Artificial intelligence & virtual regret (appeared on L'Express, Mauritius, May 2011)



In this column last week, we looked at how artificial intelligence (AI) grew as a discipline, and how it is slowly filling the gap between technological realities and tall expectations from AI enthusiasts. In this article, we examine some of the more recent progresses in artificial intelligence, and dare to peek into the future.

Science fiction creators have, of course, let imaginations run wild. Science prehistoric times, human like artifacts have aroused interest and fantasy – overwhelmingly destructive in nature, as epitomized by Frankestein. And sometimes positive, like Isaac Asimov’s helpful robots. So, the typical scenario is machines turning into monsters with minds of their own. There is usually not much room for regret, sharing, altruism, and other such consciousness–based faculties of men. These belong more to the realm of emotional intelligence than machine intelligence.  But this may change.

For instance, few artificial intelligence systems so far dealt with “regret” explicitly. But recently Yishay Mansour, from Tel Aviv University's Blavatnik School of Computer Science, is introducing “hindsight” (and the regret it generates) into thinking machines to anticipate the future more accurately and realistically, especially in highly unpredictable environments. Well, the computer may not “feel” the regret of past decisions, but at least it can measure the difference between the actual outcome and the would-have-been outcome and use this wisdom to adapt future behavior. Virtual regret is not just science catching up with science fiction, but has practical applications as well. For instance, Google wants to know how programs that manage unpredictable internet traffic can regret, learn and improve – on their own. Or internet businesses (like online auctions) can use some hindsight to improve decision making and maybe avoid buyer’s remorse!  

With huge strides in unraveling the magic in our brains, we may be getting closer to first understand, and then replicate human traits. A very recent neuro-robotics study by Swiss Olaf Blanke elucidates one of the most fundamental subjective human feelings: self-consciousness. In particular, how our brain gets the feeling of being at a particular position in space and for perceiving the world from there. Or robotics researcher Dario Floreano,working with biologist Laurent Keller, to find where in our DNA is our altruism encoded – they used artificial entities with programmed altruistic potential to prove their point!


Strong and weak AI

Capabilities such as virtual regret may fail to impress many, because we are still far from the real and comprehensive intelligence as displayed by humans (referred to as “strong AI”). Optimism on attainment of human-level intelligence gradually gave way to an understanding of the immense complexity of intelligence.  There has been remarkable breakthroughs in understanding and thereby building very useful artifacts around human senses and faculties, such as memory, search, vision, planning, reasoning, decision-making, hearing, learning, language-understanding (referred to “weak AI”).  These machines can see and hear, respond to questions, learn, draw inferences and solve problems – and also show signs of emoting, regretting  and feeling (even if the capabilities are not all packaged in one unit, and do not necessarily replicate the human faithfully).

The future of AI

For many, we are just at the beginning of a new and exciting era, and even close to a turning point for humanity. For example, in a famous essay in 1993, computer scientist Vernor Vinge envisions a “singularity point” in history where ultra-smart machines, that will be both self-aware and superhuman in intelligence, will be capable of designing better computers and robots faster than humans. In his 2005 book “The Singularity Is Near: When Humans Transcend Biology”, Raymond Kurzweil uses Moore’s Law (that has correctly predicted the doubling of computing power every one and half years) to foresee the arrival of post-human evolution in 2045. Accordingly to Kurzweil, rapidly increasing computing power along with Cyborgs (beings with both biological and artificial parts), would lead to a point when machine intelligence would surpass human intelligence and take over the process of technological invention, with unpredictable consequences.

Let's hope that one day we get to see ultra-smart AI capable of virtual regret, without real regret.

No comments:

Post a Comment