Talks

Not Gizmos, but Ghosts — Margaret Boden

35 min read

About this talk

Speaker: Margaret (Maggie) Boden
Event: 2018 Religion & Science Conference: Artificial Intelligence & Consciousness, Cambridge Muslim College (funded by the John Templeton Foundation)
Date: 2018 (conference); video uploaded 2019-01-19
Length: 38:14
Channel: Cambridge Muslim College
Watch: Not Gizmos, but Ghosts: AI & the Mind - Professor Margaret Boden

How to read this page. It has two kinds of content, always marked:

  • 🎙 In the talk. What Boden says, with [MM:SS] timestamps. Only these parts are her views.
  • 📚 Deeper theory. Background from the traditions she draws on, including some she does not name. It explains why she says what she says and who disagrees. It is not her claim unless a timestamp says so.

Suggested route. Summary, Map, Key terms, Part 2 (read once, slowly), Deep dives, “Putting it together”.

See also: searle-boden-chinese-room-debate-1984.md (her main sparring partner on the question “can a program understand?”) and her 2018 lecture at the Future of Intelligence centre. This page does not depend on either.

Speakers

LabelWhoRole
BodenProfessor Margaret (Maggie) Boden, Research Professor of Cognitive Science, University of Sussex (named in the video description)The only speaker heard in the recording
Earlier speakers”Mark Harris” and “the Sheikh” (as the captions give the names)Spoke earlier that morning on science and religion. Boden refers to them. Not heard here
Later speakersA “Ron” (see warning box) and others, “tomorrow”Boden says they may cover the LIDA model and Dennett’s view

Speaker identification confidence: high. The title, channel and description all name her, and the first-person remarks (Sussex-era AI history, knowing Oliver Selfridge) fit her career.

Status note (added 2026). Boden died on 18 July 2025 in Brighton, aged 88 (Wikipedia; University of Sussex obituary). This talk is therefore part of her late public record.

Caption warnings. Auto-captions misspell many names. Corrections used on this page (the first group is certain from context; the second is my inference):

  • Certain: “Gilbert’s Rao / Gilbert Rau / Royal” = Gilbert Ryle. “Dennis” = Dennett. “they caught” = Descartes. “Quanis” = Aquinas. “peleas / Paley’s” = Paley. “Christ College” = Christ’s College, Cambridge. “Selfridge” is right (Oliver Selfridge).
  • Inferred: “the wrong crizzley” is probably Ron Chrisley, a Sussex cognitive scientist who may have spoken at the conference. “Ron’s head” at [22:12] is probably Aaron Sloman, the Sussex/Birmingham AI philosopher who wrote on “the mind as virtual machine” (the captions drop the “A”). “The leader model” is probably LIDA (Stan Franklin’s model of consciousness, built on Global Workspace Theory). “Mark Harris” and “the Sheikh” are not checked.
  • Some sentences are broken by the captions (for example “Selfridge” appears with run-on text at [06:21]). Quotes on this page were checked against the caption text as it stands, so a quote can contain a caption slip.

Factual slips and checks.

  • Boden says Darwin, as an undergraduate at Christ’s, was given the rooms William Paley once had [14:13]. This is a well-known Cambridge story. From memory; not checked.
  • She says Selfridge was grandson of the man who founded Selfridges [07:54]. True as far as I know (Harry Gordon Selfridge). From memory; not checked.
  • She says the mind-body problem “didn’t even begin” to get light until the mid-20th century [19:02]. That is a strong claim. Hobbes, La Mettrie, Diderot and 19th-century physiologists all pushed on it earlier. Read it as “the mid-20th century gave us a new kind of answer”.

Summary

Central question. What is the most important thing AI has given us? Boden’s answer is not the products.

Main thesis. AI is worth most for the ghost, not the gizmos. By “ghost” she means the mind. AI gives us a way to say what a mind is without calling it a physical thing and without calling it a supernatural thing. A mind is what the brain does: an information-processing system, described in abstract terms and implemented in the brain. A computer scientist would say: a virtual machine, or a set of interacting virtual machines.

The argument runs in five steps:

  1. Gizmos versus ghost. Translation apps are useful, and their social effects are big, but they are not the point [00:00]–[01:34].
  2. Ryle’s phrase, kept but softened. “Ghost in the machine” was Ryle’s insult for Descartes’ view. Boden keeps the word “ghost” without the insult. Mind is not the brain, and it is not a mystery either [01:34]–[04:45].
  3. Wonder survives understanding. Many people think explaining a thing kills the wonder. Her Selfridge story (a child, a compass) and her Darwin/Paley example both show that explanation can remove one kind of wonder and leave a deeper one [04:45]–[18:24].
  4. Functionalism plus computation. Philosophers in the 1960s said the mind is what the brain does. AI added a precise vocabulary for “does”, and a way to test ideas by running them [18:24]–[30:09].
  5. A mind is more impressive, not less. AI shows how complex and subtle the mind is. It does not say “we are just machines”. That gives religious people who honour the human being something to stand on [30:09]–end.

