Two steps back, three steps forward. It is an understatement to say that Elon Musk’s strategy in artificial intelligence lacks clarity. While the entrepreneur signed a shock letter calling for a moratorium on the development of generative AI, THE FT teaches us that he set up his own company in parallel, baptized “X.AI“. Elon Musk has already ordered thousands of sophisticated graphics chips (necessary for training AI) and recruited several experts (from DeepMind in particular) for his new project. Last night, he explained that he was working on a project. ‘IA, nicknamed “TruthGPT”, responsible for “seeking the maximum truth” and competing with other existing software, which, according to him, does not include enough safeguards…
This weathervane behavior is indicative of one thing: Elon Musk has a somewhat wobbly vision of artificial intelligence. He overestimates and underestimates her at the same time. Visibly haunted by the old Terminators, Elon Musk displays an irrational fear of “artificial general intelligence” (AGI in English), this AI chimera that has become conscious and smarter than humans. The prowess of ChatGPT and its congeners is, according to him, the clear sign of the dazzling progress of AI in this grandiose way.
Nothing is less true. The productions of generative AIs are certainly disturbingly relevant and sometimes seem colored with emotions such as sadness, humor or anger. But ChatGPT doesn’t understand what it reads or responds to, and doesn’t consciously choose to give one response more than the other. It guesses (remarkably well) what a human might answer from what millions of humans have written in the past – including all those sentient, bloodthirsty AI-heavy sci-fi works.
“An AI as smart as a dog”
In reality, note with finesse The Verge, the AI makes humans retake the mirror test they faced around the age of two. Do you understand that you do not see another living being, but your reflection in the responses of these artificial intelligences? Passionate reader of science fiction, Elon Musk remains visibly under the influence of the spell. In doing so, he misjudges the immensity of the challenges that AIs would have to overcome to approach this concept of general artificial intelligence. Despite their prowess, today’s AIs regularly make glaring mistakes, have great difficulty understanding the physics of our world, and remain hopelessly single-tasking (most are specialized in text or image.). “Before talking about GOD-like AI (Editor’s note: an AI as intelligent as God), it would already be necessary to pass the DOG-Like AI (Editor’s note: an AI as intelligent as a dog)”, summarizes with humor on Twitter, Yann LeCundirector of Meta’s AI lab and bigwig in the field.
The illusions that Elon Musk harbors about AI have very concrete consequences: the boss of Tesla has not ceased for seven years to promise the imminent arrival of autonomous driving on these vehicles and to push it back. Despite the undeniable success that the Tesla are in the field of electric cars, they are still very far from being able to be managed 100% by driving software backed by AI. By limiting himself to humans as a reference, Elon Musk also underestimates the upheavals that AIs very different from humans can already produce. It is the same bias that pushes him to want at all costs to develop humanoid Tesla robots while non-human forms make it possible to push the physical limits much further (of speed, load displacement, etc.) at a price well lower. It remains to be seen whether his new team of AI masterminds will lead him to broaden his horizons.