Your coach tells you to play h3 in this structure. You don't understand why. You play it.
A grandmaster annotates a game and dismisses your idea in one sentence. You don't understand why. You drop it.
The engine shows -0.4 in a position you thought was fine. You don't understand why. You conclude your position was bad.
In all three cases, you changed your judgment without receiving a single argument. And in all three cases you were probably right to do so, which is exactly what makes the problem hard.
What Milgram Actually Did
The experiment is so famous that nobody bothers to describe it properly anymore.
In 1961, at Yale, Stanley Milgram recruited volunteers for what he presented as a study on learning and memory. The participant was assigned the role of "teacher." Another man, actually a confederate, played the "learner" and was placed in the next room.
The teacher had to deliver an electric shock for every wrong answer, moving up one notch each time. The scale ran from 15 to 450 volts, with increasingly alarming labels. The shocks were fake and the learner's protests were pre-recorded.
When the participant hesitated, the experimenter in the grey coat had four prompts, to be used in a fixed order: "please continue," "the experiment requires that you continue," "it is absolutely essential that you continue," "you have no other choice, you must continue."
In the study published in 1963, 65% of the 40 participants went all the way to 450 volts.
The result entered history, and it is almost always summarized the same way: ordinary people torture a stranger because a man in a lab coat asked them to.
What the Archives Revealed
Now for the part that almost always goes missing, and it's serious.
In 2013, Australian psychologist Gina Perry published Behind the Shock Machine, the result of several years in Milgram's archives at Yale and interviews with original participants. Two main findings, both uncomfortable.
Many participants didn't believe the setup. By tracking down one of Milgram's own analysts, Perry established that a substantial share of subjects suspected the shocks were fake. A participant who doesn't believe isn't demonstrating obedience: he's playing along. Milgram had this information and chose not to foreground it.
The experimenter didn't follow the protocol. Analysis of the audio tapes shows that John Williams, who played the role, improvised frequently. He repeated prompts, delivered them out of order, invented new ones, and applied pressure well beyond what the published paper described. The pressure was therefore neither standardized nor comparable from one participant to the next.
This second point is the more destructive of the two, because it attacks the independent variable itself. If the pressure varies and isn't measured, we don't know what the 65% figure is measuring.
The Reinterpretation
A third critique, more theoretical, comes from Alexander Haslam and Stephen Reicher. On their reading, participants didn't switch off their moral sense in order to obey blindly. They signed up to a project.
The experimenter never said "do it because I say so." He said "the experiment requires that you continue," which means: this serves science, and you are contributing to it. Haslam and Reicher call this engaged followership rather than obedience: people agree to cause harm when they identify with the cause of the person directing them and believe their action is justified.
That reading is itself disputed, notably by Hollander and Turowetz in 2018, working from the same archival recordings.
What survives. The partial replication run by Jerry Burger in 2009, stopped at 150 volts for ethical reasons and covering 70 adults, found rates only slightly below Milgram's forty-five years earlier. So the phenomenon isn't a complete artifact. But its nature has changed: it looks less like mechanical submission and more like a delegation of responsibility to an authority you judge legitimate and whose goal you have adopted as your own.
Hold on to that phrasing. It transposes exactly.
The Chess Player's Authorities
There are four of them, and they don't work the same way.
The coach. He gives instructions, he evaluates, he has status. Our article on attachment theory in chess shows that he often occupies a place well beyond teaching.
The titled player. His rating is public and numerical, which gives his authority an objectivity few fields can offer. An opinion from a 2600 lands differently than the same opinion from an 1800, before anyone has even heard it.
Theory. "It's the main line." Nobody decided this, no body promulgates it, and yet it functions as a norm. Our article on survivorship bias in chess shows the filter through which it was assembled.
The engine. And that one is a different animal entirely.
The Engine Is a New Kind of Authority
Here is the core of this article.
Every human authority shares one property: you can question it. The answer may be poor, condescending or flat-out wrong, but a channel exists. You can ask your coach why h3, and he'll say something about the g4 square or about the bishop. Milgram's experimenter himself had four sentences, and those sentences contained a justification, however thin: the experiment requires it, it is essential.
The engine produces a number.
It doesn't say why. It can't say why. Its evaluation is the output of a search over millions of positions and, in the case of neural networks, of a function whose parameters don't translate into any language at all. This isn't a usability flaw that the next release will fix. It's constitutive.
So we're dealing with an authority figure that combines two traits which had never coexisted in history:
It is almost infallible. No human can claim to be right against it on an evaluation.
It is structurally incapable of justifying itself. Disagreement isn't merely discouraged, it's materially impossible: there's nothing to contest, because there's no argument.
And this thing is available free, permanently, to fifty million players.
