The correct model of AI CEO behavior

I concur with Alex that the usual model of regulatory capture does not fit this situation, but I believe he overlooks the most appropriate model available. In a recent review of Kevin Roose's upcoming book, The AGI Chronicles: The Inside Story of the Race to Create an Artificial Superintelligence, I found a central theme: the intense rivalry and competition between Dario, Altman, and Musk.

Each individual is convinced that their rivals cannot be trusted to acquire super-powerful AI capabilities. Consequently, their actions, when observed, seem not to be driven solely by financial gain (as they already possess ample resources). Instead, their motivations appear to stem from a genuine belief that their respective companies would serve the world best by winning the race to develop highly advanced AI.

When these CEOs claim altruism, it is crucial to take their statements seriously, even though their actions may also be motivated by selfish desires that align well with their objectives. While opinions may vary on who should emerge victorious in this race, I favor the outcome where Meta and Google ultimately tie for first place, maintaining a highly competitive market.

Although these CEOs may appear selfish in a certain sense, their perception (whether accurate or not) is that "selfish and altruistic" notions are closely correlated when it comes to the concept of power. By "power," I mean the general ability to achieve goals, rather than implying any intention to engage in coercive or malevolent actions.

Furthermore, the primary objective is not merely to "capture the regulator," but rather to gain a significant competitive advantage and influence outcomes before others do. Therefore, while these CEOs express sincere concerns about various safety issues, they are equally concerned about not being the ones to achieve the race's ultimate success.

In essence, when citing these CEOs as authorities, the primary takeaway should be that "who gets it is the most important thing of all."

An important consideration to note is that despite the frequent mention of a "pause" in the development of AI, the competition to invest in computational resources continues to escalate and intensify. This ongoing race for computational power should be the key variable to monitor. It is important to recognize that a theory explaining CEO behavior centered around compute investment, which requires substantial financial expenditure, may be inadequate if it fails to account for this critical aspect.

While I do not necessarily view these actors as evil or believe they can operate with little constraint, the competitive landscape and their inflated egos certainly warrant a more nuanced perspective, akin to Adam Smith's observations on standard competition. The article "The correct model of AI CEO behavior" originally appeared on Marginal REVOLUTION.