MIT Technology Review’s five-part case for skepticism about AI

MIT Technology Review’s five-part case for skepticism about AI

MIT Technology Review published this piece on July 22, 2025. Its author offers five ideas about AI: generative AI is improving across media, but it also makes things up; widespread use raises energy questions; and how large language models work remains poorly understood. The author argues that these systems’ ability to produce outputs that often match reality makes their fabrications especially important to understand—not a flaw that a future version can simply be expected to lose. The piece also challenges talk of artificial general intelligence: without agreement on which humans, tasks, or measures count, the term can amount to “AI, but better than what we have today.” The author’s conclusion is that machines can mimic some human behavior while still having serious flaws, and that imagining a humanlike mind behind them encourages exaggerated assumptions and feeds culture wars between techno-optimists and techno-skeptics. The author says it is right to be amazed and skeptical, and that “it’s all up for grabs.” The piece notes that major companies have shared little information about AI’s energy use; it does not establish how much energy the technology uses overall.

Publication: 6f4a283f-581d-1047-3eb4-a2ab20d7cb02

attribution — pass: The reading names MIT Technology Review as publisher and attributes each argument to the piece's unnamed author. The captured text does not name the author, so this restraint is appropriate. The piece is a single first-person newsletter essay based on a SXSW London talk, not a listing. The related-story links at the end are page furniture.

This story originally appeared in The Algorithm, our weekly newsletter on AI.

summary — uncertain: The title calls this a 'case for skepticism', which flattens the author's two-sided position. The piece opens with 'so good it's scary' and 'do not underestimate it', and it closes by saying it is right to be both amazed and skeptical. The summary also omits the author's sharpest AGI claim, that there is 'zero evidence' progress will continue. Smallest repairs: retitle as the author's five ideas about AI, keep 'so good it's scary' and add the zero-evidence line. Terms are explained adequately, including via the definitions provided.

There’s this absolute faith in the progress of AI. It’s gotten better in the past, so it will continue to get better. But there is zero evidence that this will actually play out.

date — pass: The reading's July 22, 2025 matches the UTC day of originalPublished (2025-07-22T09:50:27Z). This is a publisher page, so no archive disclosure is needed. The sidebar item about the tenth anniversary of AlphaGo is a likely later addition, but the reading does not use it. The page shows a modification date of 2025-07-28; the claims used read as part of the original essay.

Last month I gave a talk at SXSW London called “Five things you need to know about AI”

support — uncertain: Most claims match their quoted excerpts. However, the summary says outputs matching reality make 'their fabrications especially important to understand'. The source says instead that the remarkable thing is how often the made-up output matches reality, and draws two lessons: know what the technology can and can't do, and don't wait for a version that never hallucinates. The 'especially important' framing is an inference the source does not state. Claim 5's 'a call for' AI better than current systems also slightly reframes the source's 'just mean AI, but better'.

What’s remarkable is not that they make up nonsense, but that the nonsense they make up so often matches reality.

media — pass: No media was selected and no verified candidates were supplied, so there is no media identity or relevance issue to check.