Editorial illustration for Harvard Psychologist Urges Pragmatic AI Safety Over Apocalyptic Rhetoric
Pinker Rejects AI Doomsday Talk, Urges Pragmatism
Harvard Psychologist Urges Pragmatic AI Safety Over Apocalyptic Rhetoric
Steven Pinker has picked a fight with the paperclip maximizer. In an open letter published on Quillette, the Harvard psychologist addresses Scott Alexander, the psychiatrist and blogger who challenged him to a public debate on AI existential risk. Pinker turned it down flat, calling the format a "spectator sport" rather than a way to settle anything.
The letter is his answer instead. Pinker argues that the doomsday scenarios driving much of AI safety discourse, an AI single-mindedly converting the planet into paperclips, say, rest on a category error: they treat intelligence as if it came bundled with a drive to dominate, and assume any advanced system would lock onto one goal with maniacal focus. He's had to walk back some of his own past optimism, admitting he didn't foresee how capable large language models would get, or how fast companies would ship them without much caution.
Where Pinker and Alexander actually part ways, and what each thinks should replace apocalyptic warnings with, is the real substance of their disagreement. Alexander laid out his own numbers and remedies on his blog back in June.
The dangers Pinker takes seriously are bioterrorism, cyberattacks, and unchecked AI agents. Doomsday rhetoric won't do anything about them, he writes. What works is sober safety engineering built on independent oversight, liability, and human control.
Why this matters
Pinker's admission cuts both ways, and that's the part worth sitting with. He was wrong about how capable LLMs would get, wrong about how carelessly companies would ship them, and that track record should make anyone cautious about dismissing existential-risk arguments outright. But he's also right that "the AI will kill us all" rhetoric has become its own kind of spectator sport, one that generates headlines and Twitter threads more reliably than it generates safety engineering.
For developers and founders actually building these systems, the useful signal here isn't Pinker versus Scott Alexander, it's the observation that the loudest doom-callers often work at, or came from, the labs pushing releases fastest. That contradiction deserves more scrutiny than either side's rhetoric. If you're shipping AI products, the practical takeaway is to treat safety as an engineering discipline with concrete benchmarks, not a debate stage.
Watch for whether Alexander responds in kind, and whether that response stays technical or drifts back into apocalypse framing.
Related Reading
Policy & Regulation Akamai Proposes SAFE Cybersecurity Guidelines for AI Threats
Policy & Regulation After ditching AI fitness apps and a Fitbit, I return to Peloton classes
Policy & Regulation Sony Music, Warner Chappell Sue Anthropic Over "Blatant Theft
From Archives Sony And UMG Sue Suno Over 'Model Laundering' Claims
From Archives BaseCamp Framework Uses Six AI Agents to Automate DNA Sequencing Pipelines
Common Questions Answered
What is Steven Pinker's main criticism of AI existential risk discourse?
Pinker argues that doomsday scenarios and apocalyptic rhetoric about AI, such as the paperclip maximizer thought experiment, dominate AI safety discussions but fail to address real dangers or produce meaningful safety improvements. He contends that this type of catastrophic framing has become more of a 'spectator sport' that generates headlines rather than practical solutions to AI safety challenges.
What specific AI dangers does Pinker take seriously according to the article?
Pinker identifies bioterrorism, cyberattacks, and unchecked AI agents as the genuine threats that deserve serious attention. Rather than focusing on speculative existential scenarios, he believes safety efforts should concentrate on these concrete, near-term risks that are more likely to materialize.
What approach does Pinker advocate for instead of doomsday rhetoric?
Pinker advocates for pragmatic AI safety engineering built on independent oversight, liability frameworks, and maintaining human control over AI systems. This sober, engineering-focused approach contrasts sharply with the apocalyptic predictions that dominate much of current AI safety discourse.
How does the article acknowledge Pinker's past misjudgments on AI capabilities?
The article notes that Pinker was wrong about how capable large language models would become and wrong about how carelessly companies would deploy them to users. This track record of underestimating AI progress is presented as a reason to remain cautious about dismissing existential-risk arguments entirely, even while criticizing their rhetorical approach.
Why did Pinker refuse Scott Alexander's debate challenge on AI existential risk?
Pinker declined the public debate format because he viewed it as a 'spectator sport' that would not effectively settle substantive disagreements about AI safety. Instead, he chose to respond with a detailed open letter on Quillette that outlines his pragmatic approach to AI safety concerns.
Further Reading
- Why AI safety requires more than industry self-regulation - Brookings
- The Fight to Rein In AI Is Dividing Washington - TIME
- AI safety governance frameworks and regulation roundup - arXiv
- The Most Important Words in the Battle Over AI - Politico
- Anthropic and OpenAI sound alarm on AI safety - Los Angeles Times
