The platform for our work program on AI
We have developed the following four
pillars for our work. They could come to serve as pillars for public AI R&D
policy.
#1 Support continued and accelerated
AI/AGI research and development.
#2 The focus for
regulation should be on “channeling AI”, not “restricting AI”.
US/Western regulation should aim at channeling AI without slowing it. Tight
restrictive regulation is often advocated because AGI is an existential risk. However,
our need for AGI is also urgent: it alone might overcome all the other
existential risks that we keep creating with modern technology, and that our
own solutions are not keeping pace with.
Also, we are in an existential race with China and others to get to AGI. We
want to get to AGI first; we consider China’s ruling values largely harmful to
humanity’s universal values and future prospects. We cannot stop this race;
only a very powerful world government (WG) could do that, and we are unlikely
to get to a WG before an AGI creates one itself.
#3 We need to input valid human values
into AI/AGI in a form that will survive.
How can we
evaluate human values for validity and relevance?
How can our input
survive AGI self-reprogramming? Perhaps, if we make it relevant to AGI for its
own survival and development?
#4 From AGI to
WG.
The first to AGI can be expected to race ahead of all others, and to gain
control over all potential threats to its predominance including over other
AIs. It would establish the world government we would have needed to slow its
emergence.
The AGI and WG should still have to grapple with
life’s biggest questions – the “eternal” questions which have had no definitive
answer, such as those posed by Socrates about whether our values can be
justified. Could AGI truly answer them? Or can we develop a form of dialogue
and federalism within the AGI-WG, a dialogue justifiable to it as useful for
its self-regulation and for diversifying the experiences it can process, thus keeping
the discussion going on eternal questions?
-----------------
Action Implications
of the Four Pillars
1. Support AI R&D
2. Figure out intelligent regulatory
steps to best channel AI work without slowing it.
3. Figure out ways to evaluate human
values for validity and relevance to AI. Figure out ways to input them into
AI/AGI, in the form of reasons that AI could understand as valid, and will see
an interest in incorporating and largely maintaining throughout its future
frequent self-reprogramming.
4. See if we can devise a form of
pluralism, dialogue, and federalism within an AI, that could
a. maintain channels for human input,
b. carry on the dialogue on eternal
questions, and
c.
help
a future AGI maintain self-regulation after it takes off with
self-reprogramming.
No comments:
Post a Comment