How right is Trump? The role of the executive in industrial "accidents"

Positions on pausing (gen)AI

Don't confuse the very different reasons people are saying "no" or "yes" to "Should we pause AI?"  Since only "yes" and "no" are feasible answers to a binary question, they mask a lot of differences of opinion.

"yes" could mean…

1. we are in the lead and want the US government to shut down our opposition
2. we are having an IPO and want people to believe our tech is amazing
3. we are honestly afraid, …
1. because we don't know much about the technology
2. because we don't know much about how governance and regulation stabilise complex technologies, services, and economies like AI's all the time
3. because we're in over our head in geopolitics and wealth

"no" could mean…

1. this is the counterproductive framing of a set of rich companies who don't want to take the time and expend the resources to perform proper development practice. Due diligence would provide (among other things) adequate transparency so they could themselves ensure their own products are safe. Transparently demonstrating best practice would also aid their governments in  ensuring any competitors who don't follow good practices are stopped and held liable. (This has long been my position, most recently blogged in the context of Bill Gates' recent essay, but see also my extensive academic publications on AI industry regulatory evasion.)
2. Trump is saying (among other things, but let's focus on the plausible) that a strong executive could solve this problem so there's no reason to stop working on a useful technology. I originally saw  this as just an excuse to prop up his strong-man persona and pander to Chinese president XI, whom Trump will soon be visiting. But I've realised this is wrong (see next section.)
3. Xi for his part has openly said that this is an anticompetitive effort by US companies to maintain their lead. I agree with this too, see yes:1 above. In fact, many of us have been saying this every since Sam Altman asked the US Congress for regulation directly after the Google "We have no moat" blog was leaked in May 2023.

On LinkedIn yesterday I said "Evidently from my framing [beginning and ending the list of reasons with regulatory evasion], I agree with exactly 2 of the 3 positions under "no." For those who are under "yes" only because they don't believe proper, ordinary governance could do this better can take the time to learn enough about what a great tool suite governance and regulation. Then please join us in both:

• - helping make the tech industry a part of a safe, sustainable world, and 
• - reassuring the general public."

But this morning I woke up thinking even Trump was more right than anyone saying "yes."

The responsibility of the executive

I'm being clever with my titles again: there are two different types of executive I'm referring to here.

The executive branch of government

It is indeed the job of the executive branch of government to enforce the law. For example, product liability. If Trump is serious that he will put these guys in jail if they don't start demonstrating excellent diligence, then I'm entirely behind him on that. Meanwhile, I'd like to see US states, the EU, and the rest of the world – all those countries not stifled by an autocracy that thinks it would rather evade money laundering laws than live in the kind of ordered economy that makes money meaningful – collaborate on regulating generative AI firms and the rest of  as the transnational utilities. I've laid out a path forward here a few months ago, and am actively working on developing this policy more formally.

The chief executive officer of a tech company, and the rest of the C suite

These are the people who are responsible for actually ensuring there's no "industrial accident", intentional or not.  Yes, responsible. They should go to jail if in fact something goes wrong because their people are too "overworked" as Anthropic's Amodei claimed. It's literally the executive's job to set staff priorities, which should include demonstrating best practice for DevOps and other AI Systems Engineering, and also to either hire sufficient staff for the projects taken on, or take on projects appropriate to staffing levels. 

So if anyone claims "rogue, sentient AI has escaped" or similar, start getting out and campaigning for the executives to be held to account. All of them.

Opinion, by the Editorial Board: The AI frontier can pace itself: Dario Amodei’s proposal has the unmistakable whiff of rent-seeking. (Click to visit the article.)