Consciousness in Artificial Intelligence: Insights from the Science of Consciousness
(ArXiv, August 2023)
Media coverage: Article in Science, article in Nature
Whether current or near-term AI systems could be conscious is a topic of scientific interest and increasing public concern. This report argues for, and exemplifies, a rigorous and empirically grounded approach to AI consciousness: assessing existing AI systems in detail, in light of our best-supported neuroscientific theories of consciousness. We survey several prominent scientific theories of consciousness, including recurrent processing theory, global workspace theory, higher-order theories, predictive processing, and attention schema theory. From these theories we derive “indicator properties” of consciousness, elucidated in computational terms that allow us to assess AI systems for these properties. We use these indicator properties to assess several recent AI systems, and we discuss how future systems might implement them. Our analysis suggests that no current AI systems are conscious, but also shows that there are no obvious technical barriers to building AI systems that satisfy the indicator properties. Authors: Patrick Butlin, Robert Long, Eric Elmoznino, Yoshua Bengio, Jonathan Birch, Axel Constant, George Deane, Stephen M. Fleming, Chris Frith, Xu Ji, Ryota Kanai, Colin Klein, Grace Lindsay, Matthias Michel, Liad Mudrik, Megan A. K. Peters, Eric Schwitzgebel, Jonathan Simon, Rufin VanRullen.
Consciousness Constitutively Depends on the Low Entropy State of the Early Universe
Your time-reverse twin’s life is your life, but played in reverse, frame-for-frame. What is it like to be your time-reverse twin? I argue (developing a point of Tim Maudlin’s) that there is nothing it is like: your time-reverse twin is a zombie. Consciousness depends on computation, but computation depends on orientation toward the low entropy end of the universe. Your twin is improperly oriented. A corollary, since your life is a perfect intrinsic duplicate of your twin’s, is that consciousness is radically extrinsic: it constitutively depends on the low entropy state of the early universe.
Experiencing Left and Right in a Non-Orientable World (Analytic Philosophy, 2021)
[Awarded Winner, 2018 Marc Sanders Prize in Mind] 2021 pre-print here, 2018 prize version here.
Consider the totality of your phenomenal experience right now — your total experience. Is there a total experience which is phenomenally different from yours, but which differs only by a mirror symmetry, the way that a picture of a left hand differs from a picture of right hand, or is there no phenomenal difference between a total experience and its mirror-reversal? If you think that there is a phenomenal distinction between an experience and its mirror reversal (a position Chalmers dubs `e-categoricalism’) then you may find it intuitive that your mirror twin — someone who is a molecule-for-molecule mirror reflection of you — in general has a different experience than you. After all, if you are looking at your left hand, she is looking at her right hand. Lee (2006) argues, however, that your mirror twin must be your phenomenal twin, if relationalism about space is correct. Paired with e-categoricalism this has puzzling consequences. Here, I begin by challenging Lee. I argue that even given relationalism about space, your mirror twin can fail to be your phenomenal twin. But this result is limited. It only applies where you and your mirror twin both live in a universe with an orientable topology. If your universe has a non-orientable topology (in the sense in which Möbius strips and Klein bottles have non-orientable topologies) then your mirror twin must be your phenomenal twin after all. Moreover, this moral applies even to those who reject relationalism about space. The upshot is that everyone, non-relationalists included, must either abandon e-categoricalism, or choose between puzzling consequences, the most promising of which may be property dualism.
Transparency about Painkillers: A Remedy for the Evaluativist’s Headache (PPQ, 2019. DOI: https://doi.org/10.1111/papq.12278)
Draft version here.
Evaluativism is the view that unpleasant pain experience represents bodily disturbance as normatively bad. The killing the messenger objection to evaluativism is that while the evaluativist has a nice story about the rationality of body-directed action in response to pain (like tending to the wound), the evaluativist confronts various obstacles in accounting for the rationality of experience-directed action in response to pain (like taking painkillers). Evaluativists have overcome some of these obstacles without questioning the claim that actions like taking painkillers really are experience-directed in the typical case. Here I consider some obstacles that evaluativists have not yet overcome, and I suggest that the best way for evaluativists to overcome these new obstacles is to question that claim after all. I then develop a positive account of a version of evaluativism that proceeds accordingly, which I call naive evaluativism. According to naive evaluativism, seemingly experience-directed actions like taking painkillers are directed at the bodily badness that pain experience represents in the typical case. We experience painkillers as diminishing bodily badness, and that is why we take them.
Fragmenting the Wave Function (Oxford Studies in Metaphysics, Vol. 11. 2019, 123-148.)
[Awarded First Runner-up, 2017 Marc Sanders Prize in Metaphysics] Draft version here.
I argue that the fragmentalist framework introduced by Fine (2005) can be deployed to solve problems for B-theorist endurantists, and also to understand the quantum state in such a way that it is more or less all there is to the universe, it is grounded in its branches, and these branches are grounded in beables. Along the way, I develop a new way for something to be co-located with itself.
The Hard Problem of the Many (Phil Perspectives, 2017)
(Philosophical Perspectives, Vol. 31, Issue 1. pp. 449-468.) Draft version here.