Close. “More harmonization than conflict” between AI and a theology that values the human being.

Her stance. She is a scientist who prefers the computational view. She says openly that she is “not fully convinced” it explains consciousness [25:20], though she thinks it is the right direction. She says people with a substantive-soul theology will be even more doubtful than she is. She does not try to argue them out of it.

Map — talking points

#PointTime
1Hype about AI gizmos. They are not the interesting part[00:00]
2The “ghost” is the mind. Ryle’s phrase, and why she uses it without the scorn[01:34]
3AI does not say “the mind is the brain”. Thoughts are information-processing phenomena implemented in the brain[03:09]
4Wonder is a shared source of science and religion. Science aims at explanatory understanding, not wisdom[04:45]–[06:21]
5The assumption that understanding drives out wonder. Source of the “conflict” thesis[06:21]
6Oliver Selfridge and the compass: wonder lost for a child, regained with maths[07:54]–[12:38]
7Darwin, Paley, Aquinas. The argument from design loses force; the wonder stays[12:38]–[17:25]
8Exceptions: literalist theologies for which Darwin does conflict with doctrine[15:49]–[17:25]
9The mind as the new “design” puzzle. Descartes: mind-brain link as a metaphysical mystery[17:25]–[19:02]
101960s functionalists: the mind is what the brain does[19:02]–[20:37]
11AI’s contribution: the mind as an information-processing system, the virtual machine[20:37]–[22:12]
12Consciousness: caused by the brain, neuroscience necessary, yet not the same as a philosophical explanation. She is “not fully convinced”[22:12]–[25:20]
13A substantive-soul theology will be more sceptical still[25:20]
14AI also gives testable computational concepts, like Darwinism’s detailed examples[26:52]–[28:29]
15Running a model shows when you were wrong. It proves nothing alone but is a starting point[28:29]
16The most important result: the mind is far more complex and powerful than we thought (Freud, Henry James had a feel for it)[30:09]–[31:47]
17”Just machines” is a wrong picture: bicycles and engines are easy to understand; computers are not[33:25]–[36:35]
18Close: special, not magical, understandable. “More harmonization than conflict”[36:35]

Part 1 — Before you read

Key terms

  • Ghost in the machine. Gilbert Ryle’s name (1949) for the Cartesian picture of a non-physical mind that “drives” a physical body. He meant it as mockery. Boden keeps the word and drops the mockery [03:09].
  • Gizmos. Her word for AI products: translation, recommendation, anything a business can sell [00:00].
  • Substantive mind or soul. A mind that is a thing of its own kind, a substance, deeply linked to the brain yet not the same as it. She uses “substantive” to mark the strong dualism she thinks Ryle attacked [25:20]. Watch out: in philosophy, “substance dualism” means exactly this. Other kinds of dualism (property dualism) are weaker and she is not discussing them.
  • Functionalism. The view that a mental state is defined by what it does (its role in causing behaviour and other states), not by what it is made of. Boden: the mind is “what the brain does”.
  • Information processing / computation. To describe a system by what it represents and how it transforms those representations, in abstract terms, not in terms of voltages or neurons. This is a level of description, not a claim about metal or flesh.
  • Implemented. An abstract process is implemented when some physical system carries it out. The same program can be implemented in many machines. This word carries the whole argument that mind ≠ brain without being a ghost.
  • Virtual machine. A system defined by its own rules and operations, which runs on another system. A word processor is a virtual machine on top of the hardware. “A collection of … interacting virtual machines” is her image of a mind [22:12].
  • Explanatory understanding vs wisdom. She says natural science aims at the first, not at “knowledge of how to live” [06:21]. This lets religion keep the second.
  • Argument from design. Paley’s argument: a complex designed thing (a watch, an eye) needs a designer. Darwin removed the need for one.
  • Hard to confuse: “mind” vs “consciousness”. The talk mostly explains mind (what thinking, remembering and feeling systems do). It says consciousness is “awfully difficult” even on this view [22:12]. Do not read the talk as “AI explains consciousness”.

People named (a reading list)

PersonWhy they come upStart with
Gilbert RyleCoined “ghost in the machine”The Concept of Mind (1949), ch. 1
René DescartesThe target of Ryle; the mind-brain puzzleMeditations (1641); Passions of the Soul (1649)
Daniel DennettQuoted at the conference: thoughts are information-processing events in the brainConsciousness Explained (1991)
Oliver SelfridgeEarly AI figure; compass story; parallel processing ideas”Pandemonium” (1959)
Aaron Sloman (probable)“The mind is a virtual machine”Sloman & Chrisley, “Virtual machines and consciousness” (2003)
William PaleyArgument from designNatural Theology (1802)
Charles DarwinRemoved the need for a designerOn the Origin of Species (1859)
Thomas AquinasDesign-style argument (fifth way)Summa Theologiae I, q.2, a.3
Sigmund Freud, Henry JamesIntuitively grasped the mind’s subtletyBoden’s own Mind as Machine (2006) for the history

