dark energy problem’.Footnote ²³

The status of the CCP, old or new, as a problem has been philosophically contentious. Denoting the CCP as a genuine problem or ‘crisis’ means that the CCP is a persisting anomaly for current theory. Philosophical debates center around whether the CCP really is a problem, what the CCP is a problem for, and what permissible solutions to the CCP look like (Earman, Reference Earman2001; Koberinski, Reference Koberinski, Wüthrich, Bihan and Huggett2021; Schneider, Reference Schneider2020). Here, I’ll primarily focus on the new CCP, although the dark energy problem will make an appearance as well.

There are several reasons why philosophers of science have been critical of framing the CCP as a crisis. For one, some take issue with the assumptions required for setting up the CCP in the first place. Koberinski (Reference Koberinski, Wüthrich, Bihan and Huggett2021) identifies three problematic steps in setting up the CCP as a conflict between the calculated value of the vacuum energy density from QFT on the one hand, and the observed value for the cosmological constant on the other. First, Koberinski sheds doubt on the fact that there even is such an absolute value for $⟨{\rho }_{\mathit{\text{vac}}}⟩$, given that the true vacuum term never appears in quantum field theoretic calculations. Second, Koberinski argues that the calculation of $⟨{\rho }_{\mathit{\text{vac}}}⟩$ is problematic, because it is a divergent term. That means that to calculate $⟨{\rho }_{\mathit{\text{vac}}}⟩$, a cutoff energy scale needs to be introduced, and although any reasonable cutoff leads to a conflicting measurement with ${\mathrm{\Lambda }}_{\mathit{\text{obs}}}$, there are no grounds to assume that the divergences of $⟨{\rho }_{\mathit{\text{vac}}}⟩$ will be cured at higher energy scales. Koberinski is hesitant to trust a divergent term in QFT for which no mechanism for its regularization exists. Third, Koberinski questions how the vacuum energy density couples to gravity. As a result, Koberinski concludes that the CCP should not be taken seriously. Indeed, even though the CCP “has undoubtedly been successful as a heuristic motivator [...], no clear solution strategy has led to major progress in developing quantum gravity” (Koberinski, Reference Koberinski, Wüthrich, Bihan and Huggett2021, 276). Koberinski and Smeenk (Reference Koberinski and Smeenk2023) further argue that the CCP only arises if one assumes that effective field theoretic methods should apply for understanding $\mathrm{\Lambda }$.

Even if one buys into the technical setup of the problem, there is a further reason why some have argued that the CCP does not amount to a crisis – at least not for current theory. Schneider (Reference Schneider2020) points out that setting up the CCP requires an additional assumption: that the observed value for the cosmological constant ${\mathrm{\Lambda }}_{\mathit{\text{eff}}}$ is exhausted by the zero-point energies arising in QFT. However, as is already clear from Equation (6), there is no reason why one would not be permitted to introduce a bare cosmological constant term ${\mathrm{\Lambda }}_0$ to cancel out the contributions. There are different ways of interpreting this bare term, but as it stands, “its function in the [cosmological] model is that of a fit parameter to help match the model to empirical data” (Schneider, Reference Schneider2020, 5).

The issue with this solution is that the calculated value for $⟨{\rho }_{\mathit{\text{vac}}}⟩$ depends on the energy cutoff scale (Wallace, Reference Wallace2021). That means that ${\mathrm{\Lambda }}_0$ becomes a function both of the observed cosmological constant, and the arbitrarily chosen high-energy cutoff scale. Moreover, as Koberinski acknowledges, $⟨{\rho }_{\mathit{\text{vac}}}⟩$ will be many orders of magnitude larger than the observed value even for very low cutoff scales. ${\mathrm{\Lambda }}_0$ would therefore need to be fine-tuned to an extraordinarily high degree regardless of the cutoff scale, but differently for each cutoff scale. Wallace (Reference Wallace2021, fn.35) points out that “the relationships between bare and observed parameters in particle physics are not generally fine-tuned in this way”. So even if there is no strict empirical anomaly for current theory, the CCP does make for an unusual case of fine-tuning.

Of course, as was discussed in context of inflation (Section 2.1), establishing that fine-tuning is problematic is a notoriously hard philosophical problem, and Wallace (Reference Wallace2021, 34) concedes that one can debate whether fine-tuning to this degree is a problem at all (Wallace clearly sees it as such). At the very least, though, Wallace is right that the fact that ${\mathrm{\Lambda }}_0$ as a fit parameter is dependent on an arbitrary calculational choice signals a gap in current theory.

When it comes to the technical setup of the problem, Wallace takes a stronger stance. Wallace points out that the CCP arises out of fairly straightforward calculations in so-called ‘Low-Energy Quantum Gravity’ (LEQG), that is, “the theory one obtains by treating GR as an effective field theory, quantized through path-integral means” (Wallace, Reference Wallace2021, 36). LEQG is behind a lot of contemporary cosmology and astrophysics, and as such has proven to be highly successful. Wallace considers this success of LEQG important for two reasons. First, it gives a strong rebuttal against the technical concerns about the setup of the CCP. Second, the success of LEQG helps explain why the CCP is a central problem in contemporary physics: it is one of the few pointers that physicists currently have to a successor theory for LEQG, that is, to a more complete theory of quantum gravity.

Although Schneider (Reference Schneider2020), unlike Wallace, is not willing to take fine-tuning as sufficient to establish the CCP as a problem for current theory, Schneider does agree with Wallace on its importance as a pointer for future theory. After all, rejecting the CCP as a genuine problem raises a puzzle: if the CCP is not a problem for current theory, why is it so widely discussed by scientists?

