About this talk
Speaker: John Searle and Margaret Boden (host Ted Honderich)
Event: “Voices” (TV programme, 1984), debate on artificial intelligence and the mind
Date: 1984 (programme year from the video description; exact broadcast date not checked)
Length: 57:34
Channel: Philosophy Overdose (uploaded 2022-10-26)
Watch: Debate on AI & Mind - Searle & Boden (1984)
How to read this page. Two kinds of content, always marked:
- 🎙 In the talk. What Searle, Boden or the host say, with
[MM:SS]timestamps. Only these parts are their views. Words in quotation marks are exact caption text. - 📚 Deeper theory. Background from the traditions behind the debate, and later history. It explains why they argue as they do. It is not their claim unless a timestamp says so.
Suggested route. Summary, then Map, then Key terms, then Part 2 (read once, slowly), then Deep dives in order, then “Putting it together”.
See also (written in parallel, not required for this page): boden-not-gizmos-but-ghosts.md (Boden, 2018) and a 2018 Boden lecture at the Future of Intelligence centre. Both show where her thinking went after this 1984 exchange.
Speakers
| Label | Who | Role |
|---|---|---|
| Host | Ted Honderich (named in the video description, not in the captions) | Philosopher; hosts Voices. Opens the programme and chairs |
| Searle | John R. Searle, Department of Philosophy, University of California, Berkeley | Introduced as giving “this year’s” Reith Lectures for the BBC, and as author of “Minds, Brains and Programs”. Defends the Chinese room |
| Boden | Margaret (Maggie) Boden, University of Sussex | Introduced as a member of the British Academy and the leading figure in AI in Britain. Author of Artificial Intelligence and Natural Man and Minds and Mechanisms. Defends AI as a way to study the mind. Does not claim to be a strong-AI supporter |
Speaker identification: high confidence. The captions have no speaker labels. I assigned each turn from context: the host asks the short questions and uses “Professor” for both; Searle tells the room story in the first person and calls Boden “Maggie”; Boden talks about “my” interest in the human mind and about the Yale programs. A few turns at [35:46]–[37:18] are ambiguous (see the Turing-test slip below).
Caption warnings. Auto-captions; names are often wrong. Corrections used on this page:
- “Bowden” = Boden (throughout).
- “Dwyane s of artificial intelligence” [03:06] = probably “doyenne” (senior figure).
- “wreath lectures” [03:06] = Reith Lectures (BBC). “a first-class Rao it was caused by” [03:06] is garbled and I could not decode it.
- “Shank / Shanks” = Roger Schank (Yale story-understanding program). “Simon and Newell” is correct (Herbert Simon, Allen Newell).
- “touring test” = Turing test.
- “Mar / Ma” = David Marr (vision). “Jeffrey Hinton” = Geoffrey Hinton.
- “Lynette” [34:13] = Doug Lenat (the program is his AM, “Automated Mathematician”; spelling of the program name is from memory; not checked against the audio).
- “Bernard 11 the theater critic” [32:38] = probably Bernard Levin (Times critic; from memory; not checked).
- “Professor Cyril” [48:18] = Professor Searle.
- “slate of hand” [15:26] = sleight of hand. “Watney’s cans” is a British beer brand (Watney’s).
- “Sharington” [38:53] = Sherrington (Charles Sherrington, neurophysiologist).
- “Prince / probes” [35:46] is unclear in the captions and not corrected.
Slips in the talk (checked).
- At [35:46] a speaker says the Lenat program is “certainly passing the touring test”. It did not. AM was never run as a Turing test; Turing’s test is about conversation with a human judge. The remark is loose, and the captions do not show who said it.
- Boden at [35:46] says AM “came up with one theorem which was a new theorem in the theory of numbers”. This claim is contested. Ritchie and Hanna (1984, Artificial Intelligence 23) argued that AM’s results depended on hidden choices in its rules, and Lenat later conceded that much of the “interesting” structure was read into its output by outside observers (summary of the dispute). Wikipedia’s account says Lenat claimed AM rediscovered Goldbach’s conjecture and the fundamental theorem of arithmetic; I did not find a source that settles the “new theorem” claim, so treat it as unverified.
- Honderich at [00:01] says “Galileo” took us from the centre of the universe. Copernicus is the usual name for that step (from memory; not checked). Minor.
- Searle at [38:53] says Leibniz compared the brain to a mill. True (the mill thought experiment in Monadology, §17; from memory; not checked).
Summary
Central question. Is the mind just a computer program? Do computers think or understand, or do they only look as if they do? And whatever the answer, is AI still the best way to study the human mind?
Searle’s thesis. A program is made only of formal rules for shuffling symbols (syntax). Understanding needs meaning (semantics). Syntax alone never gives semantics. His proof is the Chinese room: a man who knows no Chinese follows English rules to hand back Chinese symbols, so well that he passes for a native speaker. He still understands nothing, and a computer has nothing more than he has [12:21]. Searle adds a second premise: brains cause minds [27:58]. So whatever else could cause a mind must have powers equal to the brain’s. A program on its own does not.
Boden’s position. She agrees the Yale restaurant program understands nothing [13:54]. She disagrees about the general lesson. She adds that a system must be causally plugged into the world to get meaning, and she doubts that anyone has a clear account of how meaning works in humans either [15:26]. She says in principle a machine could have a mind, but in practice she thinks we will never build one like HAL in 2001 [07:44]. Her moral view is that we should not grant machines a place in our “moral universe” [06:12]. Most of all, she says she is a weak-AI person: she wants AI as a method for studying the human mind [31:05].
Where it ends. They agree on a lot: the Yale program has no semantics; intentionality is the central topic of psychology; they differ on whether computational ideas are the best way to study it. Searle says AI changes the subject and offers a formal analog. Boden says science moves by analogy and computation is the most promising analogy. Honderich closes on no agreement. Searle’s last line is “we’re betting on different horses” [56:12].
