Jacques Vallée on artificial intelligence, witness interviews and the limits of classification
Source interview: Watch on Rebelliously Curious — Chrissy Newton’s YouTube channel.
Interview analysis
Jacques Vallée traces his work on early computer systems to questions that remain relevant to using artificial intelligence in UAP research: how to define a question, recognise missing information and avoid a plausible answer to the wrong problem. He proposes tools that assist witness interviews and compare claims across catalogues while retaining human judgement. Newton also asks whether apparent intelligence in machines could help interpret unusual encounter narratives, leading Vallée into a more speculative discussion of language, perception and reported anomalies. Those proposals require different tests from the practical tasks of organising testimony and checking a database.
Key themes
A clear question comes before a useful answer
Vallée’s early astronomy system worked within a bounded vocabulary and a defined catalogue. He recalls spending substantial effort on recognising questions that were ambiguous or inappropriate for the data. That experience supplies his central warning about more general systems: a fluent response can mislead when the question itself has not been made precise.
Assistance without replacing the interview
An automated assistant could remember possible explanations and suggest a question that an investigator might otherwise miss. Vallée reserves the conduct of the interview for a person who can respond to hesitation, silence and the witness’s pace. The proposal concerns better questioning and record keeping; the assistant’s apparent empathy would not establish the accuracy of the account it helps collect.
Turning a recurring claim into a catalogue query
Vallée uses the suggestion that UFOs turn blue when accelerating to illustrate a testable comparison. A catalogue could expose contrary cases and help establish how consistently witnesses reported the proposed association. The result would still depend on the reports’ descriptions, selection and coding, so an apparent pattern would need to be checked against the original accounts before becoming a physical explanation.
Recognition reaches a harder limit with an unknown target
Systems can be tested against familiar objects with known characteristics, whereas a genuinely unfamiliar target lacks that reference. Vallée extends the problem into encounter stories involving unusual speech or apparent changes in space, which he interprets through his broader ideas about the phenomenon. His account of those narratives is a separate evidential task from demonstrating that software can retrieve reports or assist an interviewer.
Assessment
Useful tools need visible questions and traceable sources
Vallée identifies practical tasks that can be evaluated without settling the origin of UAP: retrieving accounts, checking a proposed association and prompting an investigator to consider alternatives. Such tools need traceable source passages, documented categories and tests of how questions affect their answers. His warning about overconfidence applies especially when an output appears to explain an encounter whose basic facts are contested. More ambitious interpretations of language, perception or space require independent evidence for the reported events and a way to distinguish the interpretation from competing accounts.
Navigate the source interview
- 2:00 — Northwestern, astronomy and early computer work
- 5:36 — Natural-language questions about a star catalogue
- 12:30 — Ambiguous questions and answers a system should not give
- 16:46 — UFO questionnaires and adaptive witness interviews
- 19:16 — Vallée’s account of industrial and medical applications
- 23:58 — Risks in automated recognition and decision-making
- 28:38 — Understanding and testing a system’s decisions
- 32:04 — Classifying objects whose properties are unknown
- 35:09 — Why AI and UAP research need not be a new conjunction
- 39:41 — Testing claims about colour and acceleration against a catalogue
- 45:05 — An assistant for the human investigator
- 50:27 — Emotion, apparent understanding and overconfidence
- 56:16 — Encounter narratives, language and interpretations of absurdity
- 67:32 — Practical tools for interviews and coordinated observations
Questions
2:00 What did Vallée learn from developing an early natural-language astronomy system?
16:13 How has he seen AI develop, and where do ambiguous questions cause difficulty?
23:58 What risks arise when a system must recognise unfamiliar objects or support decisions?
39:41 How could catalogues test a recurring claim about reported UFO behaviour?
45:05 Which parts of interviewing a witness should remain with a human investigator?
50:27 Could an apparently responsive system encourage users to trust it too much?
67:32 Which practical applications should researchers develop first?
Selected quotations
Three full stops (...) mark words omitted from the transcript, including verbal filler or false starts. Every retained word follows the source transcript. Use the linked timestamp or full transcript for surrounding context.
it's not going to do the interview it cannot do the interview because so much is in the human connection
Vallée argues for an assistant that prompts and remembers while the investigator attends to the witness’s pauses, reactions and account.
people will overestimate the Mind behind a robot
Vallée warns Newton that fluent and apparently precise responses can invite more confidence than the system’s operation warrants.
Sources
- Source interview on YouTube — Newton’s conversation with Vallée about the history and proposed research uses of artificial intelligence.
Read the full source transcript (plain text). The supplied wording and time markers are retained.