TL;DR
Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
Scott Alexander has published an open letter responding to Steven Pinker’s arguments that AI risk scenarios rely on an overly broad idea of intelligence. Alexander argues that newer GPT models have improved at tasks earlier systems struggled with, and says there is no clear basis for assuming progress will stop before reaching human-level capabilities. The supplied excerpt does not establish that Pinker has replied to this letter.
Scott Alexander has published an open letter answering Steven Pinker’s arguments about artificial intelligence, disputing Pinker’s skepticism that AI systems will continue to gain broad capabilities. Alexander points to improvements between successive GPT models and argues that there is no compelling reason, based on the evidence he discusses, to assume progress will stop just short of human-level performance.
The letter appeared on Astral Codex Ten after Alexander said he had challenged Pinker to a public debate in response to Pinker’s social-media posts. Pinker replied with an open letter of his own, according to Alexander, arguing that in-person debates can feel confrontational and may not support truth-seeking. Alexander accepts that criticism of the format, but says he will answer the arguments Pinker set out in writing.
Alexander’s first substantive response concerns what counts as intelligence. Pinker’s position, as Alexander describes it, is that some AI-risk arguments treat intelligence as a single quantity that can be extrapolated from animals and humans to artificial systems and then to superintelligence. Pinker instead emphasizes that intelligent systems are mechanisms suited to some problems but not others, and are limited by the information they can acquire through observation and experimentation.
Alexander counters that ordinary usage supports describing newer models as more intelligent than earlier ones. He cites a sequence of examples: GPT-2 and GPT-3’s reported difficulty with simple arithmetic or reasoning questions, later systems’ improved performance on such tasks, and an OpenAI model he says was just beyond GPT-6 that solved the Navier-Stokes equations. Those model and performance details are claims in Alexander’s letter; the supplied material does not independently document them. He argues that further improvement is plausible, even if progress eventually levels off.
Why AI Capability Forecasts Differ
The exchange focuses on a central disagreement in public debate about AI: whether improvements in systems’ performance justify expecting continued growth in general capability, or whether their strengths remain bounded by task-specific limits. That distinction affects how readers interpret forecasts about future AI systems and the risks or benefits associated with them.
Alexander’s argument is that observed gains make it hard to identify a particular point at which improvement must stop, especially below human-level performance. Pinker’s counterargument, as represented in the letter, cautions against treating intelligence as a single scale that applies cleanly across biological and artificial systems. Neither position is settled by the excerpt: the disagreement is about how to interpret evidence and make predictions, not a report of a new independently verified capability milestone.
The exchange also illustrates how a dispute framed initially as a proposed public debate has moved into written argument. Readers can assess the claims and examples directly, while the format gives each writer room to explain a position without relying on a live confrontation.
As an affiliate, we earn on qualifying purchases.
From Debate Offer to Open Letters
Alexander says he issued the debate challenge after becoming frustrated with Pinker’s social-media posts about AI. He later apologized for the attention generated by the challenge, describing an influx of offers from podcasters, debating societies and other hosts. The humorous account of that response is Alexander’s own characterization; it is not an independently verified account of the organizers’ actions.
Pinker then wrote an open letter to Alexander, according to the source, apologizing for any offense caused by his posts while denying Alexander’s accusation of personal attacks. Pinker also set out arguments against what he sees as weak assumptions in AI-risk scenarios. Alexander’s new letter takes up those arguments, beginning with intelligence and the interpretation of capability gains. It also refers to earlier exchanges involving AI researcher Scott Aaronson and cognitive scientist Gary Marcus, but the excerpt does not reproduce the full history of those discussions.
““I argued that any intelligent system is a mechanism which is good at solving some problems but not others.””
— Steven Pinker, as quoted and summarized by Scott Alexander
Evidence Behind the Capability Claims
The excerpt does not include Pinker’s full letter, so his position is available here through Alexander’s account and selected quotations. It also does not provide technical reports, evaluations or other independent evidence for the claims about GPT generations, the Navier-Stokes result, or the systems’ comparative abilities.
It is not clear from the supplied material whether Pinker has responded to Alexander’s latest letter, whether the two will continue their exchange, or whether a debate will take place. The source also gives no exact publication date, beyond Alexander’s description of the debate challenge as occurring earlier in the month.
Whether the Written Exchange Continues
The immediate next development would be a response from Pinker or a further installment from Alexander addressing the other arguments in Pinker’s letter. Alexander says he plans to proceed through Pinker’s arguments; the supplied excerpt develops the first one, concerning intelligence and the trajectory of AI improvement.
No debate date, host or agreement is reported, and the excerpt does not say that Pinker accepted the original challenge. Readers looking to evaluate the technical claims will need evidence beyond this exchange, including clearly specified model versions, task conditions and independent assessments of performance.
Key Questions
What is the news?
Scott Alexander published an open letter responding to Steven Pinker’s arguments about AI intelligence and the possibility of continued improvement in AI systems.
What is Alexander’s main argument?
Alexander argues that newer GPT models have improved on tasks earlier systems struggled with, and that the evidence cited does not show where future improvement must stop. He allows that progress may eventually level off.
What is Pinker’s position, according to the letter?
Alexander describes Pinker as arguing that AI systems are suited to some problems but not others, and that their capabilities are limited by what they can learn through observation and experimentation. The excerpt does not reproduce Pinker’s full letter.
Has Pinker agreed to a public debate?
The supplied source does not say that he has. It reports Alexander’s debate challenge and Pinker’s written response, but provides no agreement, host or date for a debate.
Source: rss
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