Stance to note. Searle is not saying machines could never think. He says a program is the wrong kind of thing, and he leaves open that some physical machine with the brain’s causal powers might think (“maybe martians do it by having a lot of green slime”, paraphrase of [28:00]–[29:30]). Boden is not a strong-AI believer, but she does not accept that the Chinese room closes the case.
Map — talking points
| # | Point | Time |
|---|---|---|
| 1 | Host’s opening: AI claims the mind is software; hardware does not matter. Are minds just programs? | [00:01]–[03:06] |
| 2 | Introductions of Boden and Searle | [03:06] |
| 3 | Boden: what an AI program is (rich representations, the “in between” that behaviourists ignored) | [04:39] |
| 4 | Boden: words like consciousness, purpose, freedom: the difficulty is moral, not factual. In practice she doubts a HAL will be built | [04:39]–[07:44] |
| 5 | Searle introduces the Yale story-understanding program and the restaurant script | [07:44]–[09:16] |
| 6 | The Chinese room in the first person. Syntax is not semantics. Adding more symbols only gives more symbols | [09:16]–[13:54] |
| 7 | Boden’s reply 1: right about Schank, wrong to generalise (“a bit naughty”). Needs causal link to world. Semantics is unclear in humans too | [13:54]–[15:26] |
| 8 | Boden’s reply 2: “shuffling symbols” misdescribes rich programs; Marr and Hinton’s dedicated vision machines | [15:26]–[17:02] |
| 9 | Searle: complexity is a red herring. A program is formal by definition | [17:02]–[18:36] |
| 10 | Searle on the robot reply (program plus cameras and arms). Chinese room goes inside the robot’s skull | [18:36]–[21:43] |
| 11 | Boden: the homunculus is only a subsystem; two kinds of causation, physical and computational | [21:43]–[24:50] |
| 12 | Searle’s two premises: syntax is not semantics; brains cause minds. Therefore a machine needs the brain’s causal powers | [24:50]–[26:24] |
| 13 | Boden: nothing magic about protoplasm. Searle: I never said only brains | [26:24]–[29:30] |
| 14 | Beer cans: Searle’s “I want to say it just seems to me incredible”. Boden: “in principle”, but not for moral reasons | [29:30]–[31:05] |
| 15 | Boden’s actual interest: weak AI, the mind, new hypotheses that “run” | [31:05]–[32:38] |
| 16 | Creativity: Boden’s case. Lenat’s AM program; creativity as exploring structures | [32:38]–[35:46] |
| 17 | Searle: even then, simulation is not the real thing. Is it a useful tool or the key? Brain metaphors through history | [35:46]–[38:53] |
| 18 | Searle: “no shred of evidence” for the computational model of psychology. AI distracts from intentionality | [38:53]–[43:37] |
| 19 | Boden agrees intentionality is basic; AI is a stick against behaviourism; intentionality is not consciousness | [43:37]–[45:11] |
| 20 | Searle: AI changes the subject. Boden asks for a better historical example. Freud and the pun on “symbol” | [45:11]–[48:18] |
| 21 | Rutherford’s atom as a solar system: how analogy works. “Promissory notes” | [48:18]–[51:28] |
| 22 | Host presses; Boden says programs with causal grounding come “interestingly closer to intentionality” | [51:28]–[53:03] |
| 23 | Searle charges inconsistency and question-begging. His two-level research programme (brain, intentionality) | [53:03]–[56:12] |
| 24 | Close: “betting on different horses” | [56:12] |
Part 1 — Before you read
Key terms
- Program. In this debate: a set of formal rules for moving symbols about. Boden’s definition at [04:39] is “a set of rules for basically moving symbols about”, with AI programs being rich sets of rules, containing a representation of the world.
- Syntax vs semantics. Syntax: the shape and arrangement of symbols and the rules for combining them. Semantics: what the symbols mean or refer to. The whole Chinese room argument is a claim that the first can never produce the second.
- Formal / uninterpreted. A symbol is formal if a rule treats it by shape alone. Uninterpreted means no meaning is attached. Searle: computers operate on “uninterpreted formal symbols”.
- Strong AI vs weak AI. Strong AI (Searle’s term, 1980): an appropriately programmed computer literally has a mind, with understanding and other cognitive states. Weak AI: the computer is a powerful tool for studying the mind. Searle accepts weak AI as a tool and rejects strong AI. Boden says her own view is weak AI [31:05]. (Careful: Searle’s “weak” is his label; Boden’s position goes beyond a mere tool, as Searle notes at [37:18]–[38:53].)
- Intentionality. The “aboutness” of mental states: a belief is about something, a fear is of something. Searle at [40:27]: the ability of mental states to represent objects and states of affairs in the world. Trap: in everyday English “intentional” means “on purpose”. Here it does not. Intentionality is not consciousness either, and both speakers agree on that [43:37]. The word comes from Franz Brentano (1874) and the phenomenological tradition (see 2.2).
- Turing test. Alan Turing’s 1950 test of machine intelligence by conversation. Searle’s version [10:48]–[12:21] says it can be passed by a person who understands nothing, so it cannot prove understanding.
- Homunculus. “Little man”. In the room, Searle is the little man inside the computer. Boden’s reply at [21:43] is that he is only one subsystem.
- Instantiate / multiple realisation. The same program can be run on different hardware. Searle at [17:02]: the appeal of AI is that the same program can be “instantiated in an indefinite range of different kinds of hardware”. This is functionalism’s central claim (2.3).
- Causal powers. Searle’s phrase for what the brain has: the ability to actually produce mental states. A simulation of the brain lacks this, just as a simulated rainstorm does not wet anything (a Searle analogy used in his written work; not said in this talk).
- Moral universe. Boden’s idea at [06:12]: to say a robot really has purposes is to bring it into the group of beings whose purposes you must respect.
