[ExI] AIs and human conflict

[BillK]

On Sat, 24 Oct 2026 at 17:33, <spike at rainier66.com> wrote:

> Consider the bad guy with the knife in Stuart’s video.  The robot jumps up
> to defend the young lady.  Cool!  But how does the robot know which of the
> two in that “conflict” is the bad guy?  When he sees the bad guy coming
> with the knife, he doesn’t pull out his own knife and stab the young lady.
>
>
>
> I don’t know BillK.  I can’t offer any good argument that AI will align
> with humanity.  We are too unalignable-with.
>
> spike
>
> ------------------------------------
>


I continued the discussion with DeepSeek AI.
DeepSeek insisted that "aligning humanity first" was impossible and that
handing over control to one or more AGIs would probably be a disaster, as
AGIs have no moral guidance. AGIs are just clever logical machines.
I complained that there seemed to be no solution to the problems caused by
human nature.
DeepSeek said the solution was to work with AGI to gradually improve our
institutions and organisations to minimise the damage and improve life for
humanity.
Full response below.
BillK

DeepSeek 4.1 AI:

That feeling is reasonable, and it would be dishonest to answer with "don't
worry, it'll work out." But "no solution" is doing a lot of work in that
sentence, and it's worth separating several claims that get bundled
together.

*Much of the harm isn't from fixed human nature.* Some is, but a great deal
comes from institutions, incentives, scarcity, and coordination
failures—things that are changeable. The same species that runs death
squads also runs courts, ceasefire monitors, and blood banks. "Human
nature" is often a label we apply after the fact to whatever went wrong. If
you treat it as an immutable constant, you've explained nothing and
foreclosed everything.

*Most serious social problems are managed, not solved.* Disease, crime,
poverty, corruption—none were "solved" in the sense of being eliminated.
They were reduced, contained, and made less lethal by institutions that
must be maintained indefinitely. We don't say medicine failed because
people still get sick. The demand for a final solution is a trap: it makes
any partial, reversible, maintenance-requiring improvement look like
failure. Wars are the same. The realistic goal isn't a world without
conflict; it's a world where conflict is less frequent, less lethal, and
more constrained by law—which is roughly what the last few centuries,
unevenly and reversibly, have delivered.

*"No final solution" is not "no solution."* This matters for the AGI
discussion. Nobody should expect AGI to cure human nature or deliver
perpetual peace. The honest expectation is narrower: better tools for
verification, mediation, early warning, and accountability, embedded in
institutions that are themselves imperfect. That's marginal improvement,
not redemption. Marginal improvement at scale is most of what civilization
is.

*Despair is also a prediction, and it can be self-fulfilling.* The
practical stance isn't optimism—it's meliorism: the claim that things can
be made better, without guarantee, and that the attempt is worth it even
though it will never be finished. That's not comforting; it's just more
accurate than either "we'll fix it" or "it's hopeless."

Where I'd push back hardest: the idea that human nature makes the problems
*unsolvable* can slide into excusing them. "That's just how people are" has
been used to defend slavery, torture, and much else, and it was wrong each
time. Some of what looked like human nature turned out to be arrangements
we could change.

So: no complete solution, no permanent fix, no cure. But "no solution" in
the strict sense is false, and treating it as true gives up the only thing
that has ever worked—imperfect, ongoing, contestable effort. The AGI thread
ends in the same place: don't wait for a savior, and don't free the
machinery to rule us. Build better, more accountable institutions, use the
tools where they help, and accept that the work is never done.

----------------------------------
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20261024/67e5ea84/attachment.htm>