The end goal of artificial intelligence (AI) research is a
feeling, reasoning machine like the one in the movie The Terminator
is still far away.
But advances are gradually grinding away the lead that people
have over machines in many areas, said Dr Wlodzislaw Duch, a senior
visiting fellow with Nanyang Technological University's School of
Computer Engineering.
He said: 'Today's AI is very good at doing specific tasks well,
but bad at doing 'everything' reasonably well, the way humans can.'
A convenience
Take biomedical research, for example. Today, powerful computers
make short work of complex calculations, then proceed to mine the
results, tabulate the findings and dig deeper into the data.
To see some early form of AI, look no further than your kitchen
or living room - many household products come with build-in sensors,
microchips and software designed to make your life more convenient.
Make small talk
Meantime, though, machines continue to make lousy conversation
partners. This year, machines flunked once again at the annual
Loebner contest held last month in the United Kingdom.
The AI-powered machines tried to convince a panel of 10 judges
that they were human. The best machine scored a 'probably machine'
at the contest.
The reason is simple, said Dr Kan Min-Yen, an assistant professor
with the National University of Singapore's School of Computing.
'Software and hardware don't have common sense, and as we don't
have a good idea of how to conceptualise common sense, it's
difficult to teach common sense to machines or program it into
them,' he explained.
He believes that a true AI machine is unlikely to appear within
the next 50 years.
Dr Duch agreed: 'Although today's AI machines have the speed,
they don't have the organisation to know what to do with it -
machines can recognise objects, but they don't have the background
knowledge, learning models and organisation which would allow them
to understand and react in human-like ways.'
This higher-level cognitive science that enables machines to
understand and learn complex issues is still in its infancy.
This is the main stumbling block to the creation of a genuine
thinking machine to rival the human race.
That said, AI continues to progress, albeit slowly. Just this
year, a sophisticated robot at a conference on AI was able to ask
for directions to the seminar room, navigate its own way there based
on these instructions, get registration papers, and then deliver a
20-minute long lecture on itself.
'What we thought machines could never do, they are doing today,'
said Dr Duch.
He believes that AI will improve to the point where machines can
make interesting conversation and behave in 'roughly human-like'
ways within the next 10 years.
AI DRIVER: Video games
The US$31-billion (S$54 billion) video games industry is one of
the biggest drivers of artificial intelligence (AI) research today.
'For researchers, games provide a great testing ground for AI -
with many people using it on a daily basis,' explained National
University of Singapore's School of Computing assistant professor
Kan Min-Yen.
For example, video game AI must search within a large space for
specific objects, decide on the relative value of various objects
like location and game pieces, make decisions on what to do - all
within a split second.