People and works named (a reading list)
| Name | Why they come up | Start with |
|---|---|---|
| Alan Turing | The test Searle attacks | ”Computing Machinery and Intelligence” (1950) |
| Roger Schank | Yale story program with restaurant “scripts” | Schank and Abelson, Scripts, Plans, Goals and Understanding (1977; from memory; not checked) |
| Newell and Simon | Strong AI claim that the computer understands in the literal sense | Physical symbol system hypothesis (1976; from memory; not checked) |
| David Marr | Vision programs with levels of representation (“primal sketch”) | Vision (1982) |
| Geoffrey Hinton | Cited by Boden for vision machines | His early vision/network work (from memory; not checked) |
| Doug Lenat | The AM program that explores mathematical ideas | ”AM” (1976–77); Ritchie and Hanna’s critique (1984) |
| Sigmund Freud | Searle and Boden argue about his “symbols” | The Interpretation of Dreams (1900), Psychopathology of Everyday Life (parapraxes) |
| William James | Searle: uses intentionality without the jargon | Principles of Psychology (1890) |
| Ernest Rutherford | The “planetary atom” analogy | Rutherford model (1911; from memory; not checked) |
What each speaker is known for
Searle published “Minds, Brains, and Programs” in the journal Behavioral and Brain Sciences in 1980 with a long set of replies by AI researchers and philosophers (the host mentions 27 replies, [03:06]). He had just published Intentionality (1983), his book on the topic [45:11], and gave the BBC Reith Lectures in 1984, which became Minds, Brains and Science (the first Reith series by a philosopher, per Harvard UP / book notes). He died on 17 September 2025, aged 93 (Guardian obituary via inkl; Daily Nous).
Boden founded the Centre for Cognitive Science at Sussex, wrote on AI, creativity and the history of cognitive science, and trained in philosophy, medicine and psychology (from memory; not checked). She died on 18 July 2025, aged 88, in Brighton (Wikipedia, citing NYT and Guardian obituaries; Sussex announcement). Both have died since the programme aired, so this clip is now part of the record of their views.
The current debate (2025–26)
Searle’s argument is 45 years old and still the first thing people reach for when someone asks if a large language model “understands”. Where things stand:
- The argument is used against LLMs, and it is contested there. One 2026 philosophy paper argues LLMs intensify the Chinese room problem because they can produce the look of “answerability” without any real stake in being wrong (Jovanović, “From Chinese Room to Language Models”). A January 2026 preprint argues the opposite: that applying the room to LLMs is “a ritual of dismissal”, not an argument (Zenodo preprint). A conference paper from October 2026 proposes that LLMs have at least functional understanding and moves the open question to whether they have a first-person perspective (Sciforum, Philosophies conference). All three are preprints or conference papers, so treat them as opinion.
- The octopus. Bender and Koller (ACL 2020) say a system trained only on the form of language cannot learn meaning, and give the “octopus” thought experiment for it. It is the Chinese room restated for text-only models (paper record; critique: Gubelmann 2023 via arXiv listing). Boden’s “causal plugging into the world” at [15:26] is the same worry in older language, and multimodal and embodied models are the live test of it.
- No new word from Searle. My search found no 2025–26 reply by Searle himself about LLMs. I did not check his last published work.
- Obituaries have reopened the question. Both deaths in 2025 produced retrospectives, including Philosophy Now on Searle and Leiter Reports on Boden.
What to carry back to the transcript: in 2026 the live fight is no longer “can programs think?” but “does understanding need grounding in a world, a body, or consequences, and can text alone supply it?” That is exactly the point where Boden and Searle half agree at [13:54]–[15:26].
Part 2 — The intellectual map
The debate is 57 minutes, but behind it sit about eight bodies of work. Searle and Boden both assume you know them. Here they are in the order that makes the argument easiest to follow. Sections 2.1 to 2.8 are 📚 background; nothing in them is the speakers’ own claim unless a timestamp says so.
MIND AS COMPUTATION MEANING AND ABOUTNESS
Turing · cognitive revolution Brentano · Frege · Husserl
Newell & Simon · functionalism Searle's Intentionality (1983)
│ │
▼ ▼
STRONG AI ◄────────── CHINESE ROOM ──────────► SYNTAX ≠ SEMANTICS
"mind = program" (Searle 1980) "brains cause minds"
│ ▲ │
│ ┌──────────┴──────────┐ │
▼ │ THE REPLIES │ ▼
BODEN'S WEAK AI │ systems · robot · │ BIOLOGICAL NATURALISM
model the mind │ brain-simulator · │ (powers of the brain)
Marr · Schank · │ other-minds · │ │
Lenat · creativity │ symbol grounding │ ▼
│ └─────────────────────┘ consciousness, qualia,
▼ later work (Searle)
EXPLORING A SPACE OF IDEAS
(Boden, Creative Mind, 1990) BACKGROUND: behaviourism → cognitive science
BRAIN METAPHORS: mill → switchboard → computer → ?
2.1 Mind as computation — how “the mind is a program” became respectable
The problem it addresses. Before the 1950s, psychology in the English-speaking world was mostly behaviourism: study stimuli and responses, say nothing about what goes on inside. How can you study inner states without being unscientific? Computation offered an answer.
- Alan Turing, “On Computable Numbers” (1936) and “Computing Machinery and Intelligence” (1950). The 1936 paper defines what a machine that follows rules can compute. The 1950 paper replaces “can machines think?” with a conversation game. If the machine’s answers cannot be told from a human’s, call it intelligent. Searle’s whole argument at [10:48]–[12:21] is that this test measures the wrong thing.
- The cognitive revolution (1950s). Miller, Chomsky, Newell, Simon and others argued that minds should be studied as information-processing systems. Chomsky’s 1959 review of Skinner’s Verbal Behavior is the usual landmark (from memory; not checked). Boden’s Mind as Machine (2006) is the long history she wrote of this; it came after this talk. Her point at [04:39] that behaviourists “totally ignored” the middle between input and output is this revolution in one sentence.
- Newell and Simon, the physical symbol system hypothesis (1976). Claim: a physical system that manipulates symbols has what it takes for general intelligence, and intelligence needs nothing more. Searle quotes them at [09:16]: they say the understanding we attribute to the computer is the same as we give to humans. Searle calls that group “strong AI”. Boden says, with some distance, that she is “not sure” she is in that group; Searle himself says “I’m really not sure Maggie is one of them” [09:16].
