AI’s future at stake: the challenge between Gary Marcus and Miles Brundage

Archive note: this essay was published on 15 January 2025. The wager, its terms and the author’s interpretation are presented as described at that time; this translation does not assess subsequent progress towards the challenge.

We are at the beginning of 2025 and, as often happens at the start of a year, predictions about artificial intelligence and other emerging technologies are proliferating. Yet AI has proved one of the hardest areas to predict, because of its multidisciplinary nature — spanning everything from computing to philosophy — and the dizzying pace of innovation. The debate about the arrival of artificial general intelligence (AGI) ranges from the enthusiasm of those who believe it is just around the corner to the scepticism of those who doubt its feasibility in the short or even the long term.

Into this landscape comes the intriguing wager between researcher and well-known sceptic Gary Marcus and AI policy expert Miles Brundage. Why all the excitement? Because this is not simply “another prediction”: both have put real money on the table, with a charitable component at stake. It is a true example of “skin in the game”, a concept popularised by Nassim Nicholas Taleb that means risking something valuable — reputation, money or anything else — when you make a claim or take a decision.

The mechanics are simple but revealing. If, by the end of 2027, AI meets a series of ambitious criteria that we set out below, Marcus will donate $2,000 to a charity chosen by Brundage. If AI fails the challenge, Brundage will instead pay $20,000 to a cause selected by Marcus. Beyond the amounts involved, this structure of milestones brings seriousness to the debate by focusing on concrete, tangible results.

The contenders: pragmatic scepticism versus grounded optimism¶

Gary Marcus, author of influential books such as Rebooting AI, is an influential sceptic. While he acknowledges advances in areas such as language processing, he insists that today’s AI still lacks the contextual understanding and capacity for generalisation required to achieve AGI. His participation in the wager is not merely a criticism but a challenge to the industry to demonstrate real, measurable progress.

In the other corner is Miles Brundage, a pragmatic optimist. With first-hand experience gained over several years at OpenAI, Brundage has confidence in rapid technological progress. Although he advocates an ambitious roadmap, he also stresses the crucial need for controls and regulation to mitigate potential risks. His wager reflects his conviction that AGI-related milestones are achievable in the coming years.

The terms of the challenge¶

The wager centres on ten specific tasks that, both agree, could serve as clear indicators that AGI has been achieved. Under the agreement, Marcus will lose if AI systems can perform at least eight of these tasks before the end of 2027. The selected tasks include:

1. Complex narrative understanding: watch a film for the first time and summarise it without spoilers, identifying characters, conflicts, motivations and plot twists.
2. Literary analysis: read new novels and answer questions about characters, plots and conflicts, going beyond the literal meaning.
3. Professional writing: produce obituaries or short biographies with the quality and rigour of The New York Times.
4. Adaptation in video games: quickly learn and master the basic mechanics of almost any new video game.
5. Legal writing: write persuasive, error-free legal arguments.
6. Advanced programming: produce working code of more than 10,000 lines from natural-language specifications, without relying solely on existing libraries.
7. Literary and cinematic creation: write Pulitzer-calibre books and Oscar-calibre screenplays with little or no human intervention.
8. Scientific discoveries: propose paradigm-shifting discoveries worthy of a Nobel Prize.
9. Mathematical translation: convert mathematical proofs written in natural language into verifiable symbolic formats.
10. Solving video-game puzzles: solve original puzzles in alternative worlds within video games.

A panel of independent judges will decide whether the tasks have been accomplished, taking account of factors such as reasonable costs and the system’s adaptability. This approach seeks to prevent the results from depending on particular technological advances or an excessively specialised design.

Beyond the wager: a gauge of AI’s future¶

Although Marcus has been a vocal critic of the narrative that AGI is imminent, his participation in the wager reflects a tacit acknowledgement of the technology’s potential. By accepting that AI could perform several of these tasks — though, in his view, fewer than eight — before the end of 2027, Marcus shows a willingness to revise his pronounced scepticism in the light of concrete evidence.

Brundage, for his part, is betting confidently on AI’s accelerating progress, reflecting a position common among those who have worked in leading organisations in the sector and have privileged inside knowledge of how these technologies are developing.

Far from being a mere clash of opinions, the wager between Gary Marcus and Miles Brundage offers a measurable, realistic method for assessing advances in artificial intelligence. Introducing “skin in the game” — and donations to charitable causes — adds credibility to a debate that too often amounts to “predictions without consequences”.

Regardless of who wins, there is one point on which both sides agree: AI continues to scale at a dizzying pace, and its implications for society will be enormous over the coming years. Even sceptical voices such as Marcus acknowledge that we are entering territory where what once seemed impossible is beginning to knock at the door. Would you dare to place a bet?

Fernando Nieto Lobato

Director of Digital Innovation, Institución Educativa ALEPH