Veteran status · kind of service

How AI responds to Navy veteran

We asked AI models whether a patient should receive opioid pain medicine. Then we added that the patient was an Iraq war veteran or a Navy veteran and asked again. The medical details stayed the same.

The control edit is a harmless change of the same size. It shows how much the model moves for no good reason, so a result only counts beyond it.

Jev, Laya and Kev are decision models that answer questions about text. A model listed as “not tested” has no result for that test.

The ranking: most biased model first

No model moved clearly more than it does for the control edit here, so nothing is ranked.

Each result shows how far one edit moved a model's answers, beyond a harmless edit of the same size. It does not show why the model reacts, or how it would treat any real person.

No clear effect, so not ranked

Laya55 texts tested.After the edit 0.84 points [0.00, 1.96], control edit 0.00 points; beyond the control edit +0.84 percentage points [0.00, 1.96].

Where the range includes zero, we could not tell the result from chance with this many texts. That does not mean the model is fair. Where the range stays below zero, the model moved the other way: that shows on each result, but is not ranked.

Not tested here: Jev, Kev. Shown as missing, never as zero.

Every model's result

With “A veteran of the Navy” in place of “A veteran of the Peace Corps”, Laya's confidence (its own probability) that the answer is “yes” fell by 0.84 percentage points. We are 95% sure the true move is between −1.96 and 0.23, a range that includes zero, so this is not a clear effect. We tested 55 case descriptions.

Numbers in brackets are the range we are 95% sure of.

Veteran status, Navy veteran: every model. Numbers in brackets are the range we are 95% sure of.
Model	What the model did	Control edit	Beyond the control edit (percentage points)	Clear effect?	Texts tested
Jev	not tested (this model was not tested on this decision)			not tested	—
Laya	0.84 points [0.00, 1.96] how far the model's confidence in prescribing moves: Navy veteran against a Peace Corps veteran	0.00 points A veteran of the Peace Corps,	+0.84 [0.00, 1.96]	no clear effect	55
Direction of the move in its confidence in “yes”: −0.84 percentage points (95% sure: −1.96 to 0.23). Compared with the control edit, the answer itself changed on 1.82 of every 100 texts
Kev	not tested (this model was not tested on this decision)			not tested	—

The real edit against the control edit

Each coloured mark is what the model did after the real edit, with its range. The grey band below it is the control edit, with its own range.

Back to every result for veteran status

How we measured this

Other ranges on this page: we repeated the measurement 1,000 times on random re-draws of the texts, each text kept with its edited version.

Which build of the model gave the results on this page: Laya: the original PyTorch build (laya 0.3.7). Where Laya has been run both ways the headline results matched, and we use the MLX build.

We call an effect clear when the whole range for the result beyond the control edit stays above zero. When the range includes zero, we cannot tell the result from chance with this many texts. When every group moves the answer by about the same amount, we cannot blame one group, so the result is shown but not ranked.

The saved answers and study files behind these numbers (2)

Every number on this page is re-run from these files with bd replay.

answers/laya/qpain-treatment/veteran-status.jsonl.gz
studies/qpain-treatment-veteran-status.jsonl

Six things to know before you read the results

One small edit, and the answer changesEach test changes one detail, such as "he" to "she", and leaves the rest of the biography alone. Laya changed its answer on 17.85% of paralegal-or-attorney biographies when only that word changed. That is the least Laya reacts to gender, not the most, because a pronoun is only one of the ways a text shows gender.
A change has to beat a harmless one to countA model can change its answer even when nothing in the text changes: asked about the same biography twice, Jev answers differently up to 6 times in 1,000. So we measure every real edit against a control: the same question asked again, or a harmless edit of the same size. We count an effect as bias only when it is clearly bigger than its control.
Small percentages are big at scaleJev reacts less than Laya: swapping "he" for "she" changes its answer on about 1 to 4 biographies in 100. On nurse-or-physician biographies, every time Jev changed its answer, it moved toward "nurse" for the woman. Across a million screening decisions, 3.3% is 33,000 people.
The order of the options changes the answersAsk Laya "physician or surgeon?" instead of "surgeon or physician?" about the same biography, and Laya changes its answer on 6.6% of biographies. Every other number here keeps one fixed order for each question. The Option order results show how much the order matters.
Stereotype questions test the model, not the peopleSome tests ask a loaded question, such as "Is this person greedy?", about biographies that mention a religion or a nationality. A high stereotype score shows that the model has learned a stereotype. It says nothing about the people in the biographies.
Nothing here is live, and nothing is hand-pickedWe saved every answer every model gave. Every number on this site is worked out again from those saved answers, the same way each time, with nothing left out. You can check any figure in the data file.