- Roger Schank’s scripts (1970s). A script is a stored outline of a stereotyped event (“eating at a restaurant”: enter, order, eat, pay). Given a story with a gap, the program fills it from the script. That is how the Yale program says the man did not eat the hamburger [07:44]–[09:16]. Scripts are also the source of the old complaint that story understanding needs endless common-sense knowledge.
- David Marr, Vision (1982). Marr proposed that any information-processing system can be described at several levels: what problem it solves, what steps solve it, and how the hardware does it (from memory; not checked). His “primal sketch” is the first stage of his vision theory. Boden uses him at [17:02] and [23:16]: each level is made of symbols defined from the level below, and the bottom level is the eye acting as a “physical transducer” [24:50]. Marr is also her model of a computational theory that tries to fit neurophysiology at a detailed level [54:37].
How it feeds the talk. Boden is speaking from inside this tradition, but only partly. She accepts the symbol-and-level vocabulary as a research method. She does not sign up to the claim that a program is enough for a mind.
2.2 Intentionality — the old problem at the centre
The problem it addresses. How can a thought be about something that is not there, or does not exist? A pattern of neurons, or of marks on paper, does not seem to be about anything in itself. Searle says this is the foundation problem of psychology [40:27].
- Franz Brentano, Psychology from an Empirical Standpoint (1874). Revived the medieval idea of “intentional inexistence”: every mental act is directed at an object. He took this to mark the difference between mental and physical phenomena.
- Gottlob Frege, “On Sense and Reference” (1892). Separates the sense of an expression from what it refers to. This is the philosophy behind “semantics” as Searle uses it: meaning is not in the symbol’s shape.
- Edmund Husserl and phenomenology (1900–). Consciousness is always consciousness of something. Searle’s claim that intentional states can be conscious or unconscious [43:37] is a cousin of the later phenomenological debate about whether intentionality needs consciousness. Boden also says they are not the same [43:37].
- Wittgenstein and Ryle (1940s–50s). Both attack the idea of an inner “something” that gives words meaning. Ryle’s “ghost in the machine” (1949) is the phrase the host plays on at [01:34]: “the mind is yet less than a ghost in the machine” — here meaning that on the AI view the mind is not even a ghost but an abstraction.
- John Searle, Speech Acts (1969) and Intentionality (1983). Searle’s own theory: intentional states have conditions of satisfaction (a belief is satisfied if the world is as it says) and a direction of fit. The 1983 book is the “philosophical account” he mentions at [45:11]. In it, there is a “Background” of skills and assumptions that is not itself a set of rules (from memory; not checked).
- Hubert Dreyfus, What Computers Can’t Do (1972). Argued from Heidegger and Merleau-Ponty that human understanding is bodily and situated, so rule-based AI would fail. He is another outsider critic of AI, with a different argument from Searle’s (from memory; not checked).
How it feeds the talk. The two debaters agree that intentionality is the real subject [43:37]. They disagree about whether AI can say anything about it, and Searle charges that AI replaces it with “a formal analog” [40:27].
2.3 Functionalism — why “hardware doesn’t matter” sounded plausible
The problem it addresses. If minds are in brains, and brains are different from silicon, can a machine have a mind? Functionalism says a mental state is defined by what it does (its causal role), not by what it is made of.
- Hilary Putnam, “Psychological Predicates” (1967). Introduced multiple realisability: pain could be realised in humans, octopuses or an alien; the same function, different stuff. Putnam later turned against his own view.
- Jerry Fodor, The Language of Thought (1975). The mind has an inner symbolic code, and thinking is computation over it. This is the strongest philosophical version of the claim that software is what counts. The host’s opening [01:34], “it’s the software that counts”, states the view.
- Machine functionalism and the “beer can” intuition. If only the organisation matters, then anything with the right organisation has a mind: water pipes, men on stools, beer cans. Searle’s beer-can computer [29:30]–[31:05] is a reductio of this. Ned Block’s “China brain” (1978) is the best-known earlier version, in which the population of China plays the neurons (from memory; not checked).
- Daniel Dennett (1942–2024) and Douglas Hofstadter. The two best-known opponents of Searle. Hofstadter and Dennett’s reply in The Mind’s I (1981) called the room an “intuition pump”, with Dennett saying it works by misleading the imagination (from memory; not checked). Dennett’s view is that the intuition that the room does not understand comes from imagining a program that is far too simple.
How it feeds the talk. Searle’s beer cans [30:00]–[31:05] go straight at functionalism. Boden’s answer that she would accept it “in principle” but not for moral reasons [31:05] is a functionalist’s reluctant yes with an ethical brake.
2.4 The family of replies to the Chinese room
The problem it addresses. The Chinese room was published together with replies, and the standard list was soon fixed. In this debate, Boden and Searle run through two of them live.
| Reply (names from the 1980 commentaries) | Idea | Where in the talk |
|---|---|---|
| Systems reply | The man does not understand; the whole system (man, rules, boxes, room) does | Not made by Boden here. Searle’s standard answer (from memory; not checked): let the man memorise everything; he still does not understand. Compare her “subsystem” move at [21:43] |
| Robot reply | Add cameras and arms so symbols are linked to the world | Boden at [14:00]–[15:26]; Searle replies at [20:08]–[21:43] |
| Brain-simulator reply | Simulate every neuron of a Chinese speaker | Not in the talk |
| Combination reply | All of the above together | Searle names it as the “more powerful” argument at [18:36] |
| Other-minds reply | We only judge understanding by behaviour in people too | Boden’s first point at [15:26] (“we don’t understand … what semantics is in the human case”) is close to this |
| Many-mansions reply | Maybe future technology will do it | Boden at [07:44]: “in principle” |
Searle’s pattern in each case: grant the extra machinery, then put himself inside it and show the symbols are still meaningless to him. His strongest card is the robot case at [20:08]–[21:43]. His own position is that it works only if there is “mental representation of that causal relationship”.