Schneider suggests that the CCP serves as a motivation for future theorizing (see also Schneider Reference Schneider2022). Specifically, different assumptions about what a future theory of quantum gravity will look like motivate different formulations of the CCP. Each formulation then fits as a ‘problem-that-is-already-solved’ by the relevant preferred future theory of quantum gravity. Schneider identifies three different categories of solutions, related to three different assumptions about quantum gravity and three different versions of the CCP:

Note that all three types of solutions don’t just solve the new CCP, they also solve the dark energy problem. That is, they give a causal explanation for the accelerated expansion of the universe, either by introducing new types of matter fields, or by modifying general relativity.

This is not obviously the case for two other types of solutions to the new CCP. The first alternative solution is just assuming that $\mathrm{\Lambda }$ is a free parameter for which the value can be determined by observations. That value should be accepted as a brute fact. There are two issues with this attitude. First, it completely ignores the CCP, which at least according to Wallace (Reference Wallace2021) is not a tenable epistemic attitude. Second, from the perspective of practicing scientists, this attitude closes off any possibility for further inquiry (Koberinski et al., Reference Koberinski, Falck and Smeenk2023; Schneider, Reference Schneider2020; Wallace, Reference Wallace2021). And since pointers for a more complete theory of quantum gravity are slim, we better make use of the few that are available.

The other type of solution that doesn’t explicitly address the dark energy problem is anthropic solutions to the CCP. This type of solution aims to give some probabilistic argument as to why the observed value of $\mathrm{\Lambda }$ is to be expected, on the assumption that we live in a multiverse (due to eternal inflation or string theory, for instance). In other words, the observed value of $\mathrm{\Lambda }$ is explained statistically, rather than dynamically. There are various issues with anthropic solutions. Similar to the measure problem in inflation, Benétreau-Dupin (Reference Benétreau-Dupin2015) has argued that there are serious issues with the probabilistic arguments. Perhaps more concerning, anthropic solutions still suffer from UV-sensitivity: the landscape of possibilities out of which the observed value of $\mathrm{\Lambda }$ is selected depends on the chosen energy scale. Finally, Koberinski et al. (Reference Koberinski, Falck and Smeenk2023, 25) point out that these solutions inevitably rely on speculative physics generating a multiverse. Without separate reasons to trust the speculative physics suggesting the multiverse, it seems premature to accept this type of solution to the CCP and to close off further inquiry into other routes.

To end this section, let me briefly note that there is an alternative way to make progress on the dark energy problem at least: observational programs. Currently, several experiments are underway to either constrain the equation of state of dark energy (the hope is that this equation of state will prove to be time-dependent) or to find evidence of modified theories of gravity on cosmological scales (effects which would have to be minute, given the strong constraints on modified gravity from solar-system tests of GR). Ideally, these observations will shed light on the physics behind $\mathrm{\Lambda }$ and break the serious underdetermination between the many theoretical proposals (Koberinski et al., Reference Koberinski, Falck and Smeenk2023; Smeenk & Weatherall, Reference Smeenk and Weatherall2023). Some are skeptical that such cosmological observations can be sufficient to break the underdetermination between different proposals for the microphysical mechanism responsible for the accelerated expansion of the universe. Instead, they are concerned that cosmology may be stuck in a state of permanent underdetermination on this subject (Ferreira, Wolf, and Read, Reference Ferreira, Wolf and Read2025; Wolf & Ferreira, Reference Wolf and Ferreira2023). Suffice to say: to be continued.

5.1 What Is a Black Hole?

Before discussing black hole astrophysics specifically, I want to take a step back and reflect on the broad range of disciplines that study black holes. Since there are so many disciplines, there are many different ways of defining black holes. Based on an informal survey of physicists and philosophers from various sub-fields, Curiel (Reference Curiel2019) identifies at least twelve working definitions which are not necessarily mutually compatible. While reviewing some of the common definitions (e.g., the classical event horizon, a singularity), Curiel also highlights that each of these definitions has shortcomings. For instance, astrophysicists in practice define a black hole as “a system of at least a minimum mass, spatially small enough that relativistic effects cannot be ignored” (Curiel, Reference Curiel2019, 30). But, as Curiel points out, this definition could also be satisfied by a naked singularity, which are supposedly forbidden by Penrose’s cosmic censorship conjecture. For each of the proposed definitions, Curiel identifies similar pitfalls, reasons why they cannot be ‘the’ universal definition of a black hole.

This raises an obvious challenge. These definitions, despite not being equivalent and sometimes even being incompatible, “are all at bottom trying to get at the same thing” (Curiel, Reference Curiel2019, 32). In other words, the astrophysicists of the Event Horizon Telescope, the condensed matter physicists studying analogue black holes (more on both later), the theoretical physicists studying semi-classical gravity, they all purport to study the same class of objects. Without an agreed upon definition – indeed, without an agreed upon common core similar to the thin definition of dark matter (Section 3.3), it’s hard to see how the case can be made that all of these definitions are studying the same ‘thing’. But, as Curiel stresses, making this case is essential for applying results from one area of black hole research to another (32).

What is to be done in response to this conundrum? Doboszewski and Lehmkuhl (Reference Doboszewski, Lehmkuhl, Boyd, Baerdemaeker, Heng and Matarese2023) identify six possible responses. Curiel’s take, they argue, aligns with pragmatic pluralism. Curiel considers “black hole”:

In other words, the fact that there are many different ways to make the nebulous concept of a black hole more precise is seen as a good thing, because this flexibility permits scientists from so many different sub-disciplines to adapt the concept to their specific research questions. Of course, on this take, it is not clear how Curiel expects it to be possible to apply results across research fields.

