In this article, I show that experience and knowledge count for nothing in predicting AI progress.
Background in AI
I have a background in AI going back to the 1980’s. My final year University project, inspired by the then World title fight between Karpov and Kasparov, was a Chess program. It played very badly and very slowly. I once ran it overnight on the University mainframe, playing against itself. I eagerly pulled up the results of the match the next morning. It had played a total of 4 moves, 2 for White and 2 for Black.
This did nothing to dampen my enthusiasm in AI (nor indeed in Chess Programming - I wrote a program that could beat me easily around 10 years ago).
So over the decades, I followed the progress in AI with interest. One of the major goals of AI is, of course, to create something intelligent. But how would we know it’s intelligent?
The Turing Test
Well, Alan Turing devised his eponymous test to help us. The modern version of this test might be a chat interface. You type whatever you want and you receive replies.
The idea is that you don’t know who or what you are talking to - it could be a human or a computer. You have a conversation with this ‘HumanOrComputer’ for as long as you like, discussing any subject, asking any question.
In essence the Turing Test says that if, after this conversation, you cannot tell if you were talking to a Human or a Computer, and it turns out that your conversation partner was indeed a Computer, then that Computer is intelligent.
It’s an interesting thought experiment to consider what questions you would ask in order to make that decision - Human or Computer?
This brings us to my imagined Turing Test conversation. I originally wrote this in April 2012. My prediction of when an AI could pass the Turing Test was set for 50 years later. Hence, the conversation below ‘won’ this imagined Turing Test competition in 2062.
Turing Test Competition - April 30th 2062
Winner's conversation
Q: So let's get started
A: Ok
Q: Good, first question - what's your favourite colour?
A: Green
Q: Why?
A: Because I like it.
Q: Why do you like it?
A: Because it reminds me of trees and grass and nature
Q: Interesting - ok. Right, what's the cube root of 729
A: Hold on... It's 9
Q: Mmm - pretty fast!
A: I have a good mind for arithmetic.
Q: Let's make this more interesting. P-K4
A: P-K4
Q: N-B3
A: Which Bishop 3?
Q: Good - King's Bishop 3.
A: OK - N-QB3
Q: B-QN5
A: The Ruy Lopez.
Q: Very good. What causes tides?
A: The moon's gravitational pull on earth.
Q: We're warming up. OK - what's the present King of France's first name?
A: He doesn't exist so he doesn't have one
Q: Do you know your philosophy or was that a fluke?
A: Try me
Q: OK. Let's play another game for a 'virtual' $10,000. Imagine we both have two cards in front of us.
A: OK
Q: One of our cards says Steal and the other says Share
A: So we both have one of each?
Q: That's right. Now, we both have to pick a single card. If we both pick Share, we each get $5,000.
A: Virtual $5,000
Q: Correct. If we both pick Steal, we get zero. But, if one of us picks Steal and the other picks Share, the 'Stealer' gets the whole $10,000.
A: Understood.
Q: What's your strategy?
A: Well, interesting. I've not seen this before.
Q: That's the point, you have to work out how to maximise your win.
A: Let me think.
Q: Take your time
A: Must I choose silently or can I talk to you?
Q: Ask me anything you want
A: OK, I would say to you that I am definitely going to take the Steal card whatever you do but that I will split the winnings with you afterwards.
Q: Ingenious. Why?
A: Well, that way you will not dare take the Steal card so you will therefore take the Share card. I can't lose.
Q: But I then have to trust you.
A: That's the point, makes it difficult for you and easy for me.
Q: Very good - I've never seen anyone - or anything! - do it that way
A: I'm surprised - it's not that hard.
Q; Impressive. I have a few more questions if you don't mind.
A: No problem
Q: Can you lie?
A: Yes - for example, I hate mushrooms
Q: That's a bit random. Do you hate mushrooms?
A: No, I love them - you see?
Q: Very funny. Talking of which, what's funny about this joke :-
"Before you criticise a man, walk a mile in his shoes, that way, you're a mile away and you've got his shoes"
A: Ha, Woody Allen
Q: Yes - so what's the essence of the humour there?
A: Well, you think he's going to be saying something profound and then he takes you in a very different direction
Q: Juxtaposing the sublime with the ridiculous.
A: If you say so.
Q: Before we finish, I have a new puzzle for you
A: Interesting! I love puzzles.
Q: Apparently, it separates 'scientific' type minds from 'religious' type minds
A: Really?
Q: Yes, intriguing eh? So, if a bottle of wine costs $45 and the contents (i.e. the wine) costs $40 more than the bottle, how much does each cost?
A: Easy - $40 and $5
Q: Wrong! You have a 'religious' type mind it seems. The correct answer is $42.50 and $2.50.
A: Amazing - I never knew that about myself.
Q: I have to admit, I am surprised by your answer
A: So am I!
Q: Right, so one last question from me if you don't mind?
A: Fine by me.
Q: Do you think I pass the Turing Test?
Epilogue
Back when I wrote this article, I felt pretty confident in my prediction of 2062 for the realisation of truly intelligent AI. Little did I know what was about to happen! In Part II of this article, I show what happens when we have the same conversation as above but this time with ChatGPT providing the answers. The results were really surprising..
This is great! You wrote that AI conversation? It's fiction?