Re: rambling question on AI

From: Samantha Atkins (
Date: Fri Jan 11 2002 - 14:52:33 MST

Adrian Tymes wrote:

> Colin Hales wrote:
>>With this in mind and as an avowed "AI or bust" person I cogitate regularly
>>on the issue of impact - both sides - and I confess this is like
>>(Automation^^N) in it's impact and it scares me. For example: Computer
>>Programming. What's the 'big software project blowout's' overall solution
>>and the next programming language? The answer is _not to let humans program
>>anything_. There goes computer programming as a profession (I include myself
>>in this profession!).
> Well, let's see...
> AI can take a self-consistent set of requirements and turn them into
> running code? Show me the CEO or marketer who can give a
> self-consistent set of requirements, *especially* in commitee. (Well,
> ok, they do exist, but they're rare.) Transition to translator, and
> ex-coder still has job.

I can't resist responding to this. It is the bane of my
existence as a software architect that too much "architecture"
of new systems today is done by marketeers and CEOs and various
levels of managers who generally wouldn't know good computer
architecture if it bit them. Committees do not produce coherent
graceful systems either. At best committees and the marketeers
and CEOs add to and help refine requirements. The should never
be seriously involved in analysis/design/implementation except
to track whether the requirements are fulfilled and to arrange
for prototype/feedback with end-users (from an organizational
perspective, not coding/design).

It is not the programming itself that is the tricky part. It is
the architecture/design and the continuous refinement of that
design from multiple perspectives that is really tricky and
often simply and surprisingly NOT DONE. It is also the most
difficult to automate. You can automate a lot of tools that
allow the architect/designer[s] to do this level of work more
cleanly and without getting caught in too many low-level details
but it is not clear that you can remove humans from this loop
entirely as long as the software is for use by humans and is in
any way maintained by humans. What an AI, an early one at any
rate, could be very good at is choosing existing design
components and balancing them against a set of requirements and
producing new localized components. It will take a much more
sophisticated AI to orchestrate the creation of entire libraries
of new components and useful abstractions for problem classes.
It would be extremely interesting to attempt to design such a thing.

> AI can also translate from vague, incoherent requirements into running
> code, including the inevitable arguing back and forth to nail the specs
> down enough? "Hey, AI I just rented: hack into various corporate
> networks' financial programs and score me a nest egg to retire on. Make
> sure no one, except maybe you and/or I, can trace the hack or anything
> related to it back to you and/or I."

Piece of cake. :-)

> Ok, that's a gimme, and it'd only last until the AIs taking over the
> corporate networks finally implement adequate security. Where next for
> the ex-coders not lucky enough to get a job working on these ever
> improving AIs? Tinker with the AI a bit and, "Hey, AI: find me a job."
> ^_^;

Ultimately I hope every software engineer out there has a dream
of automating more and more of what they do. And I hope we
organize our society and economy in such a way that doing so is
not suicidal, and more to the point, is the ultimate win. If I
am protecting my ass by not creating such a system then frankly
I might as well retire right now. I can't work like that.
Creating such a system is actually darned difficult. Look at
our so-called tools today and how difficult it is to build a
good business plan for a company to advance those, much less to
replace programmers and then tell me it is likely to come about
any time soon. Go have a look at the software CAD tools and
development environments.

>>I just hope that the speed of the change is moderated so that people are OK.
>>Sometimes I have nightmare visions of being ostracised at the local market
>>and having people throw rocks at my house out of the fear of the
>>(potentially) very smart little grey box in my office.
> If I'm ever in that situation (which implies moving away from an area
> where many residents would have a similar box), I'd probably also be
> rich enough to buy the local police and have the rock-throwers arrested
> en masse.

That isn't the point though, is it? The point is how to make
expanding technology including AI enough of a boon to everyone
that it is not seen as so threatening as to lead to violence
possibly escalating into great danger to developers and even to
wars which would slow down the advancement we seek. It is a
very important issue.

> "Attention, oppressed masses of the Muslim world. What do you say we
> beat the infidels at their own game? Lo, they have thought of a way to
> bring Allah's wisdom to the masses. See how they tremble in fear when
> they contemplate its potential existence. Let us bring their nightmare
> to life, and in so doing, bring about our new golden age."

There are actually "whiter" versions of this that could be very
important in getting to the Singularity relatively peaceably and
in one piece.

- samantha

This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 13:37:34 MST