Regulating AI Development

Date: Fri Mar 01 2002 - 13:31:14 MST

I searched through the archives and could not locate prior discussions on
the topic of if we should and, if so, how we should, regulate AI
development through some formal function or body.  I confess I have not
thought very deeply on this subject but after the last few days of postings
it seems it would be productive to discuss if and how AI development should
be regulated.  Substantial discourse is ongoing about what approach might
be successful in achieving certain thresholds of AI.  However, I am fearful
that once the core method for developing AI of a respectable level of
sophistication (Sophisticated AI) is achieved we'll be faced with a sudden
and huge problem of how this particular technology might be advanced beyond
this point without imperling the human race.

There are many schools of thought as to where we go from Sophisticated AI.
I believe Eliezer's general approach (correct me if I'm wrong) is we need
to achieve Friendy AI as soon as possible before potential "evils" such as
self-replicating nanotechnology and/or "unFriendly" AI hit the scene.
Kurzweil's general approach seems to be to enhance ourselves and thus have
"humanity" and all the baggage you get with it built-in to any AI.  There
are other approaches out there, some we are aware of and many others we
might not be aware of.

Should we have a regulatory framework in place to determine as a community,
as a race, how we proceed with developing the seed AI we use for a takeoff
event?  Others have written about this for other profound technologies
(e.g., ).  I can't think
of a more profound product of technological development than a Power.
Thus, if you agree with development-by-committee approach of regulation
then I would presume you would agree that regulation of AI development is

How should AI be developed? I'm trying to resist the "mad scientist working
in his basement" scenario but it or some variation comes to mind when
considering how to regulate AI development.  Regulating weaponized nuclear
capacity is fairly easy due to the huge financial resources and easily
identifiable physical plant requirements to develop such capacity.
Regulating nanotechnology is more troublesome - especially once the general
blueprint for self-replicating nanotechnology is understood.  However,
regulating AI development would seem to be the most difficult to regulate.
I would offer up that the way to regulate AI development is to (1) jointly
determine what the correct approach should be and then (2) invest so much
financial and intellectual momentum into this approach that it is
actualized before any other approach.  Regulation by domination or "beating
them to the punch" seems the best chance we have for ensuring a
human-friendly post-Singularity environment.  This requires faith in the
quorum though - or whatever mechanism is created to make decisions.  What
if the U.S. federal government creates a panel populated with the
_perceived_ gods of the AI pantheon, i.e., Minsky, Lenat, Kurzweil,
Hopfield, Penrose, Moravec, etc.  Perhaps these guys get it all wrong.  How
might we otherwise form such a panel?

Maybe I'm just getting a realization others have already made  through a
different avenue (or perhaps the same), but it seems that identifying the
preferred AI development path and then putting the momentum behind that
approach to get there before something else hits the streets first is the
best way to maximize our chances of survival in a post-Singularity

The information contained in this message may be privileged and
confidential and protected from disclosure. If the reader of this message
is not the intended recipient, or an employee or agent responsible for
delivering this message to the intended recipient, you are hereby notified
that any dissemination, distribution or copying of this communication is
strictly prohibited. If you have received this communication in error,
please notify us immediately by replying to the message and deleting it
from your computer. Thank you. Ernst & Young LLP

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:37 MDT