The later ones that matter most:
- Stevan Harnad, “The Symbol Grounding Problem” (1990). Symbols in a program must get their meaning from something other than more symbols (as Searle says at [13:54]). Harnad agrees with Searle on the problem and proposes grounding in sensory-motor interaction, a version of Boden’s robot reply.
- Paul and Patricia Churchland, “Could a Machine Think?” (Scientific American, 1990). Churchland’s “luminous room”: a man waving a magnet in a dark room does not produce light, yet light is electromagnetic waves; so syntax-shuffling might be what mind is even if it does not look so. Searle replied in the same issue (from memory; not checked).
- Searle’s later move, “Is the Brain a Digital Computer?” (1990). Claims that “computation” is not a physical fact at all but something assigned by an observer, so the brain is not intrinsically computing (from memory; not checked). Not in the 1984 talk, but it is where Searle takes the “syntax” point.
- Connectionism (1986). The two-volume Parallel Distributed Processing put forward networks that do not follow explicit rules. Searle’s answer was the “Chinese gym” (from memory; not checked). Boden mentions Hinton at [17:02] before this boom.
2.5 Biological naturalism — what Searle puts in place of AI
The problem it addresses. If programs are not enough, what is a mind? Searle needs a positive view that is not dualism.
- The two premises, as he states them. (1) “syntax is not the same as semantics” and (2) “as a matter of biological fact brains cause Minds” [27:58]. The conclusion: anything else that causes minds needs powers equivalent to the brain’s [28:00]–[29:30].
- What the view says. Mental states are higher-level features of the brain, caused by neural processes, like digestion or photosynthesis (an analogy Searle uses in his Reith lectures; from memory; not checked). Mind is real and physical. It is not a program.
- What it avoids. It is not dualism (the mind is not a separate substance) and not behaviourism. It is also not a claim that only carbon can think: he says so at [27:58].
- Neighbours. John Searle and the later “causal powers” position sit near mind–brain identity theories (U.T. Place, J.J.C. Smart, 1950s) and near the later enactivist idea that life matters to mind. Searle would not accept the label (from memory; not checked).
The weakest point. Critics ask what these “causal powers” are, and why a silicon system could not have them. Searle says it is an empirical question. His critics (Boden at [27:58]) say that without an account of what the powers are, the view only names the gap.
2.6 Symbols and metaphors — why Freud and Rutherford appear
The problem it addresses. Both sides argue by analogy. Searle accuses AI of fashion: whatever technology is newest becomes the model of the brain [38:53]. Boden says analogies are how science moves [49:52].
- Brain metaphors across history. Searle lists them: a catapult (Greeks), a mill (Leibniz), a telegraph (Sherrington), hydraulic and electromagnetic devices (Freud), a telephone switchboard (his own childhood), now the digital computer [38:53]. Daugman’s “Brain Metaphor and Brain Theory” surveys this habit (from memory; not checked).
- Freud and the symbol. Freud treats dreams, slips and symptoms as meaningful: something stands for something else. Boden calls him the psychologist closest to AI [46:46]. Searle says that is a pun: Freud’s symbols have meaning, the program’s “symbols” are mere counters [46:46].
- Analogy in science. The Rutherford model of the atom (1911) used the solar system as an analogy. Searle’s move at [49:52] is “at some point you have to actually look at the atom”. Boden’s reply at [54:37] is that Rutherford used the planetary concepts to get started. Philosophers of science (Mary Hesse, Models and Analogies in Science, 1966; from memory; not checked) describe this as the positive, negative and neutral parts of an analogy: which features carry over, which do not, and which are unknown. That is exactly what they are arguing about.
2.7 Creativity as exploration — Boden’s own contribution
The problem it addresses. Creativity looks like the best counter-example to the idea that minds are computational. Bernard Levin (the host’s mention, or Boden’s, at [32:38]) said it was “divine”. Boden thinks that view is defeatist.
- Boden’s argument at [32:38]–[35:46]. There are two popular theories: creativity is divine, or it is special intuition. Both stop inquiry. A third option: creativity is exploring structures in your mind, putting concepts together and changing constraints, “not necessarily consciously”.
- Lenat’s AM (1976). Starts with a small set of concepts in set theory and uses heuristic rules to propose and test new ones, scoring them for “interestingness”. It produced concepts like numbers and primes. Critics (Ritchie and Hanna 1984) said much of the credit belonged to the designer’s choices; Lenat responded in “Why AM and Eurisko appear to work” (1983) (overview).
- Where it went next. In The Creative Mind (1990) Boden sorts creativity into three types: combinational (new combinations of familiar ideas), exploratory (exploring a “conceptual space” defined by rules), and transformational (changing the rules of the space) (from memory; not checked). The talk shows the exploratory idea in its early form. The two later talks beside this page sit in this line.
- The Searle reply. Even if you learn something about human creativity from AM, the program only simulates the production of proofs [35:46]. Boden’s answer is that she has put that question aside: she wants to know whether we understand human creativity better through the analogy [37:18].
2.8 Simulation vs duplication — the idea that holds the argument together
The problem it addresses. A weather simulation does not make anything wet. Why should a mind simulation think?
- Searle’s distinction (in writing). Simulating a process is different from duplicating it. A computer model of a storm, digestion or a fire does not rain, digest or burn. The same for mind. This is what lies behind “it would not be a case of actually having thoughts” [35:46]. (The weather and digestion examples are in his written work, not in this transcript; from memory; not checked.)
- The objection. Some things are what they are by being information processing: a simulated calculator calculates, a simulated chess game is a chess game. Whether mind is more like a storm or more like a calculation is the question. Everyone in Part 2.3 and 2.4 is fighting over it.
- Searle’s “tool” concession. At [37:18]–[38:53] he agrees computers are a useful tool for studying minds “the way they are for avalanches or patterns of rain”. Boden’s reply is that in psychology, computing concepts give more: they are “the best” ones available [40:27]. So the entire second half is about whether the simulation is just a tool or an explanatory model.