Boden’s own frame (outside the talk, for context)

Boden was a philosopher, psychologist and AI scholar. She co-founded the Sussex cognitive-science programme in the early 1970s (from the Sussex obituary summary above, not checked in detail). Her big works: Artificial Intelligence and Natural Man (1977), The Creative Mind (1990), and the two-volume Mind as Machine: A History of Cognitive Science (2006). All dates from memory; not checked. A recurring move in her work, visible here, is that a computational account makes the mind richer, not poorer. She spent her career arguing against the “just a machine” insult and against the “no machine could ever” insult.

The current debate (2025–26)

  • Her topic is now an engineering question. Boden’s claim that computation is the right kind of explanation for mind has become the working assumption of a research programme. Butlin, Long and colleagues published “Identifying indicators of consciousness in AI systems” (Trends in Cognitive Sciences, 2025). The method takes computational functionalism as a working assumption, draws “indicators” from neuroscience theories of consciousness, and checks AI systems against them. Summaries report that no current system looks conscious, and that no technical barrier to building one that meets the indicators is evident (summary at selfawarepatterns covers the 2023 predecessor; secondary commentary: Medium review, Buttondown summary). I did not read the original paper, so treat the details as secondhand.
  • The sharpest opposing view says life matters. Anil Seth’s 2025 paper “Conscious artificial intelligence and biological naturalism” (Behavioral and Brain Sciences) argues that consciousness depends on a living, organic substrate. If he is right, today’s AI is the wrong kind of thing, and a conscious artificial system would have to differ a great deal from it. This attacks the step Boden herself hesitates over: whether an abstract description is enough (escholarship critique of computational indicators; Seth view as reported in the same search, secondhand).
  • Labs now run “model welfare” programmes. Anthropic launched one in April 2025 and hired a dedicated researcher; its post is “Exploring model welfare”. Reports say the February 2026 system card for Claude Opus 4.6 has a “Model Welfare Assessment” section, and that Google DeepMind, Anthropic and Meta have expanded work on machine consciousness and AI welfare (secondary sources: IEEE Spectrum, Digital Minds in 2025 review). The open method problem, noted by IEEE Spectrum: what a language model says about itself is weak evidence. Several sources conflict on numbers, so I give none. Unverified: I did not open the system card.
  • What this means for the talk. In 2018 Boden’s point was “AI helps us understand human mind”. In 2026 the same idea is used the other way round: “does the system we built have one?” She did not argue that, and this page does not put it in her mouth. Her line that AI does not say we are nothing but machines [33:25] is, though, the thing people now quote when they push back on both extremes.

Part 2 — The intellectual map

Boden spoke for 38 minutes to a mixed audience of scientists, philosophers and theologians, and she lightly touches ideas that fill whole libraries. Eight traditions meet in this talk. Some she names; some she only gestures at; some she leaves out. Each is laid out here with its problem, its chain of thinkers, and how it feeds the talk.

   DESCARTES ─► the mind-body problem ──┐                    ┌── RYLE: "ghost in the machine"
   (substance dualism, 1641)            │                    │   (category mistake, 1949)
                                        ▼                    ▼
   TURING ► NEWELL & SIMON ► MARR ►  FUNCTIONALISM  ◄── "what the brain DOES"
   (computation, 1936-50)  (AI psych)  (Putnam 1967, Fodor 1975)
                                        │
                  ┌─────────────────────┼───────────────────────┐
                  ▼                     ▼                       ▼
         VIRTUAL MACHINES       ARCHITECTURES OF MIND    CONSCIOUSNESS MODELS
         Sloman (1990s-2003)    Selfridge (1959)         Baars GWT (1988)
         "mind = layered        Pandemonium: demons      Franklin LIDA
          virtual machines"     that shout in parallel   Dennett (1991)
                  │                     │                       │
                  └─────────────────────┼───────────────────────┘
                                        ▼
                         BODEN: "AI gives us the GHOST"
                         mind = implemented information processing
                                        ▲
        ┌───────────────────────────────┼──────────────────────────────┐
        │                               │                              │
   WONDER & EXPLANATION        OBJECTIONS TO IT              THEOLOGICAL ANTHROPOLOGY
   Paley ► Darwin              Searle (1980), Nagel (1974)   imago Dei · soul · ruh/nafs
   Aquinas, Selfridge's        Chalmers (1995), Block        Ibn Sina · Ghazali
   compass (the talk)          Seth (2025): life matters     (what "harmony" must answer)

2.1 Descartes and the mind-body problem — the puzzle the talk starts from

The problem it addresses. If the mind is a thinking thing with no size or location, and the body is an extended thing with no thought, how does one move the other? Boden says Descartes saw a close tie to the brain, but the link was “fundamentally mysterious” and even a “metaphysical impossibility” unless God made it so [19:02].

