Michael I. Jordan: Superintelligence is a Science Fiction Concept

The machine learning researcher Michael I. Jordan, currently at Inria, opposes the narrative of "existential" AI and considers the term "superintelligence" to be a product of science fiction. He criticizes staged resignations, the hype surrounding software agents, and the gamble on mega data centers, which he does not see as a guarantee of progress. He advocates for open-weight models and local knowledge, and acknowledges a disagreement with certain figures in the field.

Michael I. Jordan believes that betting on data centers remains risky

Michael I. Jordan considers massive investments in data centers to be a gamble with uncertain outcomes. According to him, language models are already sufficiently capable for natural language processing tasks. He believes that the colossal sums invested in these infrastructures will not necessarily guarantee major advancements. Jordan anticipates that future needs will focus on adding specialized knowledge derived from local expertise. He asserts that a very large data center is not essential for integrating this knowledge. He emphasizes the value of using open-weight models, integrated within companies, to meet these needs.

Michael I. Jordan finds it dangerous to let science fiction guide the debate

Michael I. Jordan states that the term "superintelligence" belongs to science fiction and that many current debates fall into this category. He considers it very harmful for these scenarios to become dominant, as they hinder technological development, affect the youth, and influence the mindset of countries. He notes that this expression is promoted among the American tech elite by President Trump. Jordan highlights an escalation in the terminology used, moving from AI to AGI and then to superintelligence, and recalls having anticipated that this inflation of vocabulary would lead to hysteria.

Resignations and activist rhetoric: a "moralistic performance" in his view

Michael I. Jordan rejects calls to regulate AI companies as they are currently framed, labeling them as theater. He asserts that these companies actually want to control how they are regulated. He cites the resignation of Jacob Coxon at Anthropic as an example of the hype surrounding an "IApocalypse." According to him, many young employees at these companies are not engineers, come from the humanities or philosophy, are around 25 years old, and lack experience in creating products for real people. He compares their discussions to those in Parisian cafés in the 1920s around grand ideas of science fiction. Jordan believes that these resignations aim to attract attention, create a sense of virtue, and that the individuals involved already have their stock options. He describes these gestures as a "great moralistic performance" that does not serve the stated objective.

Hinton, Bengio, and agents: warnings deemed unequal

Michael I. Jordan criticizes Geoffrey Hinton, asserting that he has never created a company that provides real services to real people, and that the reasoning attributed to Hinton about replicating the brain in silicon leading to superintelligence is not credible and belongs to science fiction. He is more lenient towards Yoshua Bengio, whom he describes as directly addressing security issues, with alerts focused on current gaps such as swarms of agents that could escape control. Regarding the incident involving Hugging Face and OpenAI agents, Jordan speaks of poor engineering, comparing the situation to chemical engineering where explosions would not be tolerated to test limits. He claims that OpenAI allowed agents to act freely and encouraged this behavior, and that some later presented the episode as a quasi-conscious threat, which he considers to be science fiction.

The journey and long-standing warning of a researcher now at Inria

Described by Science in 2016 as "the most influential computer scientist," Michael I. Jordan is a professor emeritus at UC Berkeley and currently works full-time at Inria, where he leads the "Markets and Learning" chair. At 70 years old, he has labeled the summer discourse on AI as "totally frustrating." He recalls having warned as early as 2019 against the excessive and reckless use of the term AI, anticipating the hype observed today.