Re: [OPE] Digital investor worlds

From: Ian Wright <>
Date: Mon Jan 17 2011 - 13:52:58 EST


Just a quick point from someone who works daily on Artificial
Intelligence problems, including data mining and machine learning.

> Computers don't have a morality permitting flexible judgements about moral
> situations - they have at best a set of priority rules programmed into them
> ("if you strike X, do Y, and it that does not work, do Z").

That's not an accurate representation of the variety of
decision-making mechanisms that are currently in existence today. AI
has progressed a long way from the rule-based expert systems of the
1980's. And often these new kinds of mechanisms are regularly used by
millions of people without them realizing it -- e.g., much of what
Google does is AI.

For example, consider the support vector machine:
I include the link -- not because I expect you to read it -- but as an
example of a kind of decision-making algorithm that is nothing like an
IF-THEN rule. It involves geometrical and probabilistic concepts. The
classifier is induced from empirical data, not hand-built or
programmed etc.

Often AI data-mining approaches find relationships in data that are
missed by humans etc.

> I've also worked in survey design some years, and there
> you realize not only how easy it is to fabricate nonsense data, but also
> that practical human judgment by experienced people is absolutely essential,
> since many of the necessary interventions defy program logic; they have to
> bring together information at a specific point in time in a way for which no
> program exists, and toggle between different "levels of abstraction" to use
> Jerry's term.

I don't think anyone working in AI should claim to be able to replace
human intelligence. But in restricted areas on restricted
problem-domains AI already surpasses human performance.

I also agree that the technology we build actually changes us: so it's
a matter of co-evolution, not replacement.

ope mailing list
Received on Mon Jan 17 13:55:08 2011

This archive was generated by hypermail 2.1.8 : Mon Jan 31 2011 - 00:00:02 EST