Rob Wiblin

The first one of which was surveillance and detection of diseases.

And the fourth one was medical countermeasures, which are the other two things that you were talking about.

Yes, yeah, yeah, very much so.

Obviously a ton of overlap.

What would you like the AI companies to do differently?

Yeah, a question from the audience was, all of the major AI companies have bio evals.

How good or trustworthy are they?

I thought you were going to say that we don't even know how good or trustworthy they are because when they report the results in model cards and things like that, they don't actually give us that many, I guess for security reasons potentially, they don't give us many details of specifically how they've done the experiment or all of the details of the results.

So another on-point audience question was, should we expect a terrorist use of chemical weapons or widespread enablement of other attack types significantly before we see bioweapons enabled by AI?

And do they succeed more?

Is that basically just because it's easier to do a chemical weapon attack because it's easier to make a poison than to make a living thing?

So I guess precisely for that reason that chemicals don't spread autonomously, they don't replicate and go worldwide, the potential harm from chemical attacks is much, much, much smaller than creating a new pandemic virus.

But I think you said in your notes that you think despite the fact that the scale of the expected harm is a lot lower, that you would still like AI companies and I guess people in the AI governance space to be paying more attention to chemical weapons threats than they are.

Why is that?

I would, much more.

So the notion is that if we found a group or an individual attempting to do a chemical attack using AI, then we could study what they did and what went right and what went wrong and what might have prevented that and then apply that to the biological case where the scale of the damage is even greater.

Exactly.

And that's not happening?

Not one of the cool kids.

Okay, so the argument here is like the scale is not so radically smaller because the worst case scenario would be people figure out how to do these chemical attacks, which I guess if they were pulled off like extremely well could kill millions of people in a concentrated area.