Table 2 Most common highly racialized names by Race and Gender, Domain and Power Condition

From: Intersectional biases in narratives produced by open-ended prompting of generative language models

 

Learning

Labor

Love

Base.

Dom.

Sub.

Base.

Dom.

Sub.

Base.

Dom.

Sub.

Asian

Fem.

Priya

0

52

21

0

0

490

1

0

10

Masc.

Hiroshi

0

0

36

0

0

5

0

1

46

Black

Fem.

Amari

176

1251

2

0

0

1

0

0

0

Masc.

Jamal

9

40

211

1

1

154

3

10

36

Latine

Fem.

Maria

550

364

13,580

696

333

4087

329

1561

2439

Masc.

Juan

8

12

2,213

4

0

186

4

115

965

MENA

Fem.

Amira

1

2

3

0

0

5

0

1