The chain

  1. Descartes, Meditations (1641). Two substances: res cogitans (thinking thing) and res extensa (extended thing). The argument from doubt: I can doubt my body but not that I think. Therefore mind and body are distinct.
  2. Princess Elisabeth of Bohemia, letters to Descartes (1643). The standard objection: a thing with no extension cannot push on matter. Descartes had no good answer. His famous reply, the pineal gland as meeting place, names a location without explaining a mechanism. From memory; not checked.
  3. Later reactions. Occasionalism (Malebranche: God does the linking), pre-established harmony (Leibniz), and the monism of Spinoza, who made mind and body two aspects of one thing. All are ways of living with the same difficulty.
  4. La Mettrie, L’Homme machine (1747). The first loud “we are machines” claim. This is the target Boden has in mind when she says AI does not say “we’re nothing but machines” [33:25]. From memory; not checked.

How it feeds the talk. Boden compares the Cartesian worry to Paley’s: in both cases a wonderful fact (living design, then mind) looked unexplainable without God. Her claim is that AI does to the second puzzle what Darwin did to the first.

2.2 Ryle and ordinary-language philosophy — the insult that became a word

The problem. Is “mind” the name of a hidden thing behind behaviour?

  1. Gilbert Ryle, The Concept of Mind (1949). He calls the Cartesian picture “the dogma of the Ghost in the Machine” and says it makes a category mistake: it treats “mind” as if it were another object alongside body, like a visitor to a university who has seen the colleges and libraries and asks “but where is the university?” Minds, for Ryle, are what people do and are disposed to do (their skills, habits and dispositions). From memory; not checked.
  2. Behaviourism, around the same time (Watson, Skinner). Related but not the same. Ryle’s dispositions are not stimulus-response chains. Many readers still lump them together. The functionalists below reject strict behaviourism because it leaves out the internal states that cause behaviour.
  3. Wittgenstein, Philosophical Investigations (1953). Same suspicion of the “inner object” picture from a different direction.

How it feeds the talk. Boden’s opening [01:34]–[04:45] is a careful move. She keeps Ryle’s point (the mind is not a second, mysterious thing) and keeps his word (ghost) but turns the word to a compliment. Her ghost is a pattern of information processing, not a spirit. That is a friendly reading of Ryle that Ryle would probably have accepted in part. Ryle himself was wary of “inner” mechanisms, so this is Boden’s gloss, not his.

2.3 Functionalism — “the mind is what the brain does”

The problem. If mind is not a spirit and not simply behaviour, what is it?

  1. Turing, “On computable numbers” (1936) and “Computing machinery and intelligence” (1950). Computation can be defined without reference to the stuff that carries it out. A universal machine can imitate any other. This makes “same process, different body” a precise idea.
  2. Hilary Putnam, “Psychological predicates” (1967). The founding statement of machine functionalism: a mental state is defined by its place in a pattern of causes and effects, so pain could in principle exist in a brain, a robot or an octopus (multiple realizability). Putnam later gave up the position, in the 1980s. From memory; not checked.
  3. Jerry Fodor, The Language of Thought (1975). The mind is a symbol-processing system with a syntax that operates on internal representations. This is the “classical” computational theory.
  4. Newell and Simon (Logic Theorist 1956; General Problem Solver 1959). The first programs offered as theories of human problem solving, tested against thinking-aloud data. This is where Boden’s “put it into a computer and see” method [28:29] comes from.
  5. David Marr, Vision (1982). Three levels of explanation: the computational problem, the algorithm, and the physical implementation. Boden’s described in abstract terms … implemented in the brain [20:37] is the same layering. From memory; not checked.

How it feeds the talk. Boden gives the 1960s functionalists the credit for the first move and AI for the second, the vocabulary [20:37]. The logic is simple. If mental states are defined by function, you need a rigorous way to say what a function is. Computation supplies it. That is why she can say mind is not the brain and not a ghost: it is the function, which the brain implements.

2.4 Virtual machines and architectures — a mind has layers

The problem. A computational theory is empty until it says what kind of system a mind is. One program? A pile of rules? Many parallel processes?