Doboszewski and Lehmkuhl (Reference Doboszewski, Lehmkuhl, Boyd, Baerdemaeker, Heng and Matarese2023) have a different view. They review the relationships between some definitions and point out that these relationships “typically flow from the empirical and conceptual adequacy of an exact solution of Einstein’s field equations, in particular the Kerr and Schwarzschild solutions” (240). Given this commonality, it seems plausible that these exact solutions are the common core concept of a black hole. Whence then the plurality of definitions that Curiel’s survey revealed? That, they argue, is due to the introduction of various auxiliary assumptions that are required by the specific research context in which the common core concept is to be applied. In other words, the common core definition, that is, the exact solution to the Einstein field equations, is a highly idealized concept that needs to be de-idealized in various ways depending on the specific purpose.

While I am sympathetic to this position, there remains a lingering question. What does all of this imply for the metaphysics of black holes (also discussed by Allzén Reference Allzén, Boyd, Baerdemaeker, Heng and Matarese2023)? If the common core concept of a black hole is known to be a highly idealized concept that does not actually apply anywhere in nature (this is part of the concern Curiel raises about the classical definitions), and it is not clear how each practical definition relates to the common core, what then, can the realist commitment latch on to? More work is required to clarify this point.

5.2 ‘Direct’ Detection of a Black Hole

Even within the sub-discipline of black hole astrophysics, several epistemic puzzles appear. In the last decade, two milestones were reached in black hole astrophysics: the first detection of gravitational waves by the LIGO/Virgo consortium, and the first image of the event horizon and its immediate surroundings by the Event Horizon Telescope (EHT) collaboration. Interestingly enough, both collaborations explicitly or implicitly made claims of ‘directness’: LIGO/Virgo claimed the first “direct” detection of gravitational waves and the first “direct” observation of a binary black hole merger (Abbott et al., Reference Abbott, Abbott and Abbott2016), while the EHT touted its image of Sgr A* (the second image they released) as the “first direct visual evidence” that Sgr A* is a supermassive black hole (Event Horizon Telescope, 2022).Footnote ²⁶

A historical study by Skulberg and Elder (Reference Skulberg and Elderforthcoming) shows that language of “directness” in black hole imaging has a storied history. Still, this type of language is surprising to encounter in context of black hole research. Black holes, much like dark matter, are generically described as unobservable. Black holes are perfect absorbers (Hawking radiation would not be observable to a human observer – more on this in the next subsection). Black holes also have an event horizon, that is, they are a region of no escape. For an external observer, it is impossible to observe what happens inside the event horizon. Eckart et al. (Reference Eckart, Hüttemann and Kiefer2017, 555) therefore describe (supermassiveFootnote ²⁷) black holes as “unobservable entities” or “only observable by indirect means”. They further claim that we can only ever infer their existence based on their interactions with their surroundings, for example, their effects on a binary companion, an accretion disk, or through gravitational lensing effects.

There seems to be a tension in the description of the recent breakthrough results in observational black hole astrophysics, and the scientific reality that black holes are not in any intuitive sense directly observable. Elder (Reference Elder2020, Reference Elder, Patton and Curiel2023a, Reference Elder, Boyd, Baerdemaeker, Heng and Matarese2023b, Reference Elder2025) has taken on the task of clarifying the epistemology of gravitational wave astrophysics, including the use of ‘direct’ or ‘indirect’ in describing various observational results in black hole astrophysics. As Elder stresses, these terms are socially and politically laden, but there are also certain epistemic implications that may follow from them.

Elder (Reference Elder2025) focuses on the LIGO/Virgo-results, which claimed the first “direct detection” of gravitational waves, and the first “direct observation” of a binary black hole merger (see Collins Reference Collins2017, for a detailed history of this result). When it comes to the gravitational-wave detection, the directness qualification was meant to distinguish the new results from the 1970s observations of the Hulse-Taylor pulsar, a binary star system. Those observations revealed orbital decay in the pulsar: the distance between the two companions had decreased over time due to a loss of energy in the system. The energy loss corresponded exactly to the expected energy loss due to gravitational waves. This is widely considered to be convincing evidence for the existence of gravitational waves. In contrast, the LIGO/Virgo consortium detected gravitational waves for the first time in a terrestrial detector: the gravitational waves caused a slight disturbance in two perpendicular laser beams being reflected off hanging mirrors. For their first detection (as well as most subsequent ones), the signal was identified through matching it with an expected wave form for gravitational waves. Based on the detected signal, it was then inferred that the gravitational waves were due to a binary black hole merger.Footnote ²⁸

Building on work from Hacking, Shapere, Franklin, and Parker, Elder (2025) argues that the LIGO/Virgo collaboration’s directness claim was justified in the case of gravitational wave detection, thus setting it apart from the Hulse-Taylor pulsar. Elder first points out that both the LIGO/Virgo detection and the Hulse-Taylor pulsar observations require complex detectors (interferometers or telescopes) and various modeling assumptions. Thus, both of their detections are in fact inferences from raw data (instrument readings) to the phenomenon of gravitational waves. However, the LIGO/Virgo detection was direct because:

The reason why the interferometer can be black-boxed in the inference from the raw data to a gravitational-wave signal, is that the interferometer has been adequately calibrated and tested as a gravitational-wave detector. Contrast this with the Hulse-Taylor pulsar observations, which Elder classifies as indirect. There, the instruments that scientists could intervene on to calibrate and test was the radio telescope. The instrument was a detector of electromagnetic waves, not gravitational waves. The inference to gravitational waves required a further layer of inference about a separate target system (the pulsar). This layer of inference could not be black-boxed based on pre-existing knowledge of the detector (10). Importantly, Elder does not want to attach any general evaluation to directness or indirectness: both types can be epistemically powerful or suspect. Rather, the distinction indicates that there are different challenges to overcome in securing the reliability of the detection depending on whether it is direct or indirect (10).

