The Machine Intelligence Research Institute has published a companion guide for viewers of the documentary “The AI Doc: Or How I Became an Apocaloptimist,” a 104-minute film described as currently in theatres. The post, framed as a Q&A, addresses questions the organisation expects viewers to have after watching the film.
MIRI opens with a statement of its own position: it describes itself as having started as an accelerationist organisation and as having changed its views after investigating AI progress, now believing “there is an urgent need for the international community to come together and push for a ban on frontier AI development.”
On the pace of AI capability improvement, the post cites several data points. It notes that on launch, ChatGPT scored in the bottom 10 percent on the bar exam and that four months after launch it was scoring in the top 10 percent. It notes that AI systems have reached Gold Medal performance at the International Math Olympiad. It describes a graph — not reproduced in the post — showing improvements in AI performance on long-horizon programming tasks over several years. Yann LeCun, described as “a notorious skeptic of AI capabilities,” is cited as having said that reaching human-level AI “will take several years if not a decade,” a position the post presents as evidence that even sceptics now accept near-term timelines.
On the question of whether AI systems can be trusted to behave safely, the post summarises what it describes as a fundamental limitation: modern AI is “grown” by feeding neural networks large amounts of data and using gradient-based algorithms to adjust model weights toward better outputs. The post states that engineers cannot reliably explain why their models behave as they do, citing a 2024 statement from an OpenAI interpretability researcher who said “we don’t really understand how neural networks work.” The post does not name the researcher.
The post addresses recursive self-improvement — a scenario in which AI systems improve themselves without human involvement — and describes it as a threshold that is difficult to predict in advance. It cites the CEO of Anthropic as having described the possibility of “a country of geniuses in a data center” as a near-term scenario, without specifying the source of the quote.
The post presents Sam Altman’s description of OpenAI’s safety approach from the documentary: “You create a new model. You study and test it very carefully. You put it out into the world gradually, and then more and more. You understand if that’s safe or not. And then, if it is, you can take the next step.” MIRI describes this as “fairly typical of how labs are operating” and argues it is insufficient given the pace of capability development.
The post is primarily addressed to a general audience and links to the documentary’s ticketing page. MIRI states two of its staff were interviewed for the film but that it was not involved in its production.