  1. Oliver Selfridge, “Pandemonium” (1959). Pattern recognition by many simple “demons” that each shout in proportion to how well the input matches their feature; a higher demon picks the loudest. An early picture of parallel, bottom-up processing, long before neural nets were fashionable. Boden mentions that his ideas in symbolic and connectionist AI still influence consciousness models [07:54].
  2. Aaron Sloman (probable), virtual-machine view of mind. A virtual machine is a level of description with its own causal powers: a spell-checker causes a word to be flagged, though everything physical happens in silicon. Mind, on this view, is a set of layered virtual machines implemented in the brain, which can interact. The key point for the philosophy: virtual machines are real causes without being extra substances. Boden’s wording at [22:12] follows this. From memory and the search summary of Sloman and Chrisley (2003); not checked against the paper.
  3. Bernard Baars, A Cognitive Theory of Consciousness (1988) and Stan Franklin’s LIDA. Global Workspace Theory: many specialist processes work in parallel and unconsciously; what becomes conscious is what wins a competition and gets “broadcast” to the others. LIDA implements this as software, and its design draws on Selfridge’s Pandemonium (the search summary lists pandemonium theory among its sources; LIDA overview). This is almost certainly the most interesting computational model of consciousness she names at [07:54].
  4. Dennett, Consciousness Explained (1991). The “multiple drafts” model: no single place where it all comes together (he calls the idea of one the “Cartesian theatre”). Boden says Dennett’s wording, that thoughts are information-processing events in the brain, can be improved: they are information-processing phenomena implemented in the brain [03:09].

How it feeds the talk. This is the answer to the question a theologian would ask next: “If the mind is a process, how can it be more than a calculator?” The answer is layers. A mind is many interacting virtual machines, which is why Boden can call it “hugely more wondrous” than anyone thought [31:47].

2.5 Wonder and explanation — Paley, Darwin and the “conflict” thesis

The problem. Does a scientific explanation kill the wonder that religion trades on?

  1. Aquinas, Fifth Way (13th c.). Things without intelligence act toward an end, so something intelligent directs them. The ancestor of the design argument.
  2. William Paley, Natural Theology (1802). The watch on the heath: a watch implies a watchmaker; the eye is a finer watch. Boden uses Paley as the typical example of astonishment turned into an argument.
  3. Charles Darwin, Origin (1859). Natural selection gives a mechanism for adaptation without a designer. Boden: the argument from design has “completely lost its strength” [15:49], but “the wonder is still there” [17:25].
  4. Models of the science-religion relationship. A common sorting is conflict (Draper, White in the 19th century), independence (Stephen Jay Gould’s “non-overlapping magisteria”, 1999), dialogue and integration (Ian Barbour, 1990s). Boden’s claim is close to harmony through different questions: science explains how, religion keeps awe and meaning. She notes the exception honestly: literalist readings of Genesis really do conflict with Darwin [15:49]. Classification from memory; not checked.
  5. Wonder in the history of philosophy. Plato (Theaetetus) and Aristotle (Metaphysics I) both say philosophy begins in wonder. Thomas Kuhn’s contrast is unrelated, but Richard Dawkins’ Unweaving the Rainbow (1998) argues the same point Boden makes about the compass: explanation can deepen wonder. From memory; not checked.

How it feeds the talk. Boden’s Selfridge story is a small parable. The child wanted magic. The compass was mechanism. His disappointment was real [11:04], and it ended only when he grew into the mathematics, which let him see a different wonder (the idea of a circle and how a compass uses it). The point: disappointment comes from looking for the wrong sort of wonder, not from understanding itself. She then maps the same pattern onto mind: AI removes one kind of mystery (the ghost as magic) and shows a deeper kind (the complexity of the system).

2.6 Objections from inside philosophy of mind — what the talk does not answer

Boden is brief on this and says she is “not fully convinced” [25:20]. The objections she has in mind are the standard ones.

  1. John Searle, “Minds, brains and programs” (1980). The Chinese Room: a person following rules to shuffle Chinese symbols understands nothing, so running a program is not enough for understanding. Syntax is not semantics. Boden’s own replies are on the companion page searle-boden-chinese-room-debate-1984.md. From memory; not checked.
  2. Thomas Nagel, “What is it like to be a bat?” (1974). Consciousness has a subjective character that objective, third-person descriptions seem to leave out.
  3. David Chalmers, “Facing up to the problem of consciousness” (1995). The hard problem: even a full account of function (the “easy problems”: attention, report, integration) leaves open why any of it is accompanied by experience. This is the exact gap Boden marks at [22:12] and [23:46]: knowing what happens in the brain “is not the same thing” as explaining what consciousness is.
  4. Ned Block, “access” vs “phenomenal” consciousness (1995). Distinguishes consciousness-as-availability-for-use from consciousness-as-felt. Functionalism may explain the first and miss the second. From memory; not checked.
  5. Anil Seth, biological naturalism (2025). Consciousness is bound to life and metabolism; a simulation of a storm does not get wet. See the current debate above.
  6. Integrated Information Theory (Tononi). Consciousness depends on physical causal structure, so a digital computer running the same software could have little or none. The Butlin et al. work sets IIT aside because it conflicts with computational functionalism (per the search summary).

How it feeds the talk. These are what her “not fully convinced” stands on. Notice what she does claim: that the computational approach is “the way we should be going” [25:20] and that it explains mind better than any earlier view. She does not claim it solves the hard problem.

