AGI: Judea Pearl Opposes Causality to Performance

AI leaders announce that artificial general intelligence (AGI) has already been achieved or is imminent. Judea Pearl, a computer scientist, offers a different perspective: in his view, the feats of deep learning do not constitute intelligence without an understanding of causal relationships.
Brockman and Huang declare AGI achieved
Greg Brockman, co-founder and president of OpenAI, and Jensen Huang, former CEO of Nvidia, have claimed that the arrival of recent models marks access to AGI. These statements are part of a series of positions taken by leaders presenting this threshold as crossed or near.
Former U.S. President Donald Trump has also mentioned "superintelligence," suggesting that the term "artificial intelligence" has become redundant. However, artificial superintelligence, or ASI, already corresponds to a defined concept in science, and no basis has been provided to assert that this milestone has been reached.
Pearl makes causality the missing criterion
For Judea Pearl, current systems remain far from AGI: their ability to imitate does not replace an understanding of causes and their effects. He emphasizes that these systems rely on statistical models to generate results, lacking true causal understanding. This distinction, he argues, separates spectacular capabilities from authentic intelligence.
Le brief IA que les pros lisent chaque soir
Les 7 actus IA du jour, décryptées en 5 min. Gratuit.
Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.
Choisissez votre rythme
Gratuit · Pas de spam · Désabonnement en 1 clic
A pioneer in AI, Pearl contributed to the probabilistic approach that underpins deep learning, earning him the Turing Award in 2011. In The Book of Why, which focuses on the role of causality in scientific transformations, he confronts the initial vision of intelligent machines with contemporary systems right from the first chapter.
Transformers remain the foundation of language models
Image recognition, multilingual translation, and real-time conversations are among the advancements made possible by deep learning. Modern language models are based on the transformer architecture, stemming from foundational work by Google scientists. The Book of Why was published a year after this advancement.
The upcoming GPT-6 Astra and Gemini 4 families from Google are announced as forthcoming models, but they would continue to rely on the same frameworks and architectures as existing language models.
Brief IA — L'actualité IA en français
L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.