What is the AI futures debate (AI 2027 and others)?

Between 2024 and 2026, a string of forecasts and scenarios about when AI smarter than humans will arrive, and what the world will look like then, set off a major debate. People on the Plurality side responded by asking, rather than where one AI is headed, how society prepares and who holds the steering wheel.

Follow the news about AI and you hear two stories: that AI will surpass humans within a few years, and that things will not change so fast. Between 2024 and 2026, several pieces of writing laid out this question as concrete timelines and storylines, and they were argued over around the world. This article introduces the main documents and how people on the Plurality side responded.

In short

Between 2024 and 2026, essays and scenarios describing when AI smarter than humans will arrive, and what the world will be like when it does, came out one after another and set off a major debate. The main documents are introduced under "In more detail" below.

People on the Plurality side, such as Audrey Tang, E. Glen Weyl and Vitalik Buterin, respond in a distinctive way. They ask not only what will happen with the single most advanced AI, but how society prepares, and who holds the steering wheel [3][11][12].

For example

AI 2027 (April 2025). A scenario (a storyline) published by the AI Futures Project, led by Daniel Kokotajlo and others. A fictional US company automates programming in 2027, and AI accelerates AI research [1].

It has two endings. In the "race" ending, AI that goes on being developed takes power and eliminates humanity in 2030. In the "slowdown" ending, dangerous models are halted and the United States and China reach an agreement, though the problem remains that power concentrates in the oversight committee [1]. The authors later explained that 2027 was the single year they thought most likely, not their median forecast, and they continue to revise their forecasts [2].

Buterin's response (July 2025). Buterin criticized an assumption implicit in AI 2027: that only the leading AI gains godlike economic and destructive power, while everyone else's power stays about the same. If the same scenario pictures cancer and aging being conquered by 2029, he pointed out, defenders should gain superhuman powers too [3].

He then wrote that a strategy of letting one AI take control and win the race could actually reduce safety, because it removes any means of pushing back if that dominant AI goes wrong [3].

Tang's response (2025). In a July 2025 interview with Nikkei Asia, Tang offered a counter-image to views like AI 2027. "I think that is the bet of cooperative AI. It means that these systems, such as Community Notes, that foster human coordination, can get exponentially stronger once AI capacity becomes exponentially stronger. And it's not just, as you say, the kind of dominating view of AI 2027, but rather, something like we the people with the AI now become superintelligence together" [4]. (Community Notes is a system in which users add context to social media posts.)

In more detail

The main documents in the debate

Situational Awareness (June 2024) — A piece by Leopold Aschenbrenner. It says the race for artificial general intelligence (AGI) has begun and that true superintelligence will arrive by the end of the 2020s. It predicted a project led by the US government and all-out competition with China [5].
Machines of Loving Grace (October 2024) — An essay by Dario Amodei, chief executive of the AI company Anthropic. It describes the upside if powerful AI goes well, in five areas: biology, neuroscience, the economy, peace and governance, and work and meaning. It says most people underestimate that upside [6].
AI as Normal Technology (April 2025) — An essay by Arvind Narayanan and Sayash Kapoor. It argues that AI should be seen not as a "separate species" or a candidate superintelligence, but as a normal technology (though one that still transforms society), like electricity or the internet. Its position is that adoption will be slow, and that policies prioritizing control and resilience are needed [7]. In November 2025, the authors of AI 2027 and of this essay also published a joint document listing points on which they agree [8].
Gradual Disempowerment (January 2025) — A paper by Jan Kulveit and others. It argues that even without a sudden takeover, as economies and states come to depend less on human labor and thought, human influence, such as the choices made in voting and shopping, could be lost gradually and irreversibly [9].
AI 2040: Plan A (July 2026) — A scenario by the same team as AI 2027, written as a recommendation of what should be done rather than a forecast. In its storyline, the development of superintelligence is delayed until 2040, all AI research is made public, and many companies are allowed to catch up with the frontier [10].

Response 1 from the Plurality side: a balance of power, not one AI

Buterin says that whether the leading AI can easily wipe out everything else is not a fixed assumption. It is a dial that can be adjusted by spreading defensive technologies [3]. This connects to Buterin's d/acc (the idea of prioritizing technologies that help defense and decentralization).

Response 2: the frontier AI problem and the local problem are different

The FAQ (frequently asked questions) by Tang and Caroline Green separates two questions. The first is aligning the values of frontier general AI with humanity. The second is: in this setting, to whom is this AI accountable, and who acknowledges that? The AI ethics framework 6-Pack of Care is for the second question, it writes [11].

The FAQ describes AI 2040: Plan A as a convergence arriving from the risk side, since in Plan A the good ending runs through "a handoff to an ecosystem, not to a single corporation-within-a-corporation". But, it continues: "What that scenario leaves open is the second question: its public receives a Citizen's Dividend and ratifies bargains negotiated by AIs, but the rooms where people live never become authors — that half is our work" [11].

Response 3: society's preparation is left out

In a July 2026 essay in the magazine Noema, Weyl and co-authors, while agreeing with some of Amodei's policy proposals, pointed to an imbalance in the debate. There are concrete, ambitious plans for building smarter AI and for aligning AI with human values, they write, but hardly anything on preparing society to receive it [12]. This is the starting point of reverse alignment (the idea of rebuilding society's institutions, rather than AI).

Response 4: the people set the pace of slowing down

According to an essay by Hélène Landemore and Tang, on September 12, 2026 Amodei called for slowing the pace of frontier development, and OpenAI's Sam Altman and others agreed [13]. In response, Landemore and Tang argued in Noema that the US–China competition is closer to a relationship in which cooperation is possible if there are mutual commitments. As a witness to watch over those commitments, they proposed a standing "Global Citizens' Assembly on AI" of 1,000 people chosen by lottery from around the world [13].