Elder further evaluates whether the claim of a ‘direct observation’ of a binary black hole merger by the LIGO/Virgo collaboration was justified, and concludes that it was not. The inference to the binary black hole merger was based on a model of both the gravitational wave detector and the detected gravitational wave signal (Elder, Reference Elder2025, 10). As such, it was more similar to the Hulse-Taylor pulsar observations as evidence for gravitational waves, than the LIGO/Virgo-detections.

Elder’s analysis of (in)directness can also be applied to the evidence for Sgr A* being a supermassive black hole. Recall that the first strong evidence consisted of long-term observations of stars orbiting the object in the galactic center. Their orbital speed was so high, that the object needed to be extremely massive. The inference to the presence of a supermassive black hole required both modeling the telescope used for observations, as well as the orbiting stars. Skulberg and Elder (Reference Skulberg and Elderforthcoming) conclude that this qualifies as an indirect inference on Elder’s account.

This indirectness also seems to be at the root of some philosophical discussion as to the strength of the evidence for Sgr A* being a supermassive black hole prior to the EHT results. Given its intrinsic unobservability, Eckart et al. (Reference Eckart, Hüttemann and Kiefer2017) raise an underdetermination worry: while a supermassive black hole is one possible explanation for the motion of the orbiting stars, there are some alternative explanations like GRAVASTARS or boson stars. In other words, the models were not specific enough to empirically rule out alternatives. Doboszewski and Lehmkuhl (Reference Doboszewski, Lehmkuhl, Boyd, Baerdemaeker, Heng and Matarese2023, 233-234) argue that the inference to a supermassive black hole takes the form of an inference to the best explanation. This reconstruction does not remove the underdetermination – in their view, this would require probing the spacetime geometry right outside of a black hole – but it helps understanding the nature of the underdetermination.

The underdetermination is not pernicious. Even without conclusive evidence to eliminate alternatives, there are several reasons to not be too concerned about the underdetermination of the nature of Sgr A*. First, as Eckart et al. (Reference Eckart, Hüttemann and Kiefer2017) themselves argue, many of the alternative explanations suffer severe theoretical challenges, including the fact that some likely would have collapsed into a black hole during the formation or accretion phase. Second, the EHT has provided new empirical evidence for Sgr A*: it has significantly tightened the size estimates for the horizon to the order of the Schwarzschild radius. This also explains why the EHT press release announcing the Sgr A* results touts the observations as the “first direct visual evidence” that Sgr A* is a supermassive black hole (Event Horizon Telescope, 2022).

Finally, how do the EHT imaging efforts play out as far as Elder’s account of direct or indirect observation goes? The EHT uses a technique called ‘very long baseline interferometry’ (VLBI). The basic concept is the following: telescopes positioned around the earth simultaneously observe the same object, Sgr A*, several days in a row. This generates enormous amounts of data. By cleverly combining the images from the different telescopes together, scientists are able to ‘fill in the gaps’ between the different telescopes. As a result, they in practice have created a telescope that is the size of the entire Earth – a size that is essential to be able to observe an object as far away and as compact as Sgr A*.Footnote ²⁹

Elder (Reference Elder2025, 10-11) (see also Skulberg and Elder, Reference Skulberg and Elderforthcoming) recognizes that this case is difficult to assess. On the one hand, the raw data is about visibilities, not about a black hole. To get to the image of the black hole published by the EHT collaboration, complex imaging algorithms were required. On the other hand, one could argue (and this is reflected in the language used by scientists) that the relevant instrument in this case includes all the individual telescopes as well as the imaging algorithms. The relevant instrument is the telescope the size of the Earth. On the former reading, Elder suggests the EHT images would count as indirect; on the latter as direct.

The case is further complicated by the fact that we need to be more precise about what exactly the target phenomenon is: the black hole interior, exterior, or the event horizon (Doboszewski & Lehmkuhl, Reference Doboszewski, Lehmkuhl, Boyd, Baerdemaeker, Heng and Matarese2023, 13)? There are reasons why the interior of a black hole is in principle unobservable for an external observer: it is surrounded by an event horizon. If the EHT claims to have made a direct observation, it can surely only be of the exterior of the black hole, and specifically its effects on the surrounding materials in the form of accretion disks and jets. Of course, this suggest that it’s plausible that no observation of a black hole will ever be direct. So be it. As Elder stresses, there is no reason why this should prevent us from learning a great deal about black holes from indirect observations anyway.

To conclude this discussion of direct detections of gravitational waves or black holes, I want to reiterate that none of the so-called direct observations are even remotely similar to observation with the naked eye. All of them heavily rely on instrumentation and various kinds of simulations. Doboszewski and Elder (Reference Doboszewski and Elder2025) have recently analyzed the role of simulations in the LIGO/Virgo-results and in the EHT observations, and in what context these simulations lead to circular reasoning in the inferences drawn. I will return to the topic of computer simulations in Section 6.3.

5.3 ‘Indirect’ Detection: Hawking Radiation and Analogue Experiments

It should be clear by now that black holes are notoriously hard to get empirical access to. The previous section described some features of black holes that can be detected through their effects on their environment, but there are other predicted features that are not detectable. One such a feature is Hawking radiation, a predicted black-body radiation emitted by black holes. This phenomenon is one of the most crucial predictions from black hole thermodynamics.Footnote ³⁰ It is also much fainter than the CMB. This makes it unlikely that any human technology will ever be able to detect Hawking radiation. Evans and Thébault (Reference Evans and Thébault2020) classify Hawking radiation therefore as ‘unmanipulable and inaccessible (in practice)’.

