Project Alpha

When a well funded new psi laboratory opened at a university in the late 1970s, James Randi wrote to its director offering, at no charge, to advise on protocol design against conjuring methods. The offer was not taken up. Randi then arranged for two young amateur magicians of his acquaintance, Steve Shaw and Michael Edwards, to apply as subjects, having coached them on how to behave: never claim too much, allow failures, do not perform when the controls are tight, and above all volunteer nothing. They were accepted, tested over an extended period, and reported on as promising. They bent metal, made images appear on sealed film, and produced apparent effects on equipment in sealed containers, all by ordinary means.

Randi sent the laboratory a list of the specific vulnerabilities in its procedures that a conjuror would exploit, without naming his participants. When the two were finally tested under conditions that closed those routes, the effects stopped. Randi disclosed the whole arrangement in 1983. The coaching is the part worth attending to. Randi did not teach them an elaborate act; he taught them restraint. A performer who succeeds every time invites suspicion, and one who volunteers explanations invites testing, so the instruction was to fail often, to say nothing about method, and to decline politely whenever the controls tightened. That is a description of how a good deception survives scrutiny, and it applies well beyond conjuring.

The methodological point was accepted widely and quickly, including by parapsychologists who disliked Randi personally. It is now unusual for a serious test of a claimed individual ability to be designed without input from someone trained in conjuring, and where that expertise is absent, reviewers ask why. In that narrow sense the demonstration worked better than any number of articles making the same argument had. The objections are real and were pressed hardest by people sympathetic to the argument itself.

Researchers were used as unconsenting subjects in someone else's experiment, their reputations damaged to prove a general point about a field rather than a specific failing of their own. The disclosure was made to the press rather than to the laboratory first. And the demonstration establishes only that a laboratory can be fooled by someone actively trying, which nobody seriously denied; it does not tell you whether any particular published result was produced that way. Randi's answer, that gentler methods had been tried and ignored, is on the record, and readers can weigh it. This atlas records the episode as a demonstration with a disputed ethics, not as an experiment with a result.