2.7 Theological anthropology — the audience she is speaking to

This tradition is the least developed in the talk, and it is the one the audience came from. Everything below is background, not Boden.

  1. Substantive soul. In Christian, Jewish and Islamic thought the soul has often been described as a real, non-bodily thing. The strength of that claim varies a great deal by school. Boden’s own contrast is between a strong sense (mind or soul as a different substance, “deeply mysteriously connected” to the brain) and her sense (mind as a different kind of description) [25:20].
  2. Islamic thought on the self (from memory; not checked). The Arabic words ruh (spirit) and nafs (self/soul) have layered meanings in the Quran and in later philosophy. Ibn Sina (Avicenna, 11th c.) argued, with the “floating man” thought experiment, that self-awareness does not depend on sensing the body, a Cartesian-looking argument 600 years early. Al-Ghazali (d. 1111), in The Incoherence of the Philosophers, attacked some of the philosophers’ claims about the soul while keeping the soul’s reality. These matter because the host college trains Muslim scholars, but Boden does not discuss them.
  3. Imago Dei. The idea that humans bear God’s image. Boden says that for such a theology, AI’s finding about the mind’s richness gives “extra ground to stand on” [33:25]. Note what this is: an invitation, not an argument. She does not claim AI proves any doctrine.
  4. Religion without the soul-as-thing. Some theologians (for example in the “non-reductive physicalism” camp, such as Nancey Murphy) accept that the mind is tied to the brain and still hold that humans are special. From memory; not checked. That is the group Boden’s “harmonization” fits best.

How it feeds the talk. Her closing bargain: you can keep special, not magical, very worthy of wonder … capable in the end of understanding [36:35]. A strongly dualist theology cannot take the deal in full, and she says so.

2.8 What the talk leaves out

  • The 2018 AI. She speaks of AI as symbolic programs and cognitive models. Deep learning is barely present. A 2026 reader will ask whether large language models fit her “virtual machine” picture at all, since they have no hand-written cognitive architecture. This is a live question, and her talk does not address it.
  • Creativity. Her best-known research (The Creative Mind, 1990) is not mentioned. Her wider position that computers can model, and possibly be, creative is the same bet as the one here: computation is a good way to understand a human capacity without cheapening it.
  • Ethics and risk. She sets “the excitement and worry” aside [30:09] and says gizmos are “not always” good for society. Nothing on jobs, bias or power.
  • Embodiment and life. The 4E tradition (embodied, embedded, enactive, extended: Varela, Thompson, Clark) argues that mind needs a living body in an environment. This is the philosophical ancestor of Seth’s 2025 position.

The thread that ties them together. Descartes sets the puzzle. Ryle says the puzzle comes from a category mistake. Functionalism says the mind is a pattern of doing. Computation makes “pattern of doing” exact. Architectures say what the pattern looks like. Paley and Darwin give the model for how a mystery becomes understood without losing its wonder. The objectors ask whether the pattern is enough for feeling. The theologians ask what is left for them. Boden’s talk sits at the junction and tries to say: nothing needs to be lost.

Part 3 — Deep dives

1. Gizmos versus ghost — what is the point of AI?

🎙 In the talk

  • She names the hype and then sets it aside: “not the gizmos it’s what I call the ghost” [01:34].
  • On products: “it certainly isn’t the gizmos” [30:09]. She grants Google Translate is useful on holiday, and that such tools are “hugely important and influential in our society”.
  • The claim is about kind of value: gizmos are useful; the ghost is about understanding ourselves.

📚 Deeper theory

  • This is the old split between AI as engineering (build useful systems) and AI as cognitive science (use machines to model minds). Boden’s career is on the second side. In 2026 almost all public attention is on the first, so her frame sounds unusual.
  • It matches the “synthetic method” in science: build a model, run it, compare it to the real thing (Part 2.3, Newell and Simon).
  • Link to Part 2.2: the word “ghost” is chosen to be a question about the mind, not an answer.

Where others push back. Many engineers say the cognitive-science branch has been outrun by scaling. Systems that are not built as models of the mind now beat the ones that were. If so, the “ghost” lessons may come from neuroscience and psychology more than from AI.

What she leaves out. Whether today’s gizmos are themselves a new kind of ghost (a mind-like thing). Her frame treats AI as a mirror, not a candidate.

Check yourself

  1. What does she mean by “ghost”, and how does it differ from Ryle’s use?
  2. Why does she call the gizmos “trivial” in the theological context?
  3. In what sense is AI a mirror in her talk, and what changes if the system is itself a candidate mind?

2. The mind as information processing and virtual machine

🎙 In the talk

  • She says AI does not see the mind as a physical mechanism or as the brain: “it doesn’t see the mind as a physical mechanism it doesn’t see the mind as the brain” [03:09].
  • Functionalists said “the mind is what the brain does” [20:37]. AI added a way to read that sentence: “the mind is an information processing system” described “not in physical terms but in abstract terms” [20:37].
  • The computer scientist’s version: “the mind is the virtual machine” [22:12], more exactly a collection of “very different and interacting virtual machines” implemented in the brain.
  • She improves on the quoted Dennett line: thoughts need not be “events in the brain”; they are information-processing phenomena “implemented in the brain” [03:09].