Given that Hawking radiation is a cornerstone of black hole thermodynamics, physicists have tried to find ways around its inaccessibility: so-called analogue gravity experiments. Following an initial suggestion from Unruh (Reference Unruh1981), these experiments aim to create ‘dumb holes’, manipulable systems that are similar to black holes such that one would expect an analogous effect to Hawking radiation to be observable in the manipulable system.Footnote ³¹ Since the first proposal, the field has grown significantly (Field, Reference Field2021a). But the hope still remains that one can learn about black holes from fairly straightforward laboratory experiments. This raises the question whether this hope is warranted, that is, whether analogue experiments can offer genuine confirmation for hypotheses about black holes or not.

Before looking at the question of confirmation, some more detail on what sets analogue experiments apart from regular experiments is needed. In what follows, the ‘source system’ refers to the experimental system that can be actively intervened on in the lab, and the ‘target system’ is the system one wants to draw conclusions about based on what is learned from the source system. According to Crowther, Linnemann, and Wüthrich (Reference Crowther, Linnemann and Wüthrich2021), what sets apart analogue experiments from more conventional experiments is that in conventional experiments, the source and target system are “supposed to be [...] the same kind of system for the purpose of interest” (S3709). In analogue experiments, the source and target system are similarly “supposed to be [...] the same kind of system, [...] [but the target system] is inaccessible under the relevant conditions for confirming that it is actually the same kind of system as [the source system] for the purpose of interest” (S3709-S3710). Thus, the main difference between conventional and analogue experiments is whether we have access to the target to ensure that it is of the same type as the source system.

Evans and Thébault (Reference Evans and Thébault2020) push back against this definition by showing that the inaccessibility of a target does not automatically imply that it is impossible to draw any kind of inferences about that target. For instance, stellar interiors were confirmed to be of the same type as conventional laboratory systems in nuclear physics, despite stellar interiors being inaccessible. Instead, Evans and Thébault focus on the nature of the inference between source and target. While in conventional experiments source and target are assumed to be of the same type, this is not the case for analogue experiments. In analogue experiments, the source and target systems are of different-but-analogous types. More precisely, following Field (Reference Field2021b): in analogue experiments, the source and target are of a different type at the micro-level, meaning that the modeling frameworks to describe their micro-level behaviors are distinct (e.g., gravity and fluid dynamics), but they nonetheless display similar macro-level behavior. It is because the source and target systems are of different micro-level types that the reliability of source to target requires a new kind of justification compared to regular experiments.

Field’s classification clarifies why the question of confirmation is contentious for analogue experiments. The relation between source and target is one of analogy, and, as all authors in the literature agree, analogies are not confirmatory (Crowther et al., Reference Crowther, Linnemann and Wüthrich2021; Dardashti, Thébault, and Winsberg, Reference Dardashti, Thébault and Winsberg2017; Field, Reference Field2021b).

Despite this difficulty, Dardashti et al. (Reference Dardashti, Thébault and Winsberg2017) argue that analogue experiments can provide confirmation if the following conditions are met: the modeling frameworks for source and target systems are adequate within their respective domains, and the two frameworks are sufficiently similar such that there is an isomorphism between them within the relevant domains of applicability. In the case of analogue gravity experiments, they are confident that the isomorphism and the adequacy of the model of the source system are sufficiently established in the relevant domains of applicability. The difficulty lies in demonstrating the adequacy of the modeling framework for the black hole: how can this be established, given that scientists had to resort to analogue experiments due to the inaccessibility of black holes in the first place? Dardashti et al. argue that universality arguments can give reason to believe in the adequacy of the black hole modeling framework.

Crowther et al. (Reference Crowther, Linnemann and Wüthrich2021) take issue with this last step. They point out that these universality arguments assume that black holes can be modeled, as Hawking originally did, within semi-classical gravity. There are, however, well-known issues with this approach – not in the least the trans-Planckian problem (addressed by Crowther et al. Reference Crowther, Linnemann and Wüthrich2021; Thébault Reference Thébault, Dardashti, Dawid and Thébault2019). Thus, Crowther et al. are concerned that the universality argument that is required for analogue experiments to be confirmatory assumes the adequacy of the modeling framework it is supposed to establish.

Building on a Bayesian analysis of analogue experiments from Dardashti et al. (Reference Dardashti, Hartmann, Thébault and Winsberg2019), Field (Reference Field2021b, Sec. 2.2.1) gives a more precise analysis of the circumstances under which analogue experiments can be genuinely confirmatory. Field shows that the conditions for genuine confirmation are very weak: it only requires that there is a universality argument with undetermined validity, and that there must be a chance that this universality argument is positively relevant to both the target and source system. Field thus agrees with Evans and Thébault (Reference Evans and Thébault2020, Sec. 4.1.1) that denying the confirmatory power of analogue experiments collapses into inductive skepticism.

However, Field further shows that this does not imply that the confirmatory power of analogue experiments is significant. This because significance requires the universality argument to be actually positively relevant to both the target and the source system. And the problem in the case of analogue gravity experiments is that, as argued by Crowther et al. (Reference Crowther, Linnemann and Wüthrich2021), there is currently no such universality argument.

Field identifies two ways in which relevance of universality arguments can be established: either knowledge of the micro-structures of the target and source system reveal that they belong to the same universality class, or empirical tests of the macroscopic behavior of target and source reveal that they belong to the same universality class. It goes without saying that, as of right now, neither are available for black holes due to their current inaccessibility. Nonetheless, Field is hopeful that progress can be made – for example through the observations discussed in the previous sections, such that there is a future for analogue gravity experiments.

