the dungeon has...

The Poplog Version of Eliza

As a child I remember going round to a friend’s house and playing on their computer. At home we only had a Spectrum, which had failed to catch my attention, but my friends had a PC. This meant little to me at the time – apart from the fact that they had a specific computer program called Eliza. Eliza was a program originally created in the 1960’s by Joseph Weizenbaum, and this marked a change in the way that computers were viewed. I did not know that at the time, I was more concerned with how it worked. At first, as a child, it seemed convincing enough, as a psychiatrist-doctor program which ran on the computer. My grasp of the English language was limited, as was my understanding of computers. It seemed to make perfect sense, until we became bored and began to test the programs limitations in a way that only children can. I have fond memories of spending many Sunday mornings trying to upset it, or offend it, or being just plain abusive to it. We quickly became bored when we realised that it did not understand, that we could get no reaction out of it that we had not already received. In this brief essay I shall attempt to look at one version of this program and how it can be related to the much larger topic of computer intelligence.

I ‘talked’ to Eliza (the poplog version) in order to determine how well the program understands the English language. In the appendix are some examples of the conversation. Eliza does go far towards emulating human conversation, however this has it’s limits as can be seen from the examples. The program works to pick up on specific words and to reply with a set phrase, or to reverse any statement made and respond as a question. This works well until it becomes faced with some of the more intricate parts of grammar, or until it becomes faced with object-orientated conversation where it can begin to suggest the impossible. For example, in response to the statement ‘I do not have a sister’ it responded with the statement ‘suppose you did have a sister’, in order to promote further discussion. However, it appears to lack an understanding of the world, as shown by the statement that followed ‘perhaps you really have a sister’. The program also shows the ability to retain parts of the conversation, and to feed it back later e.g. ‘Earlier you said sometimes you are’ and ‘I recall your saying you do not have a problem’. This initially seems quite impressive, as it can promote the notion that it was understanding and remembering the conversation, however this is also limited to specific phrases which are retained in its memory.

The programming in Eliza is impressive in some ways, it can take ‘I’ and convert to ‘you’ in an output sentence, it can also pick up on key words, which in the right circumstances may actually produce a very realistic conversation. Although Eliza does have limitations in its use of English it has, at times, been mistaken for an intelligence. This really is the fundamental question, how far can we go to say that Eliza is an intelligent program. To answer that we must first consider exactly what we mean by the term ‘intelligence’ and look at what other people have said on the subject.

The philosopher Bertrand Russell said: ‘mental events are events in a living brain’. Which, taken as said, does not allow for the possibility of mental events happening in anything that is not a biological brain. Therefore if we take intelligence to be a mental act it would never be possible for a machine to be considered intelligent. Chalmers raised this question in terms of consciousness in computers, does consciousness only apply to human beings as a unique property of the universe? An alternative viewpoint was given by Gilbert Ryle, who attempted to dispel the myth of the ‘Ghost in the Machine’. He considered that we could not rely on any conception of a soul>

Alan Turing, in the philosophy journal ‘Mind’, wrote a paper called “Computing Machinery and Intelligence” in which he considered the logical possibility of machine intelligence. In this he tried to create a criteria for ascertaining intelligence, which is now known as the ‘Turing test’. Briefly stated he considered that if a machine behaved indistinguishably from a human then it must be intelligent. In practical terms, for example, if you are having a conversation with a machine and cannot tell the difference between it or a human then it must surely have intelligence. This argument stems from a Behaviourist viewpoint, because it gives an operational definition of intelligence through behaviour. However, the question remains whether intelligence can be tested entirely by behaviour.

