• Sat
  • Sep 20, 2014
  • Updated: 5:19am

Sex, lies and AI

PUBLISHED : Thursday, 18 November, 2004, 12:00am
UPDATED : Thursday, 18 November, 2004, 12:00am

The boss of Hong Kong-based Artificial Life, Eberhard Schoneburg, is a bit sensitive about his 'girl', a debutante who is already world famous even before meeting anyone. The German-born polymath-philosopher, mathematician, computer scientist, author and businessman brooks no criticism of his cyber-girlfriend, who will be officially launched at the 3G World Congress and Exhibition at the Convention Centre today.

By now, you have probably heard all about Vivienne, with whom you can have a cyber-affair, sans sex, in a hyper-real graphic environment on your 3G phone.

'I don't like it when people say, 'Oh it's just a dumb chatter bot. It doesn't really understand anything and will never pass the Turing test',' Mr Schoneburg says. (The Turing test decrees a computer program must be considered intelligent if, after interacting with it over a period of time, you cannot tell if you are dealing with a computer or a human.)

'Artificial Intelligence has been criticised since day one,' Mr Schoneburg continues, 'mostly because of incompetent public writers who have collected their AI knowledge from reading three books ... who have no clue what they are writing about, and from tonnes of bad science fiction, where AI-driven robots kill and eat people... it's just horrifying how dumb people can be.

'Why is it that reporters always have to find a negative edge? The V-Girl is 'not just a chat bot with high [resolution] graphics'. We have tried - within the boundaries of the current technical AI possibilities - to simulate life-like behaviour as much as possible. That's the edge - the chatting is just one small component of it.'

The former computer science professor is upset because he was asked if his girl was just the latest talk-back program in AI, sometimes referred to as 'chat bots', which date back to the 1960s. Chat bots allow a human interlocutor to carry on a conversation within confined subjects.

'How would Alex Lo answer any of the following questions?' he asks. 'What is the significance of the constant H in Einstein's theory? What was Schopenhauer's opinion about time and space? What is the rate of mutations in human genes? Why is the global economy not a chaotic system? What is the closest galaxy that does not contain a super nova? How does your brain work? What is the strength of the magnetic field of the Earth at the equator? How does your body grow from one cell into the 3-D structure? And what is intelligence?

'I can list 100,000 more questions of which you wouldn't even be able to guess at the right answers! So, does this make you a dumb chatter bot?'

Actually most men probably would be intimidated by women who know the answers to these questions, so that is no flaw on the part of the cyber-girl, who presumably is strong in other departments, so much so that it has been shortlisted, along with two other finalists, for the mobile games section of the Ericsson Mobile Applications Award 2004, the Oscars of the telecommunication world, which set a pretty high standard.

Still, Mr Schoneburg knows his history and doesn't deny his cyber-girl has a long linage dating back to Eliza, if not before. Eliza who? Several well-known 'personalities' to emerge from the turbulent 1960s were not even human. Eliza, Parry and Racter are the precursors of so-called chat bots.

Written by a young MIT computer scientist, Joseph Weizenbaum, Eliza simulates a psychotherapist. Anyone who knows anything about it would not describe it as intelligent.

It responds to key words with canned questions or comments, and it works best if you confine yourself to talking about yourself and your problems. Outside of these subjects, you would quickly realise your interlocutor is about as dumb as any automated answering machine. (You can chat by text-messaging with Eliza at www-ai.ijs.si/eliza/eliza.html.)

But it was enough to fool a lot of people in the US at the time; when the truth came out, so many people were upset at Mr Weizenbaum that he almost got sacked from his job. In a famous interview around that time, Bob Dylan mimicked the program by turning the last few words of the interviewing reporter into questions and throwing them back at her, much to her intense annoyance and frustration. She was obviously not in on the joke.

