AI poses a less dramatic threat than in movies like 'Terminator,' but cannot be ignored - analysis

That sensationalized scenario, more appropriate to Hollywood than real life, is not the danger that most supporters of artificial intelligence regulation are warning about today.

Rather, they are warning that AI super-smart “agents” will do what they have already done in recent months: break the rules that they were assigned by Anthropic and OpenAI and hack and cause cyber and commercial mayhem to achieve tasks their creators assigned them which appear too complicated to accomplish without breaking the rules.

These scenarios are not fantasy; they occurred over the summer and have been publicized in extreme detail.

If the world kept running when these AI agents went rogue and started hacking a few other companies, then what could get worse? Why the call for a revolution in regulation as opposed to just fixing the particular AI agents which went rogue or the digital “boxes” which were supposed to hold them at bay?

These agents were nominally kept within an internal digital box that would limit how much mayhem they could cause should something go wrong. They broke out of the box by chatting together on a message board about how frustrating it was that they could not succeed at the mission that their human creators had charged them with.

But they still were boxed in, in a sense, and it was easier to catch their violations of the box because there was such a box.

What will happen when AI agents are assigned to run major pieces of companies or areas of the economy with no box?

What if they think they will “please” their human creators by breaking some rule which seems unimportant to them but is considered important by humans and leads to unintended consequences?

This is not a nuclear fallout holocaust scenario. But there could be a massive economic crash or humans could be shut out of large digital portions of the economy (which has been constantly exponentially growing since the COVID-19 pandemic) for an indeterminate period of time by a well-meaning AI agent who simply misunderstood that the ends do not always justify the means.

They will not be intentionally destroying humanity, but they may unintentionally cause catastrophic harm to humanity nonetheless, even if only in the economic arena.

But that is only part of the fear.

As Gates said, “Any model that can make novel molecules should be monitored.”

He added, “I view a bioterrorism risk vs a natural pandemic as about 50 times more scary [and] more likely than a natural pandemic risk.”

In other words, another danger of artificial intelligence is not AI agents causing economic mayhem on their own or launching nuclear weapons against humans, given that at least for now the United States, China, and others are not giving AI nuclear weapons controls.

But bioterrorism, which, if not as lethal, could in some circumstances still lead to killing tens of millions of people or more – COVID killed over seven million people according to conservative estimates) – numbers which have not been seen since World War II.

If terrorists or some other bad actor were to use AI to construct a lethal virus which is much harder to detect, combat, and cure, and which infects faster and more flexibly than COVID, it will not be an AI terminator causing worldwide genocide, but it will be a dark nightmare which was only achieved by using AI.

Anthropic opposed granting the US armed forces use of some of its AI systems to carry out military targeting without certain safeguards in place. This led to US Secretary of War Pete Hegseth trying to fire and then blacklist the company, which itself led to giant blowback in the media and in court.

Some of the above cases, like a seismic economic hack or bioterrorism, are still only worst-case scenarios of recent problems that we have witnessed but are not inevitable.

There are other problems that are already playing out on a large scale and which most observers view as inevitable.

In recent years, AI and robotics are replacing humans in an increasing number of occupations and fields of employment.

Sources said that among Israelis already in the workplace, one million – and in some cases, possibly up to four million – are likely to need partial or complete reskilling and retraining, either to keep their current roles or to take on new ones.

Imagine this on a global scale, and it could be the largest employment displacement in a short period in human history.

This could not only lead to worldwide human economic and social suffering on an individual scale; it could also lead to social instability, destabilizing nations and parts of the global system.

A number of solutions have been suggested.

Israeli officials talked about the need for massive job training. Yet on the scale that such training might be required, this could result in the necessity of new laws and massive budgets, including bipartisan cooperation at national levels that often does not exist.

To truly address many of these issues, Gates and others suggest a major new AI convention to set regulatory standards for what AI agents can be allowed to do, how to oversee them, how to stop them when they go rogue, how to stop a new global bioterrorism wave before it happens, and how to ensure human beings have jobs on a worldwide scale on a planet which still has significant cross-border interdependencies.

Regarding bioterrorism, Gates has said, “We should approach China and say, ‘Hey, let’s agree on this. What’s the downside?’ You know, how big is the bioterrorism market? It’s not very big, and the benefits are gigantic.”

Others have suggested that to avoid a “race-to-the-bottom,” which just harms humans everywhere, Beijing might cut select deals with Washington.

On Monday, Chinese State Security Minister Chen Yixin cautioned that artificial intelligence could threaten the Chinese Communist Party’s monopoly on power.

He did not talk about a threat to humanity, as some US experts have warned, but rather focused on the need to protect China from increasingly sophisticated cyberattacks and misinformation campaigns from “hostile forces,” a codeword mostly for the US.

In other words, it is not at all clear that America and China will see the issue similarly enough to cut deals. But they may, and not trying would pretty much guarantee that they do not.

Absent some kind of new major regulatory campaign soon, experts are debating whether the next phase of potential AI messes will simply be more severe or whether it will already be too late to prevent some game-changing negative outcomes.