HackAtari: the test that reveals our AIs understand nothing

They dominated video games. Posted superhuman scores. And then, one day, the games were made easier. The result? They collapsed. Why? Because they had never understood what they were doing. And that is where everything changes. In a study as brilliant as it is unsettling, Quentin Delfosse and his team lift the veil on an illusion of intelligence: that of systems that excel… as long as nothing changes. They devised HackAtari, a clever test based on simplified versions of Atari games. A test that, instead of making tasks harder, makes them simpler, and yet it reveals a gaping flaw. Because when you remove the obstacles, the AIs stumble. Where a human adapts and understands, the machine collapses. What does HackAtari tell us? That an AI can shine in the exam without ever grasping its meaning. That it can repeat, optimise, correlate… without ever reasoning. And what if our AIs were, in reality, very good pupils… who recite without understanding? This is not a test of performance, but a test of truth. A test that measures not what an AI does, but what it understands. And which raises, implicitly, this unsettling question: do our AIs really understand what they are doing? Find the full transcript of this episode, along with all the sources and plenty more, at: HackAtari: the test that reveals our AIs understand nothing