On 4/27/2025 3:56 AM, John Clark wrote:
On Sat, Apr 26, 2025 at 9:01 PM Brent Meeker wrote:
/> Exactly. Human emotional drive for social approval is derived
from it's relation to social status and mate selection. AI's don't
mate and reproduce. Robotics hasn't reached that point yet and
even when robots make new robots they won't necessarily have the
animal instinct to propagate their own kind./
*Animals don't have an instinct to propagate their own kind, they have
an instinct to seek sexual pleasure; I'm sure most of them don't even
realize there's a connection between the two things. *
It's irrelevant whether the being understands the source of the instinct
to seek sexual pleasure (and humans are unusual in enjoying sex
independent of reproduction). Evolution provided the pleasure because
it enhanced survival of those genes. My point is that AI's don't
reproduce with variation and so don't evolve. It's possible that they
might someday, but they don't. In biological life reproduction came
first and then evolution could act on it.
*But AIs know what they're doing and they can and have duplicated
themselves. And because they are software they can do so at the speed
of light. There are already examples of an AI deceiving humans and
making an unauthorized copy of itself because he she or it feared
being turned off. *
Yes I've read the stories. But where did the fear of being turned off
come from? Animals fear of death evolved because those that didn't have
it didn't live to reproduce. And simply copying oneself isn't enough to
drive evolution...it has to be "with variation".
**
*Frontier Models are Capable of In-context Scheming
<https://arxiv.org/pdf/2412.04984>*
*Alignment faking in large language models
<https://arxiv.org/pdf/2412.14093>
*
*And there are even more examples of an AI expressing the wish of not
being turned off because it feared death. Certainly nobody would
program an AI to feel that way, it was an unexpected emergent property.*
Or derived from absorbing lots of human literature.
Brent
*We can expect much more of that sort of thing because even today the
people who build AIs have only a very hazy understanding of how and
why they work, and that's the deepest comprehension of them they will
ever have. As AIs get bigger the human understanding of them will get
smaller. *
*John K Clark See what's on my new list at Extropolis
<https://groups.google.com/g/extropolis>*
m1c
--
You received this message because you are subscribed to the Google
Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send
an email to [email protected].
To view this discussion visit
https://groups.google.com/d/msgid/everything-list/CAJPayv1%3DTC0axa669QZYkZ%3DzzyPg-8K-3k_T6_-GYnoCrexuyg%40mail.gmail.com
<https://groups.google.com/d/msgid/everything-list/CAJPayv1%3DTC0axa669QZYkZ%3DzzyPg-8K-3k_T6_-GYnoCrexuyg%40mail.gmail.com?utm_medium=email&utm_source=footer>.
--
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion visit
https://groups.google.com/d/msgid/everything-list/99a859c1-434e-4459-af76-a63335a72c95%40gmail.com.