Two final notes are in order. First, Mathie (Reference Mathie, Boyd, Baerdemaeker, Heng and Matarese2023) gives a further argument in favor of the confirmatory power of analogue experiments. Mathie shows that the analogical inference between black holes and dumb holes that underpins analogue gravity experiments and the analogical inference underpinning black hole thermodynamics are crucially linked. This link suggests that there might be a tension between holding on to black hole thermodynamics and Hawking radiation on the one hand, and dismissing analogue gravity experiments on the other. Second, Field (Reference Field2021a) argues that today, confirmation of hypotheses about inaccessible targets (black holes, but also the early universe) is no longer the sole focus of analogue experiments. Field identifies two new roles: exploring the behavior of analogue systems in themselves, as well as investigating whether certain phenomena, like Hawking radiation, generalize. Thus, while we should remain modest for now about the potential for analogue gravity experiments to affect our understanding of black holes, these new roles do suggest that laboratory experiments could play an interesting role more broadly in context of cosmology and astrophysics.

6.1 Refuting the Lack-of-Experiment Objection

One cannot put the universe, a galaxy, or a star in a lab. This obvious fact was taken to be a huge challenge for astronomy and astrophysics by Hacking (Reference Hacking1989). Examining the case of gravitational lensing, Hacking argued that astrophysicists merely construct various “models upon models upon models” (576). This abundance of models was, according to Hacking, symptomatic of a broader issue: these models allow astrophysicists to “save the phenomena” (577). By contrast, the aim of natural science is “manipulating and interfering with the world in order to understand it” (577). Indeed, Hacking argued that natural sciences “came into being” when they adopted the experimental method. Since astronomy’s methods haven’t changed since ancient times, astronomy is therefore “not a natural science at all” (577).Footnote ³² While Hacking’s phrasing might be polemic, it is plausible that the objection can have some intuitive appeal. Surely, the lack of experiment poses a serious challenge?

All aspects of Hacking’s argument have garnered significant push-back. Shapere (Reference Shapere1993) rebuts Hacking’s review of gravitational lensing, both in terms of scientific content and in terms of it being a very young research area when Hacking first described it. Anderl (Reference Anderl and Humphreys2016) points out that the use of models is ubiquitous in all of science. But most focus has gone to Hacking’s dismissal of astrophysics as a science due to the lack of intervention, which Anderl (Reference Anderl and Humphreys2016, 657) helpfully frames as a concern about underdetermination: is it the case that a lack of intervention means that astrophysics is more susceptible to concerns about underdetermination and hence can only ever hope to save the phenomena?

At least three strategies have been used to respond to the underdetermination concern: (i) argue that astrophysics does make use of experiments along Hacking’s own criteria (Boyd, Reference Boyd, Boyd, Baerdemaeker, Heng and Matarese2023; Sandell, Reference Sandell2010) – sometimes combined with pointing out that ‘use’ of entities in research and ‘intervention’ on said entities are not obviously co-extensive (Doboszewski & Lehmkuhl, Reference Doboszewski, Lehmkuhl, Boyd, Baerdemaeker, Heng and Matarese2023; Shapere, Reference Shapere1993); (ii) accept, sometimes implicitly, the observation/experiment distinction as epistemically significant but argue that the lack of experiment can be overcome without implying a loss in scientific status – this response is either explicitly linked to or mirrors discussions about the epistemology of historical sciences more generally (Anderl, Reference Anderl and Humphreys2016; Elder, Reference Elder, Boyd, Baerdemaeker, Heng and Matarese2023b, Reference Elder2024); and (iii) argue that the distinction between experiment-qua-physical-intervention on the one hand, and observation on the other cannot serve as a guide for epistemic judgments (Boyd & Matthiessen, Reference Boyd and Matthiessen2024).

As an example of the first strategy, Boyd (Reference Boyd, Boyd, Baerdemaeker, Heng and Matarese2023) discusses cases from the field of ‘laboratory astrophysics’. Examples include dark matter production experiments and analogue black hole experiments, but also spectroscopy, accelerator-based nuclear astrophysics, and laboratory supernova research, Boyd’s main case study. Laboratory astrophysics should, according to Hacking’s classification, be a contradiction-in-terms. Boyd counters that:

Boyd argues that the inference from the terrestrial experiment to the target system out in space happens through the usual argument for the external validity of an experiment, that is, by arguing that the states probed in the lab and those instantiated by the astrophysical systems are tokens of the same type (23-25). If successful, it seems that there is no reason to be more concerned about underdetermination in laboratory astrophysics than in other experimental sciences, thus undermining Hacking’s argument.

Interestingly enough, Boyd argues that in the case under discussion, the external validity arguments actually fail. However, Boyd stresses that this would not have become clear if one stayed focused on framing the experiment as ‘experimental’ and therefore ‘epistemically better’. Indeed, a corollary of this first strategy is that, sometimes, observations are better suited for a particular research question. For instance, Elder (Reference Elder, Patton and Curiel2023a) points out that LIGO/Virgo is able to probe novel regimes of general relativity, even though there is a clear lack of manipulation (see also Elder Reference Elder2025). A nice example of how observations have informed realistic models of the interior of stars is discussed by Suárez (Reference Suárez, Boyd, Baerdemaeker, Heng and Matarese2023). In modern particle physics, the first empirical evidence for the existence of neutrinos came from observations of the Sun. And the universe is thanks to Zel’dovich more generally known as the ‘poor man’s accelerator’, because the early universe and stellar interiors can reach energy scales that can never be recreated in terrestrial experiments.

