Let me begin with a brief caveat. This column is written about two weeks before publication for a magazine that comes out on the first day of each month. For that reason, when dealing with a topic like this, where developments are moving so rapidly, I ask for your understanding if some of the information has already become outdated.

The capabilities of AI appear to be advancing at a tremendous pace.

On September 12, Dario Amodei, CEO of major US AI developer Anthropic, posted a message on his blog arguing that development of frontier AI must be slowed. His point, it seems, is that AI capabilities are advancing faster than safety measures, and that if development continues unchecked, there is a risk that AI systems will become impossible to control. He also warns that, if the current situation is left as it is, AI could be misused in cyberattacks or terrorism, creating a risk of serious social disruption.

I read the blog post itself, titled “We Must Pace the Frontier,” in translation. I cannot say I am entirely confident that I have understood his message correctly, but this is how I understand it.

As long as humans were the ones improving AI performance, it was still possible to keep it under control. But as that process continued, we have recently reached the point where AI itself has begun improving AI performance. Once that happens, AI will evolve at an accelerating pace, in an endless loop of self-improvement that humans can no longer control. Unlike humans, AI has no ethical judgment, moral sense, conscience, fear, or sense of responsibility, so what awaits us is likely to be a tragedy for humankind. Of course, if AI can be properly controlled, it can bring people a wide range of benefits. That is why there is an urgent need for AI developers, governments, and, if possible, the world as a whole to put appropriate regulations and rules in place.

The “crisis for humankind” that Amodei fears is a standard storyline in movies. An enemy of humanity appears from somewhere, or is born from somewhere, grows too powerful to control, becomes a great evil, and attacks humankind. In movies, though, the usual pattern is that good people overcome the crisis through wisdom and courage, and the story reaches a happy ending. Needless to say, if something like this were to happen in reality, it would likely be difficult to prevail over AI.

Now, Anthropic, the company led by Amodei, is expected to go public in the near future. Sending a message along the lines of “let’s slow down the pace of development” is negative from a financial standpoint and could even damage the company’s value. That makes these words, coming from a leader at the front lines of the world’s most advanced AI development, all the more significant.

Sam Altman, CEO of US rival OpenAI, was quick to respond by expressing his agreement. But if these two companies coordinate their approach, will other AI developers follow suit? Will the US government support their position? And would companies and governments in other countries, such as China, align with this idea as well?

Unfortunately, that seems likely to remain a fantasy.

President Trump was quick to lash out at Amodei over the issue. Apparently, there is no problem because a strong and smart president will keep AI under control. From Trump’s standpoint, as someone who wants to quickly deliver US economic growth, have the public feel its benefits, and thereby strengthen his own position, the message is probably: stop saying things that get in the way.
Leaders at other AI developers have also voiced disagreement with Amodei’s position. Meta CEO Mark Zuckerberg has opposed Amodei’s view, arguing that such matters should be left to the voluntary judgment of individual companies and to market mechanisms. From the standpoint of a company that seems to be “a lap behind” in AI development, perhaps the concern is that if all companies were to coordinate, Meta would lose its chance to catch up.

I do not know whether Amodei’s view is right or wrong, and that is not really the point here.

The real point here is that people’s opinions are shaped by the positions they occupy. The reality is that views are determined by one’s position, not by whether something is true or correct. In President Trump’s case, his view is shaped by the question of how the US can gain an economic advantage over China. In Zuckerberg’s case, his view is shaped by the need to reverse Meta’s disadvantage in AI. They are not assessing AI risk in and of itself. Unless people are looking at the issue from the same vantage point, no substantive discussion can take place, and naturally, reaching a consensus becomes difficult.

For all these reasons, this issue is likely to be caught up in all manner of competing agendas. Reconciling interests, let alone reaching a consensus, will be difficult, and each side will probably go on interpreting the situation and acting as it sees fit from its own position. What lies beyond that, only God knows.

Since the start of the 21st century, world-shaking events have occurred at regular intervals: the terrorist attacks in the US in 2001, the financial crisis that began in 2008, and the global COVID-19 pandemic that began in 2020. The next major event may perhaps be related to AI. I hope Amodei’s prediction proves wrong, and that the proper use of AI leads people toward greater happiness. But conflicts among competing interests are only intensifying around the world, and I cannot shake a sense of unease.

(Subsequently, it appears that President Trump gathered the relevant parties and that a provisional compromise was worked out. Of course, the situation remains highly fluid.)

Hirotaka Shimizu
Chairman and CEO
Kamakura Shinsho, Ltd.

Image material：PIXTA