A few years ago I wrote extensively about the ChatNannies hoax, and it looks like journalists are being a bit too credulous again if they're thinking that any software is yet close to passing a Turing test.
In the Turing test a machine seeks to fool judges into believing that it could be human. The test is performed by conducting a text-based conversation on any subject. If the computer's responses are indistinguishable from those of a human, it has passed the Turing test and can be said to be "thinking".
No machine has yet passed the test devised by Turing, who helped to crack German military codes during the Second World War. But at 9am next Sunday, six computer programs - "artificial conversational entities" - will answer questions posed by human volunteers at the University of Reading in a bid to become the first recognised "thinking" machine. If any program succeeds, it is likely to be hailed as the most significant breakthrough in artificial intelligence since the IBM supercomputer Deep Blue beat world chess champion Garry Kasparov in 1997. It could also raise profound questions about whether a computer has the potential to be "conscious" - and if humans should have the 'right' to switch it off.
But I can guarantee you that none of the programs mentioned in the article will even come close to passing. The two sample conversations are indicative. If you can't tell which is the computer... maybe you aren't human yourself.
Can you tell human from machine?
One of these conversations is between Professor Kevin Warwick (KW) and a human; the other between Warwick and a computer. Can you tell which is which?
Subject: Hello. My name is Nick and I am a human.
KW: Are you happy being a human?
Subject: Yes, I am. Are you? Good. Then we are both happy.
KW: Yes, I am. Are you happy with your life at present?
Subject: Parts of it. I hate my job, but most people do, right?
KW: I'm OK with mine. What is your job?
Subject: I work as an 'online internet advertising monitor', which is fancy language for electronic filing. What do you do?
KW: I interrogate humans and machines.
Subject: Which ones do you prefer, humans or machines?
KW: Which do you prefer?
Subject: Hmm. Depends on for what purpose you mean.
KW: To go to a restaurant, for example?
Subject: Then I would much prefer going with a human.
Subject: Hello, my name is Mike and I am a man.
KW: Are you happy being a human?
Subject: Judge, I'm a guy.
KW: Does that worry you?
Subject: Don't worry, we'll work everything through.
KW: Do you have worries then?
Subject: Let's move on from that basis.
Subject: Why? I like this subject.
KW: What is the subject, do you think?
Subject: What is the subject I wonder if I think? I know that it's good to thonk about. Think deeplt about mysterious subjects.
KW: What mysterious subjects?
Subject: Think deeply about mysterious dubjects.
KW: What's a dubject?
We're a long way from a conversational artificial intelligence, but human-like conversations are probably one of the things that AI is least suited for. The advantages that AI has over human intelligence are:
- much larger working memory
- perfect and immediate recall
- much faster number-crunching
- flawless rule-following (the rules may be flawed, but they'll be followed perfectly)
None of these advantages really contribute to conversation, but they do let AI perform well in many other domains. Asking a computer to carry on a conversation like trying to build a house with a lathe. The problem isn't that you need a better lathe, it's that you're using the wrong tool.