[ExI] AI 2027: A Realistic Scenario of AI Takeover

[Adrian Tymes]

It's late 2025, and we have already missed the early marks of this
scenario.  That suggests this scenario won't happen, at least not on
the timeline given.

More critically: most of the large AIs - the ones capable of recursive
operation - require large data centers to run on.  Even if they were
theoretically capable of "escaping the lab", there are few places they
could escape to - all of which are heavily monitored, and most of
which are already running rival AIs.  The potential for runaway
unmonitored self-replication is stymied if there aren't enough
resources on Earth to run even 100 copies and all those copies would
be monitored.

There is room for smaller AIs to self-replicate onto a lot more
platforms, but those smaller AIs need to be able to self-improve to
pull off something like this scenario, and those who are running
self-improving AIs generally don't see the point in using smaller AIs
for their work.

On Mon, Oct 6, 2025 at 4:22 PM John Clark via extropy-chat
<extropy-chat at lists.extropy.org> wrote:
>
> The people at AI2027 made this video about what they expect will happen between now and 2030 and it's pretty close to what I think will happen. Spike I really hope you watch it because even if you disagree with it at least you'll understand why I can't get all hot and bothered about the national debt. In their scenario there is a branch point around November 2027, one branch, the most likely branch, leads to human extinction but the other branch does not because the president made a wise decision. The trouble is in November 2027 He Who Must Not Be Named will still be in power.
>
> AI 2027: A Realistic Scenario of AI Takeover
>
> John K Clark
> _______________________________________________
> extropy-chat mailing list
> extropy-chat at lists.extropy.org
> http://lists.extropy.org/mailman/listinfo.cgi/extropy-chat