[ExI] If Anyone Builds It, Everyone Dies: One Year Closer - Review

[Ben Zaiboc]

On 18/09/2026 00:12, Adrian Tymes wrote:
>
>> 5. By default, an ASI will have motives that are harmful to us
> This is where their claims break down.  The evidence presented is, at
> most, that AI has had motives that are sometimes not caring what
> humans want.  The article then conflates AIs behaving selfishly with
> evidence of active hostility or maliciousness, but these are very
> different things.  The only way I can see it justified is from the old
> slaveholder's mindset - "To an oppressor, being unable to oppress
> others feels like oppression" - and ASI was never going to be
> humanity's slaves.
>
> ASI might be a partner, to those willing and able to see the
> difference between "partner" and "enemy".  Alas, there is a large
> segment of humanity that isn't able and willing to see that difference
> - they believe that anyone of relevance who is neither their superior
> nor their servant must necessarily be personally opposed to them - and
> that segment is a self-made enemy not just of ASI, but of the rest of
> humanity.


That's a very good point.

Switching perspective from 'AIs are potentially our enemy' to 'we are potentially the enemy of AI', the danger isn't AI itself, it's certain humans in the presence of AI.

So it's not all humans that are potentially the enemy of AI, just those ones.

This perspective should be fed back into the whole argument about AI alignment.
-- 
Ben