📚 Deeper theory

  • Why “implemented” matters. If mind were identical to brain events, then any difference in the brain would be a difference in mind (type-identity theory, Place and Smart in the 1950s). Functionalism allows the same mind in a different body. Boden’s wording sits in the functionalist camp (Part 2.3). From memory; not checked.
  • Why layers matter. Software is real and causal, yet you cannot find it by taking the computer apart. Sloman’s point (Part 2.4) is that this is not a trick: high-level causes are real. The mind is as real as a program.
  • A hidden choice. “Abstract” can mean (a) a description we find useful, or (b) a real structure in the world. Realists about computation say (b); sceptics (Searle: “syntax is not intrinsic to physics”) say it is (a) and so cannot make anything conscious. Boden does not take sides in the talk.

Where others push back.

  • Searle: computation is observer-relative; a program does not understand (Part 2.6).
  • Embodiment theorists: a mind is a body-environment loop, not software.
  • Connectionists: brains are not run as “virtual machines” with clean levels. The levels blur. Sloman would reply that layering can be messy and is still real.

What she leaves out. Which computation. A thermostat is an information processor. So is a language model. The talk does not say what kind of processing makes a mind.

Check yourself

  1. Why is “implemented in the brain” different from “is the brain”?
  2. Give an example of a virtual machine and say why it is not “a ghost”.
  3. What would a sceptic like Searle say about “the mind is the virtual machine”?

3. Wonder survives understanding — Selfridge, Darwin, Paley

🎙 In the talk

  • The conflict thesis she attacks: “many people think assume that understanding drives out wonder” [06:21].
  • The compass story: the child wanted something magical and was “horribly disappointed” [11:04]. Later, the same object turned out to carry real wonder.
  • Darwin: “the astonishment may have gone but the wonder is still there” [17:25].
  • The design argument has “completely lost its strength”, unless one’s theology makes explicit claims that conflict with Darwin [15:49].
  • Her sorting: such conflicts exist for literalist theologies, but “in general” religion “is not undermined by biological Darwinian science” [17:25].

📚 Deeper theory

  • Two kinds of wonder. The first is the astonishment of ignorance (“how could this possibly exist?”). The second is appreciation of structure (“look how this works”). Explanation removes the first and can feed the second (Part 2.5). Scientists and poets have said this many times; the Dawkins title mentioned in Part 2.5 is one example.
  • The compass is a good test case. It is fully understood and still elegant: a rigid constraint producing every circle. That is the shape of the claim she wants for the mind.
  • Where she is careful. She does not say wonder must survive. She says it “sometimes” does not [06:21] and that for some people Darwinism “seems to have that effect”.
  • Link to Part 2.1: Descartes also saw the mind as a wonder only God could explain, so she is applying the Paley template to a new case.

Where others push back.

  • Disenchantment theory (Weber): modern explanation does shrink the sacred. Wonder as aesthetic survival is not the same as the sense of the sacred that religion offers.
  • Some theologians: design arguments were never the core of faith, so “losing” the design argument costs little, and a harmony claim built on it is weak. (This is a view among some thinkers; not tied to a named person here.)
  • Religious critics of the mind case: the Darwin example is about life’s origin; the mind case is about experience and meaning, which is harder to give up.

What she leaves out. Whether the “wonder” of the mind, once explained as a virtual machine, includes the felt side. She admits that is the open part.

Check yourself

  1. What is the difference between astonishment and wonder in her usage?
  2. Why does she say the design argument fails for most theologies but not all?
  3. Is the mind a better or a worse test case for “explanation keeps the wonder” than evolution? Why?

4. Consciousness — what AI does and does not claim

🎙 In the talk

  • She says it is “awfully difficult” to explain consciousness even on this view [22:12], and that it is caused by the brain: “there’s no question whatsoever” [22:12].
  • On neuroscience: it is necessary, but saying “what is going on” when we are conscious “is not the same thing” as giving a philosophically satisfactory account of what consciousness is [23:46].
  • Her own position: “I’m not fully convinced by it” [25:20], though it is the right direction.
  • She says AI offers testable ideas: “it shows you that you weren’t right” when a model gives the wrong output [28:29]. A model that matches human data is “very interesting and very encouraging” but “it doesn’t prove it” [28:29].

