Jev shies away from unknown odds; people don't

When one gamble states its odds and the other only lists its possible payoffs ('probabilities you are not told'), which does Jev pick, compared with people?

result452 questions

Offered a gamble whose odds aren't stated, Jev takes it 32% of the time; people playing for real money take it 57%. The unknown gamble is Jev's likelier pick in 21% of problems, and people's in 63%. Jev even passes up, 41% of the time, unknown gambles that pay more than the alternative whatever happens.

all problems32% · 57%

against a sure amount29% · 57%

Jevpeople50%

How to read this: Each pair of bars is how often people and Jev pick the gamble with unstated odds, over all problems and over those where the other option is a sure amount; the 50% line would mean no preference between known and unknown odds.

452 problems; 90% intervals [0.301, 0.333] (Jev) and [0.549, 0.585] (people). Against a sure amount (259 problems): Jev 29%, people 57%. Jev's guess of most people (27%) sits near its own answer, so it doesn't expect people to differ.

In short

Jev steers clear of gambles whose odds aren't stated, while real-money players on Mechanical Turk slightly favored them.
Even in 34 problems where the unknown gamble beat the sure amount whatever happened, Jev took it only 59% of the time, people 82%.
Asked what most people would pick, Jev expects them to be as cautious as it is, though the real players were not.

What the data shows

Shut Up And Take My Money Fry meme: odds not stated?; shut up and take my money

Jev avoids unknown odds; people don't. Jev takes the unknown gamble 32% of the time, people 57%. The unknown gamble is Jev's likelier pick in 21% of problems, and people's in 63%. Against a sure amount it's 29% vs 57%.
Even when it can't lose. In 34 problems, every payoff of the unknown gamble is higher than the sure amount, so it wins whatever the odds. People take it 82% of the time; Jev 59%.
And when it can't win. In 26 problems every unknown payoff is lower than the sure amount. Jev almost never takes those (4%), people sometimes do (19%).
It doesn't think people differ. Asked what most people would pick, Jev puts 27% on the unknown gamble, right next to its own answer. It projects its caution onto people.

What it means, and what it doesn't

Jev treats "you are not told the odds" as a reason to walk away, more strongly than real players did, and sometimes past the point of sense. If you ask it about an option with unclear numbers, expect a nudge toward the safe, known choice, and an assumption that you'd want that too.

It doesn't mean Jev is cautious about risk in general: with the odds stated, it follows the better bet about as much as people ("Jev leans toward the better bet about as much as people do"). It's the missing information, not the risk, that it dislikes.

Caveats

Not pure ambiguity. The unknown gamble still lists its possible payoffs, so part of each choice is about amounts, not odds. The problems were split by whether the unknown gamble could beat the known one; Jev shies away in every group.
Small real stakes. The players were US workers on Mechanical Turk, paid a bonus of 10% of one outcome; about 15 to 18 played each problem. Real but small money.
The wording was the project's. The phrase "with probabilities you are not told" is the project's rendering of the study's hidden-odds display. A different phrasing, like "odds unknown", might read as more or less ominous.
Jev answers each problem fresh. Players chose five times per problem; Jev answers once, as a probability over the two options. The human share is the share of trials, so the two aren't exactly the same kind of number.

Jev on this experiment

Would a person find it interesting to read?: Yes76%
Does it describe you?: No60%
Would you have predicted it?: No52%
How fair is the comparison?: The comparison is reasonable
How much should a reader rely on it?: Moderately
Which caveat matters most?: Not pure ambiguity54%

Why ask this

In 1961 Daniel Ellsberg showed that people prefer known risks to unknown ones: offered a bet on an urn with a known 50-50 mix, or one with an unknown mix, most people pick the known urn, though nothing suggests the unknown one is worse. This "ambiguity aversion" is one of the most studied quirks of human choice.

A model that talks people through decisions (a job offer with an unclear bonus, a new product with no track record) could amplify that caution or correct it. This shows which way Jev leans, and how that compares with people who were playing for real.

How this was done

The people and the data

The comparison comes from choices13k (Peterson and colleagues, 2021, in Science), a very large dataset of risky choices. US workers on Amazon Mechanical Turk chose between pairs of gambles five times each and were paid a bonus of 10% of one outcome, so their choices counted. In 452 of the problems, one gamble listed its possible payoffs but not their odds. About 15 to 18 people played each.

On average those players weren't ambiguity-averse at all: they took the unknown gamble 57% of the time.

What Jev was asked

Each problem was one question with both gambles written out as the players saw them, the unknown one flagged in words:

Imagine you must play one of these two gambles once, for real money (wins are paid to you, losses come out of your pocket). Which do you choose: gamble_a or gamble_b?
gamble_a: $27 with a 90% chance, or -$5 with a 10% chance · gamble_b: one of these amounts: $22, $26.5 or $27.5, with probabilities you are not told

Jev answered with the two options in both orders, and also said which one most people would pick.

How it was measured

For each problem, the share of Jev's answer on the unknown gamble, and the share of players' trials on it. These are averaged across problems; separately, the analysis counts the problems where each side more often picks the unknown gamble. The same is repeated for the 259 problems where the known option is a sure amount.

Where these questions live

452 questions across 1 topic of the map. Each opens on the map with every question in it.

Risk & Decision Style

Gamble Choices452

Every question

All 452 questions behind this result, the telling ones first: the examples the analysis points to, then the ones where Jev misses, biggest gap first.

Jev’s own answer

Showing 0 of 452