What about the second strategy? Anderl (Reference Anderl and Humphreys2016) connects the concerns about underdetermination to a debate about historical sciences. Historical sciences aim to reconstruct causal chains that led to current observations without any intervening on that hypothesized causal chain. Anderl’s response to the underdetermination concern is twofold. First, Anderl follows Cleland’s (Reference Cleland2002) argument that stereotypical historical science uses the so-called Sherlock-Holmes strategy: find a causal explanation that can unify all observed traces, and discriminate between competing explanations using ‘smoking gun’ evidence. The lack of experiment is no reason to be concerned about underdetermination, because of the time-asymmetry of causation, Cleland continues: one cause can generate a range of traces, of which only a subset can be sufficient to identify the original cause. Anderl agrees with Cleland’s conclusion that it’s not obvious that underdetermination due to unidentified auxiliaries in experimental settings is in any way less strong than underdetermination due to multiple explanations for traces in observational settings. Thus, while it is the case that astrophysicists cannot actively recreate causal chains in the lab, it is not obvious that that creates a significant epistemic hurdle compared to experimental sciences.

Second, Anderl points out that not all types of astrophysical research are stereotypically historical science. While some research questions are about reconstructing the causal history of singular events, a lot of astrophysics aims to make general statements about types of objects, from main-sequence stars to Type II supernovae or elliptical galaxies.Footnote ³³ The universe contains a lot of instances of each of these classes: it constitutes a “cosmic laboratory” (Anderl, Reference Anderl and Humphreys2016, Section 3.2). Because astrophysicists can observe so many instances of the same type, they can at the very least run ‘quasi-experiments’. Quasi-experiments compare a control and test group which are not randomly assigned. They therefore require the use of statistical tools and simulations to draw any conclusions. Similar to the first part of Anderl’s argument, I believe it’s safe to conclude that this aspect of astrophysical methodology precludes any strong conclusions about underdetermination being more concerning in astrophysics than other scientific research areas.

I would add a third point to Anderl’s response: as discussed in Section 2.2.3, it has been convincingly argued that the historical sciences have a much richer methodology than solely the Sherlock Holmes strategy (Currie, Reference Currie2018; Reference Novick, Currie, McQueen and Brouwer2020). This makes it even more difficult to draw any principled conclusions about any historical science as a whole. A clear illustration of this point comes from Yao (Reference Yao2023). Yao has recently argued that most of the inferences in contemporary astronomy don’t rely on natural or quasi-experiments, but rather on “Rosetta-Stone situations” (1389). These are cases where certain token objects have specific properties that are tied to various historical contingencies. These properties provide external information about the token of interest that helps draw inferences about the token’s internal properties. Yao gives the example of how stars being part of the same cluster implies that they are born at the same spatiotemporal point, and that only their mass will vary. This kind of information can then help construct a complex web of models, for example, of stellar evolution, that can account for a broad range of datasets. Because of the interlocking models and data sets, it is highly unlikely that an alternative complex web would be possible. In other words, underdetermination is limited in Rosetta-Stone situations. Thus, from the perspective of the epistemology of historical sciences, concerns about underdetermination will have to be judged on a case-by-case basis.

This final point about the rich methodology of historical sciences relates to the third strategy to respond to Hacking: argue against drawing any strong distinction between experiment-qua-physical-intervention and observation. Boyd and Matthiessen (Reference Boyd and Matthiessen2024) identify and dismiss two arguments in favor of the distinction and the epistemic superiority of experiments:

Boyd and Matthiessen show that both of the expressed goals, hypothesis discrimination and causal inference, can be achieved through observation and experiment alike, and that it depends on the research context which empirical method will be better-suited to that goal. They therefore propose to shift focus to specific features of empirical research, such as signal-to-noise ratio or background characterization, as a way to judge epistemic superiority.

This to me seems exactly right, and a more fruitful way of approaching the epistemology of cosmology and astrophysics. In fact, the first two strategic responses are in a way illustrations of the general point by Boyd and Matthiessen. The fact that there is a field called “laboratory astrophysics” or that black hole researchers appeal to analogue experiments reveals that scientists themselves don’t think in terms of ‘observational’ or ‘experimental’ sciences, but rather in terms of what method is most appropriate to a specific goal (see also the setup of dark matter searches discussed in Section 3.3). Moreover, aside from the reliance on various kinds of experiments, Abelson (Reference Abelson2022b) has further shown that multimessenger astronomy, where evidence from cosmic rays, electromagnetic waves, gravitational waves and neutrinos are combined, is an extremely powerful tool. Abelson argues that the value of multimessenger astronomy lies in the fact that it can lead to the elimination of different plausible hypotheses–in other words, it allows the mitigation of underdetermination. The fact that cosmology and astrophysics are able to overcome underdetermination often without actively intervening on their targets (as happened in the MOND/dark matter debate) again shows that the observation/experiment distinction as a principled epistemic guide is moot.

6.2 Generalizing Based on Limited Samples

Showing that the presumed lack of experiment does not pose a principled problem for cosmology and astrophysics does not mean that we are completely out of the woods. There are, like in any area of scientific research, genuine epistemic challenges with regard to limited empirical evidence – some common to both fields, some unique to one. Importantly, these challenges are not due to a lack of experiment. Rather, they are due to various ways in which we possess a unique vantage point on our universe and how that affects certain specific observations we can (not) make. Understanding the nature of these challenges, as well as what is required for a satisfactory response thus needs a more careful philosophical analysis. Here, I review the most important ones.

