[ExI] AI thoughts

Jason Resch jasonresch at gmail.com
Wed Dec 27 05:00:29 UTC 2023

On Tue, Dec 26, 2023, 11:42 PM Keith Henson via extropy-chat <
extropy-chat at lists.extropy.org> wrote:

> I wonder if AI might be less dangerous and less useful than it seems.
> Intelligence is the application of knowledge to making decisions. The
> LLM to date have depended on human knowledge for training. The point
> here is that there may be a limit to knowledge. For example, all the
> knowledge in the universe is not going to find a new element between
> carbon and nitrogen.

While existing AI is trained on human generated data, there's no limit to
the amount of data an AI might choose to generate itself based on
self-play, simulation, or exploration of the unlimited
mathematical/platonic reality, or what you might call the realm of

Left to its own devices, and provided unlimited computational resources, I
think an AI could overcome an initial dearth of knowledge. The AIXI
algorithm for instance, starts with zero knowledge but achieves perfect
universal intelligence. Similarly, Google's AlphaZero was provided only the
rules of chess, but through self-play quickly discovered and mastered all
the common openings that took humans centuries to discover. And Koza's
"invention machine" applies principles of evolution to come up with novel

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.extropy.org/pipermail/extropy-chat/attachments/20231227/7d660fdd/attachment.htm>

More information about the extropy-chat mailing list