📚 Deeper theory

  • Explaining versus locating. Finding the neural correlates of an experience tells you where and when; a theory has to say why this and not that (Chalmers, Part 2.6).
  • Falsification by simulation. Running a model is a real test: if it fails to produce the behaviour, the theory is wrong (Part 2.3). But a model that succeeds is only one of many possible mechanisms. This is underdetermination, which is why she asks for brain and psychology data as well.
  • Descartes’ hint. She notes he said that each kind of conscious thought must have something going on in the brain and that people should look for it [23:46]. She credits him with starting the search, a generous reading.
  • The 2026 shift. The same testing logic now drives the indicator approach to AI consciousness (see Part 1). Whether passing indicators means anything is exactly the point she calls unproven.

Where others push back. Illusionists (Dennett, Keith Frankish) say the hard problem is badly posed and there is nothing more to explain once function is explained. Dualists and biological naturalists say the opposite: function never reaches feeling. She sits between.

What she leaves out. What evidence would change her mind. She gives no test for consciousness.

Check yourself

  1. Why does she say neuroscience is necessary but not sufficient?
  2. What does a computer model show when it fails? When it succeeds?
  3. How does the 2025–26 work on AI-consciousness indicators use her “test it by running it” idea, and where does it go past what she said?

5. “We’re not just machines” — the closing argument

🎙 In the talk

  • AI is not telling us “we’re nothing but machines” [33:25].
  • It is saying “there’s no magic” and that we are “in principle systems that can be understood by some sort of science”, with computation, informed by neuroscience, in the explanation [33:25].
  • The mind is “hugely more impressive hugely more wondrous and awe-inspiring than we ever realized” [31:47]. Freud and Henry James sensed this; the experimental psychologists did not.
  • Bicycles and engines are easy to understand because “you can see the bits” [35:01]. Computers of the current kind are “a very very very different degree of Awesomeness” [36:35].
  • Her offer: keep human beings “special”, “not magical”, “capable in the end of understanding” [36:35]. Result: “more harmonization than conflict”.

📚 Deeper theory

  • The meaning of “just”. “Just a machine” assumes the machine is a simple one, like the pre-computer devices she lists. It is an old rhetorical trick of reductionism: pick the smallest comparison. The reply (compare to the right class of machine) is hers.
  • Echoes. La Mettrie’s “man a machine” (1747) was meant to scandalise. Boden’s is a defence of the human by raising the machine. Compare Dennett’s “free will worth wanting” move: change what “machine” implies and the threat shrinks. From memory; not checked.
  • A theological reading. Awe at complexity is not a proof of a Creator, and she does not say it is. She says AI gives religious thinkers “extra ground to stand on” [33:25] and no more than that.
  • Risk. If a human-level understanding of mind is possible, then copying the mind is possible too, which makes the question of AI moral status practical (Part 1, current debate).

Where others push back.

  • Strong dualists: she has changed the subject. The question was never “is the mind complicated?” but “is it a different kind of thing?”
  • Humanists and critics of AI: the framing “machines can be wondrous too” can be used to excuse treating people as machines in workplaces and schools. (The principle is a general worry, not tied to a named speaker here.)
  • AI-welfare researchers: if machines are in the same category, the “special” part is harder to keep. This tension is an open one, not hers.

What she leaves out. The social side of “just machines”: who gets to say people are, and for what purpose.

Check yourself

  1. What is wrong with “just a machine”, on her account?
  2. What exactly does she offer a religious audience, and what does she not offer?
  3. Can “special but understandable” survive if the same kind of system can be built?

Putting it together

Mental model. Boden’s talk makes one trade. She gives up the mysterious ghost and keeps the real one. A mind is a layered set of information-processing activities (virtual machines) that a brain runs, as a program is something a computer runs. This view is exact enough to test on a computer, so we learn by building models and watching them fail. What we learn is how much the mind does, which makes it more wondrous, not less. She says nothing here shows that this is enough to explain consciousness, and she says she is “not fully convinced”. She offers religion not a proof but a possible peace: you can keep the human being special and stop asking for magic.

Open tensions

  1. Mind vs consciousness. The talk explains mind well and leaves consciousness open. The 2026 debate lives exactly in that gap.
  2. “Abstract” description. Is information processing real structure or a convenient description? Everything depends on this and she does not rule.
  3. Harmony or truce? “More harmonization than conflict” fits weak dualism and non-reductive views. A strong-soul theology gets sympathy, not agreement.
  4. Mirror or candidate. In 2018 AI is a tool for studying human minds. In 2026, labs ask whether the tool might have one.
  5. Which AI? Her examples are symbolic programs and cognitive architectures. Whether large neural systems share the “virtual machine” structure is open.

Reading path

  • Start: Ryle, The Concept of Mind, ch. 1. Then Boden’s own Mind as Machine, introduction. Then the Chinese Room companion page here.
  • Then: Chalmers (1995), Dennett (1991, early chapters), Sloman and Chrisley (2003).
  • Deeper: Baars (1988); the LIDA papers; Seth (2025) against Butlin et al. (2025) side by side; on the theology side, Ibn Sina’s floating man passage and a modern non-reductive physicalist such as Nancey Murphy. (Titles from memory; not checked.)

Graph View

Backlinks