6.3 Multi-Scale Modeling

The previous section discussed challenges that arise due to our limited vantage point on a single universe. There is a separate set of challenges that relates to multi-scale modeling. Cosmology and astrophysics aim to model systems across large ranges of length- and timescales. While the relevant physics is largely deterministic, unlike, for example, in climate science where human decision-making is relevant, the range of scales raises significant hurdles.

6.4 The Ethics of Cosmology and Astrophysics

Philosophers should not just contribute to epistemological debates in cosmology and astrophysics. Another equally, if not more, important task is to participate in ethical debates about these disciplines. A naive view may suggest that cosmology and astrophysics would be morally good or at the very least neutral (who could be harmed by observations of stars?). This is far from the truth. Cosmology and astrophysics do not operate in a vacuum. They are human endeavors practiced in a particular sociopolitical, historical, and environmental context (de Swart, Thresher, and Argüelles, Reference de Swart, Thresher and Argüelles2024). For instance, the fact that these disciplines are becoming more and more computationally expensive raises questions about their climate impact and how to mitigate it. Cosmology and astrophysics (and astronomy) also have a long history of entanglement with colonialism and oppression (Prescod-Weinstein, Reference Prescod-Weinstein2021). The 1919 eclipse expedition that provided key evidence for general relativity was enabled by the combined colonial powers of Britain and Portugal (Simões & Sousa, Reference Simões and Sousa2019). And even the newly accepted model of the expanding universe was appropriated by some to defend white settler colonialism in the 1930s (Schneider & De Baerdemaeker, Reference Schneider and De Baerdemaeker2023).

Arguably one of the most complex issues is the siting of new earth-based telescopes. In a recent podcast episode, Thresher summarizes various challenges associated with telescope siting (Enander & Thresher, Reference Enander and Thresher2024). The following discussion is largely drawn from that episode.Footnote ³⁵ Optimal observing conditions for telescopes require remote locations at high elevation. That usually implies culturally significant mountain tops with delicate ecosystems in countries with a history of colonization. Building new telescopes thus comes with environmental, cultural, political, and socioeconomic impact.

With regard to environmental impact, Thresher highlights that telescope construction commonly implies disturbing unique ecosystems. For example, Mount Graham in Arizona, home to three telescopes, is also home to the endangered Mt. Graham red squirrel (Swanner, Reference Swanner2013).Footnote ³⁶ Moreover, the construction of ultra-clean detectors and mirrors often generates large amounts of waste. Decommissioning is an even more delicate exercise (the environmental impact study for decommissioning the Arecibo telescope in Puerto Rico (NSF, 2017) gives a good sense). In the past, there often were no clear plans for decommissioning telescopes that had reached the end of their lifetime, thus leaving enormous structures to slowly decay and further destroy the environment. Responsibly decommissioning telescopes comes with a hefty price tag. ‘Dealing with the issue once it arises’ is clearly not an option when the cost is that high.

Culturally, Thresher points out that telescopes are often constructed on mountains that are culturally significant for local and indigenous communities. These communities have historically not been heard in the construction of previous telescopes, which tends to create a deep sense of mistrust towards new telescope projects. This is closely tied to the political impact of telescope construction. Many telescope sites are in regions with a history of colonization or oppression. Consider, for instance, the case of Chile, one of the prime locations for Earth-based telescopes. The European Space Organization (ESO) runs various telescopes in the Atacama Desert, including the Very Large Telescope, and the future (not very creatively named) Extremely Large Telescope. In the 1990s, the construction of the Very Large Telescope became imperiled due to legal disputes: the land on which it was to be constructed was claimed to be private land which had unjustly been donated in the 1970s to ESO by Pinochet, Chile’s former dictator, whose regime was notorious for its human rights violations (Long, Reference Long1994).

Finally, Thresher describes the socioeconomic impact of telescopes. Modern telescopes represent the cutting edge of human technology, but they are rarely accessible to people living nearby. Instead, the use of telescopes is a privilege for international scientists, who can fly in for a short observing run or even run their observations remotely. Local communities are historically only offered less-prestigious technical or janitorial job opportunities. Meanwhile, credit for new discoveries tends to go to research scientists, much less to the technical staff.

The case of the Thirty Meter Telescope (TMT) at the MaunakeaFootnote ³⁷ peak in Hawai’i provides an example of how all of these factors can come to a head. Maunakea is one of the best locations for Earth-based astronomy in the Northern hemisphere, due to its remote location and stable atmosphere. It has been developed for astronomy since the second half of the twentieth century, but the mountain is sacred for native Hawaiians.Footnote ³⁸ The construction of the TMT, which would be larger than any existing telescope on the peak, has been continuously opposed, protested, and blocked by native Hawaiians. While the conflict has often been unhelpfully framed as one of ‘science vs. religion’, the truth is much more nuanced: it is about a history of colonization, environmental impact, a lack of respect for indigenous rights, questions about informed consent, and the assumption that scientific progress can come at all costs. Maunakea is a prime example of how intricately tied telescopes are to the context in which they are constructed, and how ignoring that context can significantly harm both people and science. Thresher implores that reckoning with that context and making morally informed choices is crucial for the future of cosmology and astrophysics and for a just science.

Recent efforts have shown that things can get better. The next generation Event Horizon Telescope (ngEHT) collaboration includes a responsible siting working group. This working group is tasked with developing guidelines for how to site a new telescope without causing harms in any of the aforementioned ways. One of the central questions is to understand what ‘informed consent’ can look like when it comes to siting a new telescope, if it is at all possible in a given historical context. The group consists of philosophers of science, ethicists, sociologists, anthropologists, astronomers, and more. By recognizing that responsible siting is key to the future of cosmology and astrophysics, and by actively calling on the expertise of humanities scholars and social scientists, the ngEHT hopes the avoid the harms that have often been done in the name of science.