Can a computer predict you?
The Aaronson oracle is a small pattern-counting program that tries to call your next key press before you make it. This test lets you find out how readable you are when you try your hardest to be random: you press left or right a hundred times, and only afterwards see how many it called and what gave you away. The computer scientist Scott Aaronson wrote the original for his students; the result sheet shows how his class fared.
This test is a version of that program, often called the Aaronson oracle. You press left or right 100 times. Before each press the machine looks at your last five presses, finds every earlier time you made those same five, and bets on whichever side you pressed next more often; with too little evidence it falls back to your last four, three, two or one. Each guess is computed and stored before your key is read, then turned face up. Your score is how many of the 100 it called, set against what the same machine scores on a real coin.
It is not a test of intelligence, personality or free will. It measures how patterned your presses were, once, on one screen. A hundred presses is a small sample: even against a real coin, any guesser lands between 40 and 60 about 96% of the time. No crowd has played this exact machine yet, so you get no percentile, only the coin, a few fixed strategies and Aaronson’s classroom anecdote.
What the test measures
One thing: how predictable a hundred left-or-right presses are to a simple pattern-counting machine.
The task
- 100 presses, left or right, as unpredictably as you can, with no time limit
- A five-press pattern predictor, after the method Scott Aaronson described for his f/d program: a pattern counts once it has come up three times and one side leads by two; otherwise it tries shorter patterns, and with nothing to go on it falls back on a fixed default bet
- Sealed guesses: each card is written before your press and only turned over after it, so you can watch the machine’s hit rate as you go
- Fully deterministic: the same hundred presses always get the same hundred guesses
Your result
- How many of your 100 presses it called, and a band: Coin-level, Hard to read, Readable or Open book
- How often a machine guessing a real coin would do as well, worked out exactly
- What gave you away: how often you switched sides, your longest run, your favourite three-press move, a lean to one side, and your first 50 against your last 50
- Your longest run beside a real coin tossed at the same time, and the machine’s notebook showing how it made its last call
- A shareable card and a link that challenges a friend to be harder to read
Before you start
- About a minute
- Works on a phone (two big keys under your thumbs) or a computer (the ← and → keys)
- Sound is optional; the page needs a browser with 3D (WebGL) turned on
Predictability and randomness FAQ
Is this test free? Do I need to sign up to see my result?
Yes, it’s free, and there’s nothing to sign up for: no account, no email, no payment. Your result appears on this page the moment you finish, and you can share it or play again straight away.
Can a computer really predict which key I’ll press?
That is exactly what this test lets you find out, and the answer differs from person to person. A pattern-counting program needs no access to your thoughts: it only counts which short habits you repeat. The result shows how many of your 100 presses it called and how often pure luck would do as well. Scott Aaronson recalls how often his own version called his students; that classroom anecdote, not a published study, is marked on the result sheet.
How does the Aaronson oracle work?
The Aaronson oracle remembers every run of five presses you have made and what you pressed next. Before each new press it looks up your last five and bets on the more common follow-up. The version on this page also falls back to shorter patterns when it has too little evidence, and seals every guess before reading your key.
What score means I’m as random as a coin?
About 50 out of 100. A fair coin guessed by any method scores between 40 and 60 about 96% of the time, so scores in the low 50s are within luck. In this test, 57 or fewer counts as coin-level; 65 or more happens by luck only about once in 570 tries.
Are people good at being random?
That is what the result sheet explains, after your hundred presses, because knowing first would change how you press. Psychologists have asked people to make up random sequences for decades: Willem Wagenaar surveyed the experiments in 1972, Falk and Konold measured how often people switch sides in 1997, and Tversky and Kahneman described the belief behind it in 1971. The sheet sets your own switch rate and longest run beside a real coin’s.
Can you learn to be unpredictable?
Yes, with feedback. In Allen Neuringer’s 1986 study, students pressed two keys a hundred times a round and saw statistics after each round; with that feedback they learned to produce sequences that statistical tests could not tell from a computer’s random numbers.
Is this a free will test?
No. A high hit rate shows that your presses followed patterns a simple program could count, not that your choices were not free. The test also says nothing about intelligence or personality, and one run of 100 presses varies from try to try.
Where the numbers come from
No crowd has played this exact machine, so there are no percentiles: the comparisons are a fair coin (exact binomial arithmetic), fixed strategies run through the same machine, and Aaronson’s recollection of his class (an anecdote). Your presses stay on the page; a challenge link carries only how many it called. The site does not publish its own players’ data.
- Aaronson (2006), Quantum Computing Since Democritus, Lecture 18: Free Will
- Aaronson’s description of the method, quoted in the elsehow/aaronson-oracle README
- Wagenaar (1972), Psychological Bulletin — Generation of random sequences by human subjects: A critical survey of literature
- Falk & Konold (1997), Psychological Review — Making sense of randomness: Implicit encoding as a basis for judgment
- Rapoport & Budescu (1997), Psychological Review — Randomization in individual choice behavior
- Tversky & Kahneman (1971), Psychological Bulletin — Belief in the law of small numbers
- Neuringer (1986), Journal of Experimental Psychology: General — Can people behave “randomly?”: The role of feedback
How the machine guesses
Five-press patterns
There are 32 possible runs of five left/right presses. The machine keeps a count of what you pressed after each one and bets on the side that followed more often.
Fallback
A pattern is trusted only after three appearances with one side two ahead. Until then the machine tries your last four, three, two and one presses; if none decides, it falls back on a fixed default bet, spelled out on the result sheet. Its very first guess is →.
Coin comparison
Any guesser calling a fair coin 100 times gets a binomial score centred on 50. The result shows exactly how often that luck alone reaches your score.