Trust me

One of the early promises of the 80's computing revolution was the true dawn of Artificial Intelligence. By the end of the century, so scientists said, Japanese robots would be helping old women cross the road.

AI suffered a sort of semantic breakdown, as it turned out that no one knew what intelligence was, and therefore whether it made sense to say it was artificial or not. Philosophers such as John Searls with his Chinese Room problem started to question the validity of AI as a concept.

Maybe the AI community underestimated how hard it would be to simulate even the "unthinking" parts of the brain - just vision and walking are still challenging today. But while many strands of AI have been abandoned, the brute force computations that computers have always excelled at have been improving. Hence we do have good expert systems.

I was listening to how an expert system has been developed that differentiates between genuine and fraudulent works of art by examining brush strokes in detail. While the system was able to discover some fakes that had previously passed as genuine, it also mistakenly endorsed some known fakes. Oddly, these endorsed fakes still found a value in the art market, which recognised the "authenticity" of the fraud. The fact that a fraudster does not know the original artist's PIN number is not the issue.

This leads to the modern day problem of identity and attribution. This is best seen in politics, where candidates reveal less and less about policies ahead of taking power, prefering to play what-if games. They need to persuade their electorate to trust that they would make the "right" decision in any given event, only going by previous (and possibly unrelated) decisions.

It won't be that long before an expert system will claim to accurately extrapolate what decision a leader might make based on their previous decisions, as well as other leaders in similar positions and circumstances in the past. I use the word "claim" because with all predictions, they only matter if people take them seriously. This will lead to more assiduous recording of all speeches and decisions by anyone remotely in the public eye - if that is possible. Then it can be fed into the system and matched with patterns, rules, and learning.

Somehow, a person's identity comprises of the sum of all their communications and interactions. The brain already deals with this in its own way - and creates appropriate trust models necessary for survival. We have always rightly distrusted politicians who want to set a year zero on their personal histories, which usually starts after they smoked pot or slept with their relatives. To see what Barack Obama will do in Iraq, it may be necessary to look at his behaviour in nursery. Just to seed the model with some data, you understand.

Apart from projections, a decision actually taken could be judged to be atypical, or even fraudulent. After all, if you sell yourself on a set of given behaviours, then why can't you be criticized for making a bad impression of yourself?

Artists do of course break out of their oeuvre occasionally, sometimes to great applause, sometimes not. But politicians, who want to avoid the accusation of not being "real", tend to follow the same track for their whole career. This makes them very good targets for modelling. I'll leave out any pun about being painted into a corner.