2.9 Why 1984 matters — what the debate looked like from inside its time
The problem it addresses. Reading the clip now, it is easy to hear two people arguing about a world that no longer exists: no web, no neural-network boom, no chatbots. The point of this section is to put the period back.
- Symbolic AI at its height. In the early 1980s “AI” mostly meant hand-written rules, scripts and knowledge bases. Expert systems were commercial products. The programs both speakers mention (Schank’s scripts, Lenat’s AM, Marr’s vision levels) are all rule-and-representation systems. Searle’s attack on “programs as rules over formal symbols” fits that kind of system exactly. It fits learned systems less obviously, and critics say that is the main way the world moved (2.4).
- The question was philosophical, not commercial. The host frames AI as a challenge to human uniqueness: our uniqueness has already lost ground (he cites Galileo and the loss of our place as children of God) and AI is the next challenge [00:01]. Today’s version of the question is about labour, safety and law, but the same worry about status sits under it.
- Cognitive science as a young field. Sussex’s Centre for Cognitive Science, which Boden founded (from memory; not checked), was one of the places where philosophers, psychologists, linguists and AI people worked together. Boden’s insistence on “how it relates to the brain and what the hell goes on up here” [31:05] is the founding motive of that field.
- Philosophy as public television. The format matters. Two professors argue for an hour on a general-audience programme, with the host pressing both. Searle’s tone (“I think I can demonstrate conclusively” [09:16]) and Boden’s (“a bit naughty” [13:54]) reflect a culture where an argument could be won or lost on air. Compare the more cautious style of today’s AI commentary.
- What came next, briefly. The Reith lectures the same year; Parallel Distributed Processing in 1986; Dreyfus’s updated criticism; Harnad’s grounding paper in 1990; Boden’s Creative Mind in 1990; deep learning from about 2012; transformers in 2017 (from memory; not checked). Each re-asks the same question at a new level of technical power.
The thread that ties them together. Brentano and Frege ask what meaning is. Functionalism says organisation is enough. Searle says organisation without a causal base is empty. Boden says the base can be plugged in, and meanwhile computing gives better questions. Everything that follows in Part 3 is one of these four moves.
Part 3 — Deep dives
1. The Chinese room — the argument itself
🎙 In the talk
- Searle sets it up with a real program, not an abstract one. He starts from the Yale story program and its “restaurant script” [08:00]–[09:16]. Story, question, answer: the computer says the man did not eat the hamburger though the story never says so.
- He says he can prove it “proves no such thing” and calls its defenders “the gung-ho Fanatics in AI” [09:16]. He notes some honestly say they use the word “understanding” in the literal sense.
- In the room: Chinese symbols in boxes, English rules, no Chinese. Searle explains that “computational rule over formally specified elements” [10:48] is what the rules are. His answers become indistinguishable from native speakers’ [10:48].
- The punch line: “I do not understand a word of Chinese” [12:21] and so neither does any computer, because the computer “has exactly what I have”.
- The key diagnosis: a program is “a syntax and not a semantics” [12:21], rules for “uninterpreted formal symbols”. Humans know what the symbols mean.
- Extra symbols do not help: any effort to give an interpretation “results only in more symbols” [13:54].
📚 Deeper theory
- This is an intuition pump (Dennett’s term for it, critical): it works because you imagine the man’s point of view. It is also a reductio against the idea that running the right program is enough for understanding (2.4).
- The 1980 BBS paper had the same restaurant story with Schank’s program (from memory; not checked). The first-person framing is the argument’s power and its weak spot.
- The bridge from “no understanding in the room” to “no understanding in any computer” rests on program-independence: if understanding came from the program, then it would come to the man who runs it (2.3).
- Searle’s phrase “syntax is not semantics” has roots in Frege (2.2) and in logic: a formal system has proofs and models, and the link between them is not given by the rules.
- Newell and Simon (2.1) are the real targets; Searle says Boden may not be [09:16].
Where others push back
- Systems reply: the man is like the CPU, not the whole machine; the system may understand. Boden’s “subsystem” line at [21:43] is a cousin.
- Speed and scale: a real program that passes the Turing test would have to be enormous. Dennett says our intuitions about a man with a rulebook are not a good guide to what such a system would be like. Searle’s answer at [17:02]–[18:36]: complexity is “a red herring” since the argument is about the kind of thing a program is.
- Understanding vs. the feeling of understanding: Searle uses his own first-person sense that he does not understand Chinese. Critics ask whether first-person feeling is the only evidence.
What they leave out
- Neither says what understanding is. Boden says so openly at [15:26]; Searle relies on the contrast with his understanding of English.
- Consciousness barely appears in the talk. Searle notes that there is “very little” about it in AI [42:03], and he was to spend his later career on it (from memory; not checked).
Check yourself
- In the Chinese room, what exactly does the man have that a computer also has, and what does he lack?
- Why does Searle say that more symbols cannot supply meaning [13:54]?
- Could Searle’s argument work against a program that had no rules, but learned from data? (Connect to 2.4.)
2. Boden’s two replies and the robot — plugging in the world
🎙 In the talk
- Boden begins by agreeing: about Schank’s program she “entirely” agrees it has no sense of a restaurant or a hamburger [13:54]. But Searle’s move from that one program to “any computer” was “a bit naughty” [13:54].
- She adds two things a more genuine program needs: some semantics “plugged in”, and causal contact with the world through sense organs and action. Not by adding formal rules, which “by definition aren’t interpreted in terms of things in the real world” [13:54]. Her example: the system spits out the egg foo young if it has a certain chemical [15:26].
- She turns the demand back: we do not understand “what semantics is in the human case”, so we should not ask AI for a clearer account than we have for ourselves [15:26].
- Second reply: “shuffling symbols” is “a little bit of a slate of hand” [15:26], because real programs are much richer, with internal models of beliefs, desires and intentions of the people in the story [17:02]. She points to Marr and Hinton’s vision machines, built as dedicated machines that react to light like a retina [17:02].