A problem of the many Fs arises in cases where intuitively there is precisely one F, but when you look closely you find many candidates for being that F, each one apparently as well-qualified as the next. The problem arises for mundane things like rocks, houses, and coins. It also arises for entities that present special philosophical challenges, like persons and experiencers. In this essay, I present a new argument that the problem of the many experiencers is an especially hard problem of the many, and that property dualism – the view that properties that there is something it is like to instantiate are irreducible – may be the best way to solve it. The argument given here turns primarily on normative (i.e., moral) considerations, and is independent of existing arguments for property dualism such as the conceivability and knowledge arguments. It is also independent of existing arguments deriving metaphysical conclusions from the problem of the many experiencers (or related problems of the many), such as those in Unger (2004) and Zimmerman (2010).
Mendelssohn, Kant and the Mereotopology of Immortality with Colin Marshall (Ergo, 2017).
(Ergo, Volume 4, No. 29, 2017 DOI: http://dx.doi.org/10.3998/ergo.12405314.0004.029) Draft version here.
We argue that Mendelssohn’s argument for the immortality of the soul merits more consideration than it has received, and that it has instructive parallels with other influential arguments concerning the puzzle of contact, vagueness, and more.
Vagueness and Zombies: Why `Phenomenally Conscious’ Has no Borderline Cases (Phil Studies, 2017)
(Philosophical Studies, Vol. 174 (8), 2105–2123.) Draft version here.
I argue that there can be no such thing as a borderline case of the predicate `phenomenally conscious’: for any given creature at any given time, it cannot be vague whether that creature is phenomenally conscious at that time. I first defend the Positive Characterization Thesis, which says that for any borderline case of any predicate there is a positive characterization of that case that can show any sufficiently competent speaker what makes it a borderline case. I then appeal to the familiar claim that zombies are conceivable, and I argue that this claim entails that there can be no positive characterizations of borderline cases of `phenomenally conscious’. By the Positive Characterization Thesis, it follows that `phenomenally conscious’ can not have any borderline cases.
Indeterminate Comprehension (Thought, 2014)
(Thought, Vol. 3, No. 1, March 2014, pp. 39-48). Draft version here.
Can we solve the Problem of the Many, and give a general account of the indeterminacy in the definite descriptions that give rise to it, by appealing to metaphysically indeterminate entities? I argue that we cannot. I identify a feature common to the relevant class of definite descriptions, and derive a contradiction from the claim that each such description is satisfied by a metaphysically indeterminate entity (a Comprehension principle for indeterminate definite descriptions).
What is Acquaintance with Consciousness? (Consciousness Inside and Out, 2014)
(Consciousness Inside and Out: Phenomenology, Neuroscience, and the Nature of Experience, ed. Richard Brown. Studies in Brain and Mind ed. Gualtieri Piccinini, Springer. Vol. 6, 2014, pp 103-118). Draft version here.
Distinguishes various versions of the idea that the concept ‘consciousness’ acquaints us with consciousness. Distinguishes these acquaintance theses from the premises of Jackson and Chalmers’ anti-materialist arguments; argues that the acquaintance theses are stronger. Then argues that, pace Phillip Goff, none of these acquaintance theses imply anything about the non-vagueness of ‘consciousness’.
The Protestant Theory of Determinable Universals (Johanssonian Investigations, 2014)
(Johanssonian Investigations. Essays in Honour of Ingvar Johansson on His Seventieth Birthday. eds. Svennerlind, C., Almäng, J., & Ingthorsson, R. Berlin, Boston: De Gruyter. 2013. pp. 503-515.). Draft version here.
Some, such as Ingvar Johansson, have argued that in addition to determinate universals like Red27 and Red28 there are determinable universals like Colored. Johannson defends a relatively austere theory, where only maximally determinable universals like Colored exist in addition to determinates like Red27. This paper argues that, if we must countenance determinable universals at all, we should countenance intermediary ones like Cardinal Red to intervene between the lowly determinates and highest determinables.
Truthmaker Explanations with Barry Smith (Metaphysics and Truthmakers, 2007)
(Metaphysics and Truthmakers, ed. Monnoyer J.M., Frankfurt a.M.: Ontos 2007, pp. 105–156. (Reprinted in translation as Explications Verifactionnistes (trans. Anne-Marie Boisvert), Philosophiques, Vol. 38 No. 1, 2011, pp.177-194.))
Armstrong argues that Truthmaker theory is needed to provide both a real definition of Truth and a nominal definition of ‘Truth’. This argument is flawed, since the core insight of Truthmaker theory is compatible with deflationism about truth: asking for truthmakers is a way of asking for ontological explanation in the formal mode. In this paper we argue that Armstrong’s case for Truthmaker Maximalism hinges on the flawed argument, and accordingly, his case for Truthmaker Maximalism fails. The paper also develops a characterization of ontological commitment in truthmaker terms, and discusses the methodology appropriate to the search for truthmakers, giving examples.