Why This Is a Problem Even Though It's a Good Thing
Here's the difficulty, and it's real.
In Milgram's setup, obeying caused harm. That's what made the result scandalous. In chess, obeying the engine almost always produces a good move. It's right. Following it is rational.
The cost sits elsewhere, and it's deferred.
Improving at chess means building an autonomous capacity to evaluate. That's the definition of strength: being able to judge a position without being told. And every time you delegate an evaluation without producing your own first, you get the right answer and you don't train the function.
This isn't cheating. Long term it's worse: it's a silent substitution. You accumulate correct verdicts about specific positions without building the thing that would let you generate them.
And the Haslam and Reicher analysis applies word for word. You aren't passively submitting to the engine: you've signed up to its project. You agree that there's an objective truth of the position, that this truth is accessible, and that it beats your opinion. All three propositions are true. That's exactly why the buy-in is total, and why it never feels like obedience.
The Other Three Authorities
The coach, and the move you don't understand. Playing a move you don't understand is always a loss, even when the move is good. It doesn't attach to any structure of knowledge and won't transfer to the neighbouring position. The useful rule is simple: an unexplained move should be flagged as such, and the question asked. A good coach expects that question. One who discourages it presents a different order of problem, covered in our article on gaslighting in chess.
The titled player, and competence transfer. A title certifies a level of play. It doesn't certify teaching ability, or reliable judgment away from the board, or even correct evaluation in a type of position the player rarely enters. The authority principle operates by automatic extension, and it's the extension that costs you. Our article on the psychology of persuasion in chess details how this lever underpins an entire market.
Theory, and authority without an author. This is the strangest one, because there's nobody to question. "It's theory" is a sentence with no subject. What it actually describes is a historical consensus assembled by selecting published games, with everything that filter implies.
Four Practices
Write your evaluation down before you turn the engine on. This is the most important rule in the article, and it takes ten seconds. Note your verdict on the position, one line, then start the analysis.
The useful information isn't the engine's evaluation, which is true and tells you nothing about yourself. It's the gap between its answer and yours. Without a verdict written down first, that gap doesn't exist, and you'll reliably confuse "I understood this" with "I read the answer."
Ask why from whatever can supply it. The engine gives a number; a human can give a reason. That's the one thing human authorities still have that can't be replaced. Using them for verdicts, when a machine supplies those better, wastes the only resource they hold.
Spot the sentence that replaces the argument. "It's theory," "that's what the engine says," "a GM would never play that." Each one closes a discussion while presenting itself as a conclusion. None contains information about the position.
Accept temporary disagreement. It is perfectly legitimate not to understand why the engine prefers a move, to record the disagreement, and to come back to it in six months. What's costly isn't being wrong. It's swapping your evaluation for its own without passing through understanding.
What to Take Away
The Milgram experiment is far shakier than the retellings suggest: many participants didn't believe it, and the experimenter improvised unplanned pressure that the paper never mentions. Burger's 2009 replication nonetheless shows the phenomenon isn't empty, and Haslam and Reicher's reading has shifted what it is. It isn't blind submission but buy-in to a project you judge legitimate.
In chess, that buy-in has found an unprecedented object. The engine is the first authority that is simultaneously almost infallible and constitutionally unable to explain its verdicts. You can't argue with a number.
The paradox is that obeying produces good results here, which is what makes the mechanism invisible. The cost is never the move you played. It's the evaluation function that never gets built, when building it was the whole point.
Which leaves one rule, and it costs nothing: produce your verdict before you ask for its own.
After reading: in your next analysis, write down your evaluation of three key positions before turning on the engine. Compare. The average gap between your numbers and its numbers is a measure of your real level, and you won't get it any other way.
This article is part of a series on psychology applied to chess. It complements our article on the psychology of persuasion in chess. Upcoming articles will cover conformity and crowd psychology.
Sources
- Milgram, S. (1963). Behavioral study of obedience. Journal of Abnormal and Social Psychology, 67(4), 371-378.
- Milgram, S. (1974). Obedience to Authority: An Experimental View. Harper & Row.
- Perry, G. (2013). Behind the Shock Machine: The Untold Story of the Notorious Milgram Psychology Experiments. The New Press.
- Burger, J. M. (2009). Replicating Milgram: Would people still obey today? American Psychologist, 64(1), 1-11.
- Haslam, S. A., & Reicher, S. D. (2017). 50 years of "Obedience to Authority": From blind conformity to engaged followership. Annual Review of Law and Social Science, 13, 59-78.
- Hollander, M. M., & Turowetz, J. (2018). Multiple compliant processes: A reply to Haslam and Reicher on the engaged followership explanation of "obedience" in Milgram's experiments. British Journal of Social Psychology, 57(2), 293-305.
Comments