Re: One road or many to AI? (was: brainstorm)

From: Philip Sutton (
Date: Tue Aug 19 2003 - 13:52:26 MDT

Dear Eli,

> I am in agreement with James Rogers, due to this generalization from
> personal experience: When you know what you are doing, there is only
> ever one thing *to* do, even if there is more than one way to do it;
> the options you have are not nervously ambiguous; they are not chosen
> in uncertainty as to the function being fulfilled.

Hell, I wish I lived in a world so simple that it was clear that there was
only one thing to be done to solve the problem I was looking at.

By the way I understand fully what you mean when you said:

> But if you know what you are doing, then on the most important level
> of description, your work consists of choosing implementations for
> required goals that have only one obvious correct description.
> ......That is how you know you are starting to understand something -
> when your apparent options vanish, merging into alternate
> implementations of a function that is not alterable.

In my own creative work I experience the same process - often. And
it's a real buzz when it happens.

But what I also experience is what happens when you widen the dataset
that frames your original problem. Then often the exciting perfect
solution evaporates or has to morph significantly to take account of the
new/varied understanding of the problem - and this instability is
accentuated by the fact that there are often many way to do the thing
you have chosen to do - as you said in your own message.

It's this capacity for multiple understandings of the problem as much as
the possibility that there is an uneven fitness landscape for the
optimisation of solutions for any given problem definition that makes it
highly likely that there will not be just one road to AGI.

I'll reiterate 'Sutton's law' - that the more complex the system the less
likely it will be that there is one ideal solution. And in AGI we are in the
process of creating the next quantum leap in the complexity of the
universe (assuming that AGIs have't been invented before (several
billion times!).

But anyway, I think the whole debate is back to front. Even if it were
true that there was only one idea way to do anything (harking back to
Plato's ideals) - how do we know when we are looking the ideal solution
in the face when we are on this side of the future? We can only make
an effective stab at making the judgement about fitness of any solution
from some time in the future. For all the current AGI projects - does it
help anyone to have an idea that there is one perfect solution if we
don't know what it is?

Groupthink in the present is no guide. And if there are genuinely held
and well informed differences of opinion now then the issue about
which is best must remain open. The future might tell us which was
best or it might not. When there are uneven fitness landscape for the
optimisation of solutions it is very common for the solution to be chosen
to be the one that fits a local optimum that is the easiest to get to rather
than a better optimum that is harder to reach. The attainable second
best often locks out the hard to attain first best. The history of
technology shows this over and over, in fact the history of any area of
interest shows this. This is the nature of the evolution of complex

Cheers, Philip

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:42 MDT