Response 5: sympathy with the danger of gradually losing power

Voices on the Plurality side share the concern of Gradual Disempowerment. In an October 2025 talk, Tang said: "The addictive intelligence future is that of gradual disempowerment." She contrasted this with assistive AI that "makes it very easy for communities to steer their own local norms" [14].

What is debated

Does the Plurality side take AI's dangers too lightly? On the debate show Doom Debates (January 2026), host Liron Shapira pressed Tang on AI suddenly leaping ahead by improving itself, and asked for her rebuttal. Tang answered: "My P(Doom) is NaN, or Not A Number", because for AI there is too little to go on, so we literally don't know. She then argued that places that have trained their citizens' capacities through smaller outbreaks are stronger in larger ones [15]. In a May 2025 interview, Nicholas Thompson also asked Tang whether, if something like AI 2027 happens, patient work among citizens would be "like fixing bicycles to go on a road where there are Lamborghinis" [16].
Would giving everyone an AI representative solve it? In June 2025, Kulveit argued that even if individuals have AI agents loyal to them, gradual disempowerment is not solved, because companies and states have agents too. This piece does not, however, name the Plurality side [17].
Criticism of the forecasts themselves. AI 2027 has also drawn detailed criticism of the assumptions of its timeline models (by the blogger titotal, June 2025) and criticism doubting its realism (by Gary Marcus, May 2025) [18][19]. The authors continue to revise their forecasts and to grade their own 2025 predictions [2][20]. Most responses from the Plurality side focus less on whether the timelines are right than on who holds the steering wheel. Buterin also states plainly that he does not enter the timeline debate [3].

Where it fits in Plurality

The text of Plurality was written before these forecasts, so it does not discuss them by name. Chapter 2-0 of the book groups visions of the future centered on AI, to which society would have to adapt, as "Technocracy" (the idea of a small number of experts managing society through technology) [21]. This debate can be read as an extension of that. For more, see the articles on the three paths and on superintelligence.

The Plurality side mainly brings three ideas to this debate: Buterin's d/acc (decentralized defense); Tang and Green's 6-Pack of Care and Civic AI (accountability in each local setting); and Weyl and co-authors' reverse alignment (rebuilding society's institutions).

Sources

AI 2027 Summary (2025-04-03) — https://ai-2027.com/summary
AI Futures Project, "Clarifying how our AI timelines forecasts…" (2026-01-27) — https://blog.aifutures.org/p/clarifying-how-our-ai-timelines-forecasts
Vitalik Buterin, "My response to AI 2027" (2025-07-10) — https://vitalik.eth.limo/general/2025/07/10/2027.html
Audrey Tang, Interview with Nikkei Asia (2025-07-22) — https://archive.tw/2025-07-22-interview-with-nikkei-asia ［CC0］
Leopold Aschenbrenner, "Situational Awareness" (2024-06) — https://situational-awareness.ai/
Dario Amodei, "Machines of Loving Grace" (2024-10) — https://www.darioamodei.com/essay/machines-of-loving-grace
Arvind Narayanan and Sayash Kapoor, "AI as Normal Technology" (2025-04-15) — https://knightcolumbia.org/content/ai-as-normal-technology
"Common Ground between AI 2027 and AI as Normal Technology" (2025-11-12) — https://asteriskmag.substack.com/p/common-ground-between-ai-2027-and
Jan Kulveit and others, "Gradual Disempowerment" (2025-01-28) — https://arxiv.org/abs/2501.16946
AI Futures Project, "AI 2040: Plan A" (2026-07-09) — https://ai-2040.com/
civic.ai FAQ (Q17) — https://civic.ai/faq/ ［CC0］
Glen Weyl, James Evans and Chris White, "What Humanity Needs To Flourish" (Noema, 2026-07-21) — https://www.noemamag.com/what-humanity-needs-to-flourish-in-the-next-decade/
Hélène Landemore and Audrey Tang, "Let The People Set The Pace Of Frontier AI" (Noema, 2026-09-23) — https://www.noemamag.com/let-the-people-decide-the-pace-of-frontier-ai
Audrey Tang, SEWF, "Reshaping Tomorrow" (2025-10-30) — https://archive.tw/2025-10-30-sewf-reshaping-tomorrow ［CC0］
Doom Debates (with Audrey Tang, 2026-01-20) — https://civic.ai/doom-debate/ ［CC0］
Audrey Tang, Interview with Nicholas Thompson (2025-05-21) — https://archive.tw/2025-05-21-interview-with-nicholas-thompson ［CC0］
Jan Kulveit, "Individual AI representatives don't solve Gradual Disempowerment" (2025-06-04) — https://www.lesswrong.com/posts/E96XcEPECbsipAvFi/individual-ai-representatives-don-t-solve-gradual
titotal, "A deep critique of AI 2027's bad timeline models" (LessWrong, 2025-06-19) — https://www.lesswrong.com/posts/PAYfmG2aRbdb74mEp/a-deep-critique-of-ai-2027-s-bad-timeline-models
Gary Marcus, "The 'AI 2027' Scenario: How realistic is it?" (2025-05-22) — https://garymarcus.substack.com/p/the-ai-2027-scenario-how-realistic
Eli Lifland, "Grading AI 2027's 2025 Predictions" (AI Futures Project, 2026-02-12) — https://blog.aifutures.org/p/grading-ai-2027s-2025-predictions
Plurality, 2-0 "Information Technology and Democracy: A Widening Gulf" — https://www.plurality.net/v/chapters/2-0/eng/ ［CC0］

We welcome error reports and questions ("What is…?"). Submissions become public issues.

Report an error in this article ・ Ask a question