Contemporary philippine arts from the regions_PPT_Module_12 [Autosaved] (1).pptx
Opening the door to the Chinese Room: Intersubjectivity and the Phenomenology of AI
1. Opening the door to the Chinese Room:
Intersubjectivity and the
Phenomenology of AI
Open AIED
19 September 2018
DR. ROBERT FARROW
INSTITUTE OF EDUCATIONAL TECHNOLOGY
@philosopher1978
rob.farrow@open.ac.uk
3. 3
PERSPECTIVES FROM PHILOSOPHY OF MIND
ARTIFICIAL INTELLIGENCE
1. The first thing to say is that many computer scientists and engineers are quite uninterested in the
philosophy of artificial intelligence (despite this being where AI comes from as a concept)
2. Philosophers have long been interested in whether a non-human is able to possess a mind or have
the attribute of intelligence
3. This has only been from the perspective of computational theories of mind over the last hundred
years or so, but can be traced back into the history of philosophy as the mind-body problem
4. The so-called ‘hard problem of consciousness’ is a contemporary example of the mind-body problem
5. Machine functionalism presented the mind as an information processor
6. Turing test was presented as a test of intelligence
7. Searle’s response to machine functionalism: the Chinese Room
8. Thoughts on the potential use of AI in education
9. Concluding thoughts
4. 4
WHY DO HUMAN BEINGS EXPERIENCE PHENOMENA AT ALL?
HARD PROBLEM OF CONCIOUSNESS (CHALMERS, 1995)
Information flowQualia
6. 6
HOW CAN PHYSICALISM BE RECONCILED WITH MENTAL PHENOMENA?
THE MIND-BODY PROBLEM
The mental and physical
are separate (dualism)
The mental and physical
interact (mental
causation)
Physics is causally
closed (physicalism)
7. 7
THE MIND AS INFORMATION PROCESSOR
MACHINE FUNCTIONALISM
1. Puttnam (1967) argues in ‘Psychological Predicates’ that psychological states are not in correspondence with physical
states
2. This gives rise to an interest the multiple realisability of mental properties, and hence to functional accounts of the
mental
3. For instance, pain reception can be realised in multiple ways in different organisms, but all of these systems have the
same ‘function’
4. For the functionalist, mental states are reducible to functional states: there are no ‘inner’ mental representations and
no mental/physical dualism
5. Puttnam frames his discussion of functionalism in terms of ‘Turing machines’ which read & write data along a tape
according to a finite set of internal conditions and a consistent symbolic alphabet
8. 8
A TEST OF INTELLIGENCE?
TURING TEST
Turing’s (1950) famous test of ’intelligence’
Turing’s thesis: if two systems are input-output equivalent they have the same psychological status (Kim, 1998)
9. 9
THE CHINESE ROOM THOUGHT EXPERIMENT
JOHN SEARLE (1980)
Searle similarly rejects the mind/body problem as a false dichotomy (i.e. he rejects the premise that mind and body are
separate in favour of the idea that they are different aspects of a single phenomena).
The ‘Chinese Room’ thought experiment is often presented as an argument against ‘Strong AI’ (Dennett)
Strong AI thesis: “The appropriately programmed computer with the right inputs and outputs would thereby have a mind in
exactly the same sense human beings have minds.”
More recently, Searle has framed this thought experiment in terms of the hard problem of consciousness
10. 10
AI IN EDUCATION
INTERSUBJECTIVITY?
1. At The Open University AI is being considered for personalised learner support; it would draw on the
massive amount of data retained by the university about their students
2. For instance, AI could offer personalised feedback on writing and other work
3. AI could effectively replace a large range of functions in education
4. In May 2018 the OpenAIED group heard a very interesting presentation from Mark Nichols (LTI) on
some of the thinking around AI support for students.
5. This drew on Buber’s (1923) account of intersubjectivity which contrasts intersubjective relationships
(“I-Thou”) with the world of things (I-It)
6. The suggestion was that AI interactions should be characterised by intersubjectivity and a sense of
‘human’ interaction
7. It’s important to note that intersubjectivity assumes a kind of symmetry between human relationships
which expresses itself as forms of mutual recognition – the kind of recognition which machines
cannot genuinely reciprocate even if they can simulate it convincingly
11. 11
EVALUATION & MONITORING
BRINGING LEARNING TO LIFE
https://www.theguardian.com/technology/2018/may/11/google-
duplex-ai-identify-itself-as-robot-during-calls
12. 12
SHOULD AI BE RATIONAL, OR BE LIKE HUMANS?
ARTIFICIAL INTELLIGENCE: CONTRASTING VISIONS
Russell and Norvig (1995, 2002, 2009)
13. 13
SOME THOUGHTS
IN CONCLUSION
1. Despite the recent interest in AI from an engineering and computer science perspective there remain
unresolved epistemological and metaphysical controversies
2. I hope to have at least indicated that the philosophical issues around AI do have important
implications for those building ‘practical’ systems
3. In order to resolve the mind-body problem we necessarily end up taking a reductive approach: in the
behaviourist/functionalist tradition mental operations are disregarded in favour of language and
behaviour
4. These are structural issues: machine learning has made little progress with representing higher order
thoughts, higher levels of abstraction, being creative with language, or ‘common sense’ (Russell &
Norvig, 2009)
5. In summary, we might suggest that there is massive hyperbole about the affordances and likely
futures of AI: reports of artificial intelligence are greatly exaggerated and it is human intelligence that
is being extended
6. All of the AI we refer to is essentially ‘weak’ (i.e. task orientated rather than generalised) intelligence
and should perhaps be referred to as ‘augmented’
7. This is to say nothing of the well documented ethical concerns raised by the use of AI in education
AI regularly hits the headlines these days – but has the quality of AI improved?
Machines may be intelligent, but few claim they are conscious
If information flow is all that matters from the perspective of intelligence, then what's the point of conscious experiences?
Is our phenomenal experience constitutive of human intelligence? If the answer is affirmative then what is the role of the phenomenal experiences in artificial intelligence when AI’s don’t have them (as far as we can tell)
This is a recent reformulation of the mind/body problem:
It should be noted that some philosophers in the continental tradition reject the mind/body distinction as it is customarily formulated
This is a recent reformulation of the mind/body problem: one of the three premises must be rejected to remove contradiction
It should be noted that some philosophers in the continental tradition reject the mind/body distinction as it is customarily formulated
Prior to MR it was thought that mental states and physical states were correlated with each other: if my brain state is that same as yours then we have the same mental state
The test involves interacting with an entity through a kind of chat text – if the subject cannot distinguish the responses of human and machine then the machine is considered ‘intelligent’
The test is heavily skewed towards the affordances of a machine and the bar for successful deception (confusion) is low
Is rationality or behaving like a human the more reliable indicator of intelligence?
Which is the more desirable in an AI?