Fear and loathing in AI

Rarely has the AI news cycle been as dire and urgent as it is right now. With OpenAI, Anthropic, and even xAI seemingly in lockstep regarding the upcoming AI doomsday, the broader tech community has become markedly split.

But when everyone, at once, becomes fearful and anxious, I find it most useful to ask:

Who benefits from selling us the solution to our fears?

A call to slow and regulate AI

Let’s first talk in generalities; there has been a unified call throughout the AI industry to slow AI development, ostensibly out of fears of some coming AI apocalypse.

This has given rise to a few theories from the broader public:

• Some believe AI companies are trying to delay their IPOs because they know they are in a bubble and can’t show real growth, or because their technology has largely plateaued.
• Others believe that something terrible has happened — either with respect to the Hugging Face incident or something that is still a secret — and that AI companies are truly terrified of what they have created.
• And still others believe that while AI companies truly do fear their technology, they do not want to pump the brakes while their competitors advance; they’re looking for the government to step in.

Here’s what we know to be true. The concerns outlined by the major AI companies are not new concerns, nor were they fundamentally altered by the Hugging Face incident (which Cory Doctorow breaks down better than I ever could). So, there’s a question of why now: Why would companies decide to halt development at this moment based on concerns that have been obvious for years?

And that’s not to say AI doesn’t need to be slowed and regulated, because it does. But it needs slowing and regulation no more or less today than it did 6 months ago, 12 months ago, or 24 months ago. While the average journalist, hobbyist, or enthusiast might not have realized it back then, these companies certainly did.

So again — why now?

The case for the AI bubble

First: an AI bubble does not inherently mean that AI technology, itself, is not transformative. We experienced a Dot Com Bubble even though the World Wide Web certainly transformed technology and the world. Pretty much everyone, even those bullish on AI, agrees that AI is in some form of bubble. Where people disagree is when it will collapse — and with how much aggression.

Under this theory, AI companies are delaying their own IPOs because they know the bottom is about to drop out. They can delay their own IPOs and remain noble and valuable by saying that they are simply saving the world from their incredibly advanced, incredibly dangerous technologies. And, potentially, this theory explains why AI companies would start worrying about regulation and downshifting now, as their IPOs approach.

But on the other hand, who would invest in a world-ending technology? Surely there must be a limit. Critics of this theory point out that it can’t possibly help a company’s bottom line to state that their technology is fundamentally too dangerous and must be regulated — that there is no direct financial incentive.

The case for something terrible

Social media is filled with panicked posts asserting that people do not understand the dangers of AI, that they don’t understand the importance of the Hugging Face incident, and that they don’t understand that the world could very much end very soon.

Professionals understand that these security incidents were entirely predictable; they are the result of researchers creating an infrastructure that was not properly locked down or permissioned. That being said, it’s certainly the case that these incidents provide a proof-of-concept that the technology is extremely dangerous in the wrong hands.

And perhaps the industry is awakening to this.

Compounding these fears, every year or so, an “AI researcher” will come out and speak against AI. In 2022, it was a Google engineer who swore that an LLM had gained sentience. This time, it was an Anthropic AI researcher who had been in the industry for about four years.

This has driven an underlying narrative that there was a “whistleblower” who then prompted the companies to finally take a break to reassess.

The case for all of the above

Finally, some believe that AI companies have been truly, pervasively concerned about the breakneck pace of AI development, but unable to slow because they are worried about ceding ground to the competition.

Recent attacks, like the HuggingFace attack, along with the “whistleblower,” could give these companies the opportunity to call for a mass slowdown of the entire industry. They cannot trust each other to slow down, so they instead are looking to the government.

And some wonder if this isn’t a rationale for regulating and controlling open models, such as the models coming out of China — models that, potentially, domestic companies wouldn’t be able to compete with.

Either way, the answer here is a mix of genuine concern and regulatory manipulation.

But is it possible that everyone’s stupid?

It’s not an outlandish question. In tech, we sometimes lift up “personalities” who are driven more by confidence than acumen, more by stage presence than domain expertise.

We don’t want engineers; we want visionaries. And sometimes a visionary is just a man who doesn’t understand how dumb he is.

Is it possible that the heads of these trillion-dollar companies just caught up to where the industry has already been?

Each of the theories being tossed around has its merits and conflicts. AI does need to be regulated and slowed; perhaps these companies have just realized that. The AI bubble is probably going to burst, eventually; maybe these companies don’t want to tie themselves to the pop. And no one really wants to be left holding the bag alone when the bubble finally bursts.

But what is true, regardless, is that fear is becoming pervasive.

The terrible gifts of fear

Fear is reactionary. Fear lets people be misdirected, redirected, and controlled. With this much fear circulating right now, someone is likely benefiting from that — whether that involves share prices or favorable regulatory controls.

When we do regulate AI, when we control AI, when we create rules and laws around AI, these things must be driven by logic and empathy, not fear. Environmental damage and job losses are far greater concerns than a nebulous doomsday scenario.

Because, for the most part, we do know that these fears are still broadly overblown.

As far as we know, today, “AI” technology has not done anything unexpected. It has not crossed the chasm into singularity. It has not “broken” through its constraints. It has acted as predictably as a non-deterministic solution can.

AI is not a doomsday device. It is a tool that can be used by irresponsible people.