Is Time-Travel a Problem for the Three-Dimensionalist? (The Monist, 2005)
(The Monist, Vol. 88, No. 3, 2005, pp. 353-361)
Argues, pace Ted Sider, that the four-dimensionalist has about as much trouble distinguishing between time travel scenarios as the three-dimensionalist. Still, time travel shows that three-dimensionalists must relativize property instantiation to some parameter besides (or in addition to) objective time, and this may be problem enough.
No Port, No Passport: Why Submerged States Can Have No Nationals with Heather Alexander (Wash. Int’l L.J., 2017)
(Wash. Int’l L.J. 26.2, pp.307-324.).
Territorial loss owing to sea level rise presents novel challenges to the international legal order. Nowhere is this clearer than in the case of small island states like the Maldives, Tuvalu and Kiribati, whose very existence is in jeopardy. In our recent article, Sinking Into Statelessness, we argue that the principle of presumption of continuity of state existence does not ensure that sinking states shall or may retain their legal statehood, because that principle cannot overrule the fact that territoriality is a constitutive feature of legal statehood. Here, we argue that even if, contra our previous conclusion, submerged states retain their legal statehood, territory is nevertheless necessary in order for a state to confer nationality in the sense of the 1954 Convention Relating to the Status of Stateless Persons, that is, for a state to consider someone a national under the operation of its law. In consequence, even granting that such a state could exist and have members, its members would need nationality in another state in order to avoid de jure statelessness. To establish this claim, we will argue that for a state to consider someone a national under the operation of its law, that state must be capable of complying with the duty to readmit nationals when requested to do so by another state.
‘Unable to Return’ in the 1951 Refugee Convention: Stateless Refugees and Climate Change with Heather Alexander (FJIL, 2014)
(Florida Journal of International Law, Vol. 26 No. 3 2014 pp 531-574). Final version via Hein Online here.
Argues that it is not only a point of literal construction, but also inherent in the object and purpose of the 1951 Refugee Convention, that displaced stateless persons unable to return to their countries of former habitual residence may be eligible for refugee status even if unpersecuted. ‘Unable to return’ as it occurs in the clause following the semi-colon of 1(A)2 of the 1951 Refugee Convention must be understood as a term of art subject to appropriate canons of construction in its own right. Its construal must therefore be more restrictive than many commentators have suggested, though not so strict as to preclude all but persecuted persons. Then argues that, as a case study, those who are displaced from their island nations because those nations have submerged beneath the sea will count as ‘unable to return’ in the relevant sense, and so will qualify for Convention refugee status, if they count as lacking a nationality, i.e. as stateless.
Sinking Into Statelessness with Heather Alexander (Tilburg Law Review, 2014)
(Tilburg Law Review, Special Issue on Statelessness, Vol. 19, No. 1-2, January 2014, pp.20-25)
[among Brill’s 15 most downloaded papers in Law in Q1 2014!]
Argues that, contrary to the view of many scholars, and the findings of a recent U.N. panel of experts at Bellagio, the principle of presumption of continuity of state existence can play no role in disqualifying those whose island nations are submerged beneath the sea from the status of statelessness as defined by the 1954 and 1961 Statelessness Conventions.
Boat Migrants to Australia Deserve their Refugee Rights with Christian Barry (Policy Innovations, 2012)
(www.policyinnovations.com, October 1, 2012).
Considers the claim, influential in Australia, that boat migrants are “queue jumpers”. Talks about what it takes to jump a queue, and argues that boat migrants are doing no such thing.
Formal Ontology for Natural Language Processing and the Integration of Biomedical Databases (IJMI, 2005)
(Simon, J., Fielding, M., Dos Santos, M., Smith, B., International Journal of Medical Informatics 75 (3-4), 2005, pp. 224-231.)
Demonstrates the utility of the IFOMIS project’s philosophically informed taxonomy system for the resolution of ambiguities and inconsistencies within medical databases and database tools, and for the standardization of mappings between them.
How to Be a Bicategorialist About Persistents (FOIS, 2004)
(Proceedings of FOIS 2004 (Formal Ontology in Information Systems). IOS Press. 2004, pp. 60-69 ). Draft version here.
Sketches a bicategorical logic of participation. The logic is designed to accommodate the idea that objects (endurants) participate in events (perdurants) without literally being parts of those events. A class of intended models is identified, some semantic theorems are informally (set-theoretically) derived, and then an axiomatic theory is presented from which analogues of the semantic theorems are formally derivable.
Peter Carruthers, “Human and Animal Minds: The Consciousness Questions Laid to Rest” (NDPR, 2020)
Critical review of Peter Carruthers’ 2020 argument for the following three claims 1) a first-order global workspace theory of consciousness is true, 2) for most animals (and some humans) there is no fact of the matter about whether they are phenomenally conscious, and 3) it does not matter morally that there is no fact of the matter about animal consciousness, because consciousness is not what matters morally.
Ludwig Jaskolla, “Real Fourdimensionalism: An Essay in the Ontology of Persistence and Mind” (NDPR, 2018)
Critical review of Ludwig Jaskolla’s novel case for a panpsychist theory of persistence.
Michael Pelczar, “Sensorama: A Phenomenalist Analysis of Spacetime and its Contents” (NDPR, 2015)
Critical review of Michael Pelczar’s novel, scientifically informed argument for idealism.