The Real AI Threat?

bzs at bzs at
Fri Dec 11 20:43:02 UTC 2020

Slow Friday...

One pressing problem of "AI", and might be a useful analogy, is that
we're (everyone w/ the money) deploying it, for some value of "it",
into weapons systems.

The problem is that decisions made by for example an attack drone
might have to be made in milliseconds incorporating many real-time
facts, much faster than a human can. Particularly if one considers
such weapons "dog fighting" where both sides have them.

Some decisions we're probably comfortable enough with, can I get a
clear shot at a moving target etc. A human presumably already
identified the target so that's just execution.

But some amount to policy.

Such as an armed response where there was no armed conflict a few
milliseconds ago because the software decided a slight variation in
the flight pattern of that hypersonic cruise missile -- Russia claims
to be deploying these, some with nuclear power so can stay aloft
essentially forever -- is threatening and not just another go-around.


The point being it's not only the decision/policy matrix, it's also
that when we put that into real-time systems the element of time
becomes a factor.

One can, for example, imagine similar issues regarding identifying and
responding to cyberattacks in real-time. An attempt to bring down the
country's cyberdefenses? Or just another cat photo? You have 10ms to
decide whether to cut off all traffic from the source (or whatever,
counter-attack) before your lights (might) go out and what are the

I'm sure there are better examples but I hope you get the general

        -Barry Shein

Software Tool & Die    | bzs at             |
Purveyors to the Trade | Voice: +1 617-STD-WRLD       | 800-THE-WRLD
The World: Since 1989  | A Public Information Utility | *oo*

More information about the NANOG mailing list