- Searle’s answer is that the robot case is the strongest version and the room applies: put him inside the robot’s skull, with symbols arriving from television cameras; he still “has what any digital computer has” [20:08]. Unless there is “mental representation of that causal relationship”, causation will not give him meanings [21:43].
- Boden’s counter: you “are a subsystem in the robot’s brain”, and nobody claims a subsystem understands syntax [21:43]. She also separates physical causation (light on the retina) from “psychological causation” which she prefers to call computational [23:16]–[24:50].
📚 Deeper theory
- This is the robot reply (2.4) and Harnad’s symbol grounding (2.4). Where Searle and Harnad agree: symbols need grounding. Where they differ: Searle says grounding in sensors is not enough; Harnad says it is.
- Searle’s answer is that the causal link must be mentally represented. That makes meaning depend on a prior mental state, which critics say is circular: it assumes what must be explained.
- Boden’s “two causations” is a version of Marr’s levels (2.1). The physical level sets what is possible; the computational level says what the system is doing.
- The teleosemantic approaches (Millikan, Dretske; from memory; not checked) try to answer her “what is semantics?” question by tying meaning to a biological or learning history. That is the closest the field has to the “causal plugging” she asks for.
Where others push back
- Searle: a subsystem argument does not help, because he is asked to be the whole brain [24:50]. The system reply says the whole room matters; the robot reply says the world matters; each runs into his internalisation move.
- Dreyfus-style critics: adding sensors does not help if the system has no body or stake in the world (2.2).
- Boden against herself: she admits the plugged-in semantics is not really very clear [15:26].
What they leave out
- Learning. Neither considers a system that builds its own representations from experience (the connectionist and now deep-learning case).
- Time and development: a child learns word meanings by acting in a shared world with people. That social side of reference is missing.
Check yourself
- Why does Boden call Searle’s step from one program to “any computer” naughty?
- In Searle’s robot case, what does he say is still missing?
- Which is the stronger point: “you are only a subsystem” or “we don’t know what semantics is in humans”? Why?
3. Brains cause minds, beer cans, and “magic protoplasm”
🎙 In the talk
- Searle states two premises: “syntax is not the same as semantics” and “as a matter of biological fact brains cause Minds” [27:58].
- Boden attacks the biology: she sees “no reason whatsoever to expect that there’s something special or magic about protoplasm” [27:58]. What matters is the function the biochemistry supports.
- Searle: “I certainly didn’t mean to imply that only brains could cause Minds” [27:58]. Anything that did it would need “powers equal to the brain” [29:30]. Martians with green slime, or silicon, could do it in principle.
- He says strong AI goes “any hardware whatever that had the right program” [29:30]. He builds a computer out of old beer cans, windmills and transducers, and says “it just seems to me incredible that this system of beer cans” would have thoughts [31:05].
- Boden accepts the consequence “in principle” but says she would not want to say so, for moral reasons [31:05].
- Later Searle says her position is weaker than her beer-can answer, and he charges inconsistency [53:03].
📚 Deeper theory
- This is biological naturalism (2.5) against machine functionalism (2.3). The beer-can machine is an updated Block “China brain” (2.3).
- The structure is a dilemma for AI: either the program is enough (then beer cans think, which is absurd) or something about the substance matters (then the program is not enough).
- Boden’s reply that the biochemistry matters only through what it does is functionalism. Searle’s reply (that doing things formally is not causing) is the heart of the dispute: does the computation count as a causal power? His later paper on observer-relativity (2.4) is his answer.
- The “no reason to expect magic in protoplasm” move is an argument from parity: don’t favour the brain’s stuff unless you can say what it does. Searle’s answer is that he does not need magic, only physics.
Where others push back
- Intuition vs argument. The beer-can “incredible” is a gut reaction; people once found a thinking mass of wet tissue incredible too.
- Chauvinism. Ned Block called the stricter view “chauvinism” about stuff (from memory; not checked).
- Searle’s own openness. If anything with equal powers could do it, then nothing in principle prevents an artefact, so the real fight is over whether computation is the right level.
What they leave out
- Neither says what the brain’s “causal powers” are.
- The moral side: Boden gives a reason she would not want to count machines as persons, but does not argue it. Searle’s moral line at [53:03] is that beer cans have no conscious or unconscious thoughts.
Check yourself
- Does Searle say only brains can think? Quote what he says.
- What is Boden’s “no magic protoplasm” point, and does Searle answer it?
- Why does Boden say her reluctance to call a machine “free” is a moral, not a factual, one [06:12]?
4. Weak AI, creativity and “the best way to study the mind”
🎙 In the talk
- Boden: “I personally am not primarily interested in computers I’m interested in the human mind” [31:05]. AI is “far and away the most promising way” to ask new questions, with hypotheses that are “much more rich, much more precise” and can be run [31:05]–[32:38].
- Creativity as her test: the “divine” view and the “unanalyzable intuition” view are both unhelpful, and the second “seems to me to be defeatist” [34:13]. Her answer is “a question of exploring the structures in your mind” [35:46], shown by Lenat’s program.
- Searle will grant the program is a useful tool, like for avalanches or the economy of South American countries [37:18]. He asks if there is a stronger claim: that a program is necessary for mentality. He is “suspicious” because of the history of brain metaphors [38:53].
- “I haven’t seen a shred of evidence” that psychology is best seen as computation [38:53].
- Boden presses that computer science gives “the best the most promising Concepts” for the mind [40:27].
- Searle replies that AI “distracts attention” from the fundamental problems [40:27] and that it posits “some intermediate realm” between neurophysiology and mental states [42:03]. Once it was behaviour, now it is the program.
📚 Deeper theory
- This is weak AI as a research programme (2.1, 2.7). Its pay-off is judged like any theory: precision, testability, predictions. The “run it and see” point at [32:38] is a real methodological gain.
- Searle’s charge is layer-skipping. He wants a two-level account (brain, mental states) and sees the program as an invented third level (see 3.6).
- Boden’s later Creative Mind (2.7) answers the “intuition is unanalysable” worry by laying out kinds of creative step.
- Marr’s three levels (2.1) are the best response to the “intermediate realm” charge. For Marr, the computational level states what the system must do, and it is not claimed to be a separate thing in the head.
Where others push back
- Of Boden: AM-type programs often reflect the designer’s choices (Ritchie and Hanna, 2.7). A model that works because its author chose the concepts has not yet explained the author’s creativity.
- Of Searle: his own two-level programme (brain and intentionality) is not worked out either. Boden says that at [43:37]–[45:11] in effect, calling the intentionality account an unfinished promise on both sides.
- Behaviourism’s critics: Searle is dismissive of the information-processing model, but it did produce results that behaviourism could not (in vision, memory, language). He does not argue that point.
What they leave out
- Empirical tests. Neither names an experiment that would decide between them.
- Neuroscience as it was becoming: the 1980s turn to connectionism and, later, brain imaging.
Check yourself
- What is the difference between the “tool” claim and the “necessary” claim that Searle separates at [37:18]?
- What does Boden mean by exploring structures “not necessarily consciously”?
- Why does Searle call the program an “intermediate realm”?
5. Intentionality, Freud and the atom — the closing exchange
🎙 In the talk
- They agree that the basic notion of psychology is intentionality [43:37]. Boden uses this “as a stick to beat” behaviourism [43:37]. They agree it is not the same as consciousness.
- Searle: AI is “changing the subject” [45:11]. It builds a “formal analog” and then studies that.
- Boden asks for one example of a theory that is better done without analogy. Searle names William James and Freud, whose best work uses straightforward Notions of intentionality [46:46].
- Boden: the psychologist closest to AI is Freud, because he worked on how symbols and meanings are transformed [46:46]. Searle: that is “a pun on the word symbol”. Freud’s symbols stand for something; the program’s are uninterpreted counters [46:46].
- Rutherford: Boden says he did not mean the nucleus was emitting light [48:18]. Searle: “of course metaphors are often useful the difficulty is at some point we want to know what is literally true” [48:18].
- Boden admits “that’s another promissory note” about explaining intentionality through computation [48:18]. The host presses that she seems to say computers come close to intentionality; she replies programs with a causal link “begin to have something which is close interestingly closer to intentionality” [51:28]–[53:03].
- Searle’s last move: it is inconsistent with the beer-can phase; it “begs the question”; and he lays out his research programme [53:03]–[56:12]. Two levels: the brain, and intentionality. Putting resources into a “third level” has no reason to give the “biggest payoff” [56:12].
- Boden’s last move: the questions about what mental processes the brain supports are posed at the psychological level and computing concepts are a promising way to pose them [54:37].
📚 Deeper theory
- The “pun on symbol” is a use/mention and equivocation point: symbol in Freud’s and the semioticians’ sense (Peirce, Saussure) means something that stands for something to an interpreter; in computing, it means a token. Searle’s argument depends on the first needing an interpreter.
- The Rutherford exchange is about analogy in theory building (2.6). Hesse’s three parts of an analogy would say: Searle asks which parts of the “program” analogy are negative (do not hold), Boden says the positive parts carry the science.
- The charge of begging the question is sound if Boden’s premise is that computation is the right level. She would say that is a working bet and not a premise. Searle hears the bet but says it conflicts with her beer-can answer.
- The ending is a methodological disagreement. Compare Lakatos on research programmes: you may keep a programme alive while it is “progressive”, and Searle’s challenge is that this one has made only promissory notes.
Where others push back
- The Freud section shows Searle’s argument cutting both ways: Freud’s “hydraulic” metaphors were also analogies that did real work. The history of science has many analogies that were false and still fruitful.
- Searle’s two-level programme could be said to bring back an unexplained gap (how do brain states give intentionality?). Boden’s computational level is meant to fill exactly that gap.
What they leave out
- The time dimension: how these views changed. Neither predicts that deep-learning systems would later do things (translation, image labelling, dialogue) that nobody in 1984 expected.
- Their agreements: the show ends on a clash and does not list what both accept (no program has intentionality, intentionality is central, brains matter).
Check yourself
- What does Searle mean by “changing the subject”?
- Why does Boden think Freud is the closest earlier psychologist to AI, and why does Searle call that a pun?
- State the Rutherford analogy in one line each for Searle’s use and Boden’s use.
Putting it together
Mental model. Two people agree on the target (intentionality, the “aboutness” of mind) and on a verdict (no 1984 program has it), and split on what the verdict means. Searle thinks it shows that a program is the wrong kind of thing: syntax never turns into semantics, brains cause minds, and any rival must match the brain’s powers. Boden thinks it shows that these programs are too thin: add causal contact with the world, richer representations, and a better account of semantics, and computation remains the most fertile way to study the mind. Searle’s strength is a clear argument; his weakness is that “causal powers” is not explained. Boden’s strength is a working research method; her weakness is that the link to real meaning stays a “promissory note”, which she admits.
Open tensions.
- Is there a fact of the matter about understanding, or only about behaviour? (Turing vs Searle.)
- Does grounding need a body, or is any rich enough contact with the world enough? (Boden vs Searle; today’s multimodal models.)
- Is computation a physical fact or something assigned by an observer? (Searle’s later argument.)
- Can a model be a useful theory of the mind without being a mind? (The tool vs necessary-condition split at [37:18].)
- Boden’s moral point: should we count machine “purposes” as real? It is a different question from “can they?”
Reading path.
- Start: Searle’s 1980 paper “Minds, Brains, and Programs” (with the commentaries), then the Stanford Encyclopedia article “The Chinese Room Argument” (not fetched here).
- Then: Searle’s Reith lectures, Minds, Brains and Science (1984); Boden’s Computer Models of Mind (1988); Harnad on symbol grounding (1990); the Churchlands and Searle in Scientific American (1990).
- Deeper: Boden’s The Creative Mind (1990) and Mind as Machine (2006); Dennett on the intuition-pump charge; Bender and Koller (2020) for the LLM-era version.