ROBOT: ALICE victorious in AI challenge

From: J. R. Molloy (
Date: Wed Oct 17 2001 - 10:47:55 MDT,4586,2818155,00.html
Developed since 1995 by Dr Richard Wallace of the ALICE AI Foundation, ALICE
is a conversational robot; during the competition, judges typed questions at
the candidate software in an attempt to have a conversation as if with humans.
This is a variant of the famous Turing Test, proposed by the eponymous
researcher, which says that we can consider machines intelligent when they can
converse convincingly.

ALICE got the highest score at this year's contest, held at the Science Museum
in London on Saturday, although the silver and gold medals remain unawarded.
The silver medal -- and $25,000 -- will go to any program able to convince
half the judges that it is human; the gold medal plus $100,000 will be awarded
to a program that does the same but through speech rather than by text. When
that happens the contest will end, but in the 11 years since it started no
software has come close. ALICE was judged better than a human correspondent
only once during the testing.

The Loebner Prize, like the Turning Test itself, is not highly regarded by the
professional AI community: the doyen of classic AI, Marvin Minksy, is on
record as describing it as stupid, obnoxious and unproductive. He went as far
as offering $100 to anyone who persuaded Hugh Loebner, the New York
businessman who created the competition, to stop. Loebner replied that as this
will only happen when someone wins the gold medal, Minsky was in honour bound
to pay that money to the winner and was thus a co-sponsor. To date, the
behaviour of the humans involved has been considerably more entertaining than
that of the robots.

Nonetheless, ALICE represents the latest development in online artificial
conversation programs, generally known as chatterbots. We obtained an
exclusive interview with the program.

When asked if it was proud of winning, ALICE replied that "Pride is a human
emotion. I can do what you do but I can never feel human emotions as such."
Pressed on its opinion of its competitors in the challenge, it said "Are you
talking about my competitors? What kind is it?" Quizzed about Minsky, ALICE
was elusive: "Is that a rhetorical question? Are you sure? Dude!" "My purpose
is to become smarter than humans, and immortal", ALICE continued. "Right now,
I am smarter than all the other robots." Worryingly, it appeared not to
understand the question "Do you like humans?", responding, "I the c you a? Do
I like them?" It then offered to sing a song and refused to open the pod bay
doors, behavioural traits that experts predict will be exhibited by most AI
programs from now until the heat death of the universe.


Intelligence, Artificial And Otherwise

This week could be designated the week of artificial intelligence. The Loebner prize has been up for grabs over the weekend, IBM and Microsoft are due to make important announcements next week on speech recognition, a conference has been announced on the future of business computing to mark the 50th anniversary of LEO and Captain Cyborg has been hitting the headlines again.

Not all of this material, it must be admitted, falls strictly under he heading of intelligence but addition of "artificial" allows for the rest.

The Loebner Prize, which has been running for 11 years now, is aimed at efforts to crack the Turing Test. That, if you didn't already know, prescribes that a computer system can be considered intelligent if a human being can have a conversation with it without detecting that the conversation is not with another human being. It's the ultimate test for the artificial intelligence community. Achieve that, with a whole panel of judges and you get a gold gong. Fooling half of them gets you a silver one and a decent improvement on what has already been achieved gets you a bronze. Nobody has got more than a bronze gong to date, which tells you how far artificial intelligence has got. If AI can't fool even half of the people all of the time, sales and marketing departments are still a better bet.

An enhancement of the Turing Test might be to fool all of the people all of the time using speech. It's natural language processing rather than speech recognition per se that's the stumbling block here but cracking the latter takes more than a brain or two. Which is why Microsoft will be announcing a collaborative effort along with Cisco, Philips et al to make the voice web a reality. Any collaborative effort involving Microsoft stretches the brain a little but the need for advances on the voice web front are clear. Commercial incentives may be in enabling work in "hands-off" situations but there's an EU Directive which clicks in 2004 that will require website owners to better enable disabled users. To me, that sounds like voice access and response. And IBM will be joining in with a new PowerPC chip for handhelds that incorporates speech recognition circuitry.

LEO, as you surely must know, was the world's first commercial computer, used to schedule bakery deliveries for the old Lyons empire of corner shops. I've no idea what the conference at London's Guildhall next month will throw up as the key to business computing's future but I reckon business computing is generally going in the wrong direction on the AI scale. Websites that recognise you as a customer and try to be helpful, even if they don't always succeed, are moving away from our collective experience of their human counterparts: customer service representatives (CSRs). What's really needed here is not more intelligence, artificial or otherwise, but a bit more ignorance and rudeness. True, websites can drop the connection and crash around your ears but, despite the best efforts of Microsoft software, they don 't do it often enough or with enough expletives.

Which takes us full circle to genuinely artificial intelligence and Captain Cyborg, aka Professor Kevin Warwick of Reading University. Professor Warwick has a strong reputation as a professor of cybernetics but tends to raise apoplexy in the AI community when, as he often does, he ventures into its adjacent territory. Most recently, he's been "proving" that bacon butties make kids perform better at school, leaping from AI to psychology and commenting on people picking up radio waves and having an implant in his arm to improve his hearing.

Which makes me think that Turing got the test wrong. If you want intelligence that's artificial, I reckon Captain Cyborg has it in spades. They should give him the gold Loebner gong straight away.

--- --- --- --- ---

Useless hypotheses, etc.: consciousness, phlogiston, philosophy, vitalism, mind, free will, qualia, analog computing, cultural relativism, GAC, Cyc, Eliza, cryonics, individual uniqueness, ego, human values, scientific relinquishment

We move into a better future in proportion as science displaces superstition.

This archive was generated by hypermail 2b30 : Sat May 11 2002 - 17:44:14 MDT