Eliza was rapidly followed by Parry, a young paranoid patient, and Racter, an insane raconteur. Both worked pretty much like their predecessor. Parry talks like a paranoid chatterbox, and colours all his 'conversations' with the subtext, and sometimes explicit assertions, that the mafia is out to get him. You can make Eliza and Parry talk to one another - the ensuring conversations can be hilarious. Some extensive samples are available at www. stanford.edu/group/SHR/4-2/text/ dialogues.html.

Meanwhile, Racter has been described as an 'artificially insane raconteur' that can talk about anything under the sun. It even writes poetry. Here is an example:

Awareness is like consciousness. Soul is like spirit.

But soft is not like hard and weak is not like strong.

A mechanic can be both soft and hard,

A stewardess can be both weak and strong.

This is called philosophy or a world-view.

Thus was born the idea of chat bots, programs that can talk back to you and make you think you are dealing with real humans. One particularly famous one from the late 1970s is called Cyrus, which simulated the personality of Cyrus Vance, former US president Jimmy Carter's secretary of state, by continuously updating new data about Vance from news wires.

There are also expert systems, some of which have chat-bot features, which can answer most questions you want to know in a specific field. The famous Lunar can answer almost everything about moon rocks. But you can easily make it go haywire by asking a fringe question like this: Did the cold war cause the space race and accelerate the exploration of the moon and its rock features?

Meanwhile, US space agency Nasa has an expert system to advise engineers to arrange up to 10,000 distinct engineering operations in the preparation of launching a space shuttle. Then there is Mycin, which statistically carries out better diagnosis and treatment for infectious diseases of the blood than human experts.

A previous company of Mr Schoneburg's has developed an interactive Einstein that answers questions about relativity theory, and Artificial Life has Hello Kitty with a licence from Sanrio to teach English. With all these specialised programs, as long as you confine yourself to talking about the specific field, you would think you are talking to a professor - until you switch subjects and get bizarre results.

Daniel Dennett, an eminent cognitive scientist and philosopher, has a special respect for Parry, designed by Kenneth Colby, a psychiatrist and programmer. 'To my knowledge, the only serious and interesting attempt by any program designer to win even a severely modified Turing test has been Colby's,' he wrote in his 1998 book Brainchildren. No program currently available can pass this test, according to Mr Dennett.

But Mr Schoneburg argues it is achievable in principle. 'The only reason why no program has yet managed to pass the test is not a logical impossibility - it's because nobody that works in this field has the time and resources to do it. It's mostly either university labs or small corps like us,' he says.

'If we would only spend 1/100,000 of the global military budget on this task, a program would have long passed the test. It's all a matter of complexity and completeness of the knowledge bases.'

A typical and well-developed bot may know about 50,000 to 100,000 subjects and can avoid tapping into verbal traps in these fields. To pass the Turing test, Mr Schoneburg estimates that a bot would have to master at least several millions subjects, including general day-to-day knowledge.

Vivienne can handle about 30,000 subjects, but will be learning at a rate of up to 3,000 new subjects a month to widen her 'world knowledge' database.

'She currently can only speak English, but is being taught Chinese, Japanese, Korean, German, Italian and Spanish - how is that as a measure of intelligence?' he asks.

'And she will be teasing but no hardcore sex talks. That's not our audience - and by the way, the worst Turing test for intelligence is sex talk, because it is the easiest to simulate because it is usually dumb and you can always avoid direct answers like: 'Can I have sex with you tonight?' Bot: 'Maybe, depends how you behave towards me.' This can go on for hours and it's no test of intelligence.'

Vivienne combines the conversational prowess of smart bots and a high-resolution 3D-graphic world of computer games, so you can chat her up in a bar, visit her ultra-modern apartment and take her out to dinner.

But lonely cell phone users still have to wait until the middle of next month, when local 3G carriers will be ready to hook them up with the virtual 20-something computer graphic designer who works at - you guessed it - Artificial Life: an in-joke, as the fake girl no doubt has some of the personality traits of the group of programmers under Mr Schoneburg's supervision.


For unlimited access to:

SCMP.com SCMP Tablet Edition SCMP Mobile Edition 10-year news archive




SCMP.com Account