The philosopher Thomas Nagel said in book ‘The View From Nowhere’ that “…it is hopeless to try to analyse mental phenomena so that they are revealed as part of the ‘external’ world.” This plainly goes against behaviourism and begins to question how much we can tell from behaviour. Perhaps the best example was given in 1980, by John Searle, when he published his well known ‘Chinese Room’ argument. In the argument a man who does not speak any Chinese is placed in a room. In the room there are filing systems and a dictionary. Written questions in Chinese are passed to him and using the information he is given on how to interpret them he constructs written answers. However this man still does not understand Chinese even though he is responding in it correctly. He could be considered to have inputs and outputs but understand nothing. This Searle suggests is entirely comparable to a machine, that when we attribute understanding to a machine all we can be is metaphorical. The machine is limited to its understanding of how to manipulate the ‘symbols’ of language, it has no concept of the outside world. Its ability is entirely limited to the ability of its programmer. As was once said to me ‘you cannot get semantics from syntax’. This is applicable in many systems, most noticeably computer spell-checkers. Computer spell-checkers can recognise misspellings and suggest alternatives, nut because they have no understanding of the words some words are not picked up on. Stephen Pinker, in his book ‘How the Mind Works’ said: “…to the extent that the world obeys mathematical equations that can be solved step by step, a machine can be built that simulates the world and makes predictions about it”. However, the operation word here is ‘simulates’ as there appear to be limitations on how much a computer can be said to understand what it does.

This said, we are in an age where computers are becoming more and more complex every day. In 1996, the computer Deep Blue defeated World Chess Champion Garry Kasparov in one game, drew in two others and lost in one. If computers are a match for our mental ability can we honestly say that they are not intelligent? As previously stated we need a working definition of intelligence. Personally I consider intelligence to be the ability to calculate, to remember and to learn. However questions can be raised of whether intelligence is understanding or knowing, which computers at this time are not fully capable of. So, using my definition, some computers can be considered intelligent as they have shown the capability to do all that was mentioned in my first statement. However the question could be raised of whether they need to be ‘conscious’ to fully understand what it is that they do, to have any sort of intentionality.

But is Eliza intelligent? To some it may pass the Turing test but that is a very subjective test, and personally it does not have enough working knowledge of the English language to be considered or mistaken for an intelligence. The programs obvious lack of understanding can be seen, otherwise it would not produce sentences such as ‘suppose you did have a specific problem’ followed by the question ‘would you really have a specific problem’. Unfortunately, the program is not complex enough to pass any test, and when taken from the view of Searle’s Chinese room falls down completely, as it proves that it has no conception of what the English ‘symbols’ that it is manipulating mean. Therefore I cannot help but say that this specific program fails to be called ‘intelligent’ by my working definition, and although one day it may be complex enough to pass the Turing test to everyone’s satisfaction, we may still not be able to say that it is intelligent.

- - - - - - - - - - -

Appendix

1. I RECALL YOUR SAYING YOU DO NOT HAVE A PROBLEM
2. ? i do not have a specific problem
3. SUPPOSE YOU DID HAVE A SPECIFIC PROBLEM
4. ? then i would have a specific problem
5. WOULD YOU REALLY HAVE A SPECIFIC PROBLEM
6. ? if i supposed i did then would yes
7. ? i do not have a sister
8. SUPPOSE YOU DID HAVE A SISTER
9. ? then i would be very surprised
10. WOULD YOU REALLY BE VERY SURPRISED
11. ? yes, as i do not have a sister
12. PERHAPS YOU REALLY HAVE A SISTER
13. ? no, i don't have a sister
14. YOU DO NOT HAVE A SISTER
15. ? that is what i have been saying
16. EARLIER YOU SAID SOMETIMES YOU ARE
17. ? sometimes i am what
18. WHAT IF I WERE WHAT
19. ? what if you were what what

- - - - - - - - -

Bibliography

1. New Scientist – Will Machines ever think? – Harry Collins
2. The Philosophers’ Magazine – Minds and Brains – Ted Honderich
3. The Reith Lectures – John Searle
4. How The Mind Works – Stephen Pinker
5. Theories of the Mind – Stephen Priest
6. Philosophy The Basics – Nigel Warburton
7. The Computer and the Mind – Philip Johnson-Laird
8. The View from Nowhere – Thomas Nagel
9. The Concept of Mind – Gilbert Ryle
10. Oxford Companion to The Mind – ed. Richard l. Gregory

TrackBack URL for this entry: http://bluejoh.com/cgi-bin/mt/mt-tb.cgi/390