Enrique Dans: in the AI 'wargame', the only winning move is not to play

Important notice: the content downloaded from Medium is a teaser for a 'member-only' (paywalled) article. Medium only delivers the opening of the text for free —labeled as "4 min read" but cut off after barely a couple of paragraphs— and the last sentence is left literally unfinished ("Is this enough to justify the paranoid …"). What follows, therefore, is a very brief summary, strictly limited to what can actually be read, without filling in assumptions about how the argument continues or concludes.

Enrique Dans titles his piece by playing on the film 'WarGames' (1983), whose classic conclusion is that "the only winning move is not to play". According to the available excerpt, a recent editorial in The Guardian used that same reference to talk about the current race for artificial intelligence.

The article's opening describes a contrast: the top executives of the big tech companies talk about the need for a pause in AI development that they nonetheless have no intention of applying. At the same time, China warns about "hostile forces" and Donald Trump claims to have the situation under control. Dans notes that these same leaders fear above all "regulatory capture" and consider themselves the best qualified to oversee the technology's development, while other players speak instead of matching capabilities and of safety. In parallel, all of them keep working nonstop on "improving" their models.

The text points to Anthropic and OpenAI as the main protagonists of this debate. Google, according to the article, has chosen to stay on the sidelines, while Microsoft's position boils down to a call for "common sense".

The excerpt closes by raising the big question without answering it within the accessible portion: are we heading, as some maintain, toward an existential disaster? Could AI really wipe out humanity? Dans mentions that these systems' capacity to cause harm is known when they are instructed to do so or when precautions are neglected, that those capabilities scale as the models become more powerful, and that recursive self-improvement abilities exist, in addition to the obvious risk of malicious actors using the technology for harmful ends. The text breaks off just as Dans is about to assess whether all of that justifies a "paranoid" stance toward the risk, with neither the rest of his reasoning nor his conclusion available in the downloaded content.