* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download 2101INT – Principles of Intelligence Systems
Turing test wikipedia , lookup
John Searle wikipedia , lookup
Visual Turing Test wikipedia , lookup
Computer Go wikipedia , lookup
Intelligence explosion wikipedia , lookup
Existential risk from artificial general intelligence wikipedia , lookup
Embodied cognitive science wikipedia , lookup
Functionalism (philosophy of mind) wikipedia , lookup
Ethics of artificial intelligence wikipedia , lookup
History of artificial intelligence wikipedia , lookup
2101INT – Principles of Intelligent Systems Lecture 2 Last week we covered History of automatons, in literature and reality The birth of AI and the Dartmouth conference Some applications and domains of AI since then – – – – Planning Natural Language Processing Expert Systems Neural Networks Videos demonstrated AIs operating in “microworlds” Discussion about what we think constitutes intelligent behaviour 2-3 The Turing Test Proposed by Alan Turing in 1950 to be an operational definition of intelligence A human “interrogator” asks questions of two entities – one computer, one human – behind screens and using a console, to conceal which is which If the interrogator cannot determine which is which then the computer is said to be intelligent An extension of this is the total Turing test which includes a 1-way video feed and the provision for haptic interaction 952pp The Turing Test cont. Physical simulation of a human is not considered necessary for intelligence and is deliberately avoided by the screen/console/1-way link etc. The test is purely a behavioural test of intelligence, relying only on the external behaviour of the entity and not on its internal mental states. Many philosophers claim that passing the Turing Test does not prove that a machine is thinking, just that it can simulate thinking. Turing’s response to this objection is simply that in ordinary life, we never have any direct evidence about the internal mental states of other humans. Ch.26 The Schools of Thought At the highest level, AI can be divided into two schools of thought: – Weak AI “the principal value of the computer in the study of the mind is that it gives us a very powerful tool” – Strong AI “the computer is not merely a tool in the study of the mind; rather, the appropriately programmed computer really is a mind” [Searle, 1980] These relate to the question of consciousness of an intelligent system 949 Weak AI: Can Machines Act Intelligently? Weak AI argues that machines are not actually conscious and only appear to think Computers can do many things as well as or better than humans – including tasks considered to require insight and understanding But “Weak AI” doesn’t believe that computers actually use insight and understanding in performing these tasks 949 The mathematical objection Certain mathematical questions are unanswerable by particular formal systems Gödel’s Incompleteness Theorem (GIT) and the Halting Problem being two relevant examples Some philosophers – J. R. Lucas being one – argue that these demonstrate that machines are mentally inferior to humans, since machines are formal systems and can therefore not establish the truth of their own Gödel sentence 950 Arguments against Lucas 1. Computers are not Turing Machines, they are approximations only 2. There are statements that humans cannot assert the truth of 3. It is impossible to prove that humans are not subject to GIT because human talent is not a formal system 950 The argument from informality Originally raised by Turing, claims that human behaviour is far too complex to be captured by simple rules Since computers are only able to follow a set of rules, they cannot therefore generate behaviour as intelligent as humans What this argument is really arguing against is GOFAI – “Good Old Fashioned AI” - that all intelligent behaviour can be captured by a system that reasons logically with symbols Merely acting intelligently isn’t precluded by this 954 On the nature of human consciousness Descartes considered how the soul (let’s say consciousness) interacts with the physical body This has come to be termed the mind-body problem He concluded that the mind must be distinct from the body and be two different things – hence dualism Materialism is a monist theory, doesn’t believe that the mind and the body are different things, or that there are such things as immortal souls It is simply that brains cause minds 954 Relationship to AI Dualism per se disallows strong AI, since consciousness is not a consequence of the physical system Materialism on the other hand, does allow strong AI Explaining away the mind-body problem Descartes managed to contradict himself when he proposed interactionism. In this he said that mind and body do interact, but this is restricted to the pineal organ at the base of the brain Less contradictory is the theory of parallelism. According to this, mind and matter are entirely separate, each obeys its own laws but each keeps time perfectly with the other - due to the marvellous planning of a greater being. Explaining away the mind-body problem Similar to the last is the doctrine of occasionalism. So although mind and matter can’t interact, providence intervenes on each “occasion” when they need to. And finally, epiphenomenonalism. This holds that minds play no role in the everyday running of the universe. Consciousness is merely an incidental byproduct of the system. Matter “causes” mind as in materialism, but thought has no effect on matter. So we can watch the world go by, but can’t do anything about it – any impressions to the contrary are simply an illusion 952 Strong AI: Can Machines Actually Think? Strong AI argues that a properly programmed computer is conscious “mind” in the fullest sense Functionalism The influence on functionalism is attributed to influence of computers on modern society Functionalism is still Materialism, but: – – Brain states are distinct from Mental states Behaviour is not directly related to stimulus It does not matter what the physical cause of the mental (functional) state is These functional states are responsible for our outward behaviour 954 Fuctionalism cont. Any two systems with isomorphic causal processes will have the same mental states – even if they don’t have the same brain states Recall from last week – “The conjecture that every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it” McCarthy et al. (1955) “A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence” 954 Biological Naturalism The opposite view is termed biological naturalism It argues that mental states are emergent features, caused by low-level neurological processes inside the neurons It is the unspecified properties of the neurons that matter – and equivalent mental states are not produced by something else with the same functional behaviour Of course, this theory doesn’t define why neurons have this power but the concept of a soul comes to mind 954 The Chinese Room Argument – “No one supposes that a computer simulation of a storm will leave us all wet… Why one earth would anyone … suppose a computer simulation of mental processes actually had mental processes” Searle (1980) “Minds, Brains, and Programs” Searle’s conclusion is that running an appropriate program is not a sufficient condition for being a mind 958 The Chinese Room Argument cont. Consider that you have a system consisting of a human who only understands English He is placed inside a box with a rule book and some blank pieces of paper The box has an opening, through which appear slips of paper with indecipherable symbols written on them The instructions in the rulebook direct the human to write slips of the blank pieces of paper, look at bits of paper previously used and even write some of the indecipherable symbols on paper and pass them back 958 The Analogy of the Chinese room The human is playing the role of the CPU The rule book is his program The pieces of paper are his memory 958 The Chinese Room Argument cont. What does this system look like from the outside? The indecipherable symbols are really questions and with the help of the rulebook the symbols written out are the appropriate answers So it appears that the system can “understand” Chinese, giving as it does answers appropriate to the questions asked It therefore passes the Turing test 958 The Chinese Room Argument cont. Does the person in the room understand Chinese? – No. Do the rule book and the stacks of paper understand Chinese? – No. So if none of the components of the system understand Chinese, how can the system understand Chinese? Therefore, running the right program does not necessarily generate understanding. 958 The Systems Reply – "While it is true that the individual person who is locked in the room does not understand the story, the fact is that he is merely part of a whole system, and the system does understand the story. The person has a large ledger in front of him in which are written the rules, he has a lot of scratch paper and pencils for doing calculations, he has 'data banks' of sets of Chinese symbols. Now, understanding is not being ascribed to the mere individual; rather it is being ascribed to this whole system of which he is a part.“ Searle citing ‘Berkeley’ (1980) “Minds, Brains, and Programs” 958 The Systems Reply - Response Although none of the components understand Chinese, the system as a whole does. It could be said to be an emergent property of the system. After all, if you ask it “Do you understand Chinese?” it will of course response (in Chinese) that it does Searle’s counter-argument is that if the human were to memorise the rule book and not use pieces of paper he still wouldn’t understand Chinese. 958 The Robot Reply – "Suppose we wrote a different kind of program. Suppose we put a computer inside a robot, and this computer would not just take in formal symbols as input and give out formal symbols as output, but rather would actually operate the robot in such a way that the robot does something very much like perceiving, walking, moving about, hammering nails, eating drinking -anything you like. The robot would, for example have a television camera attached to it that enabled it to 'see,' it would have arms and legs that enabled it to 'act,' and all of this would be controlled by its computer 'brain.' Such a robot would have genuine understanding and other mental states." Searle citing ‘Yale’ (1980) “Minds, Brains, and Programs” 958 The Robot Reply - Response Searle notes that this argument concedes that intelligence is more than symbol manipulation and must relate to the physical world But the addition of perceptual and motor capabilities do not add understanding Suppose you leave the human inside, but now with additional symbols (still in Chinese) that come from the camera. Some additional symbols (also in Chinese) will be given to the motors. The human still knows nothing about what any of these mean, and still does not understand Chinese. 958 The Brain Simulator Reply – "Suppose we design a program that doesn't represent information that we have about the world, but simulates the actual sequence of neuron firings at the synapses of the brain of a native Chinese speaker when he understands stories in Chinese and gives answers to them. The machine takes in Chinese stories and questions about them as input, it simulates the formal structure of actual Chinese brains in processing these stories, and it gives out Chinese answers as outputs. We can even imagine that the machine operates, not with a single serial program, but with a whole set of programs operating in parallel, in the manner that actual human brains presumably operate when they process natural language. Now surely in such a case we would have to say that the machine understood the stories; and if we refuse to say that, wouldn't we also have to deny that native Chinese speakers understood the stories? At the level of the synapses, what would or could be different about the program of the computer and the program of the Chinese brain?" Searle citing ‘Berkeley and MIT’ (1980) “Minds, Brains, and Programs” 958 The Brain Simulator Reply - Response Searle notes (once again) that intelligence must therefore be more than symbol manipulation. Even getting this close to the operation of the brain is not sufficient to produce understanding Imagine that the neurons are simulated by water pipes and that a little man runs around turning valves on and off according to rules in the book. After all the right “neural firings” the Chinese answer pops out Still the man doesn’t understand, the water pipes don’t understand and we would return to the Systems Reply 958 Summary of the Chinese Room Strong AI claims that instantiating a formal program with the right input is a sufficient condition of intelligence/understanding/intentionality Attributing these things to the Chinese Room are based on the assumption that if it looks and behaves sufficiently like something else with the same function, then it must have corresponding mental states If we knew what to attribute its behaviour to (little man, pipes, formal program) we would not make this assumption and would not attribute intentionality to it. References Haugeland, John (1985) “Artificial Intelligence – The Very Idea”, MIT Press. “Philosophy of Mind – Functionalism” http://www.philosophyonline.co.uk/pom/pom_functionalism_introduction.htm Searle, (1980) “Minds, Brains, and Programs” http://www.bbsonline.org/documents/a/00/00/04/84/bbs00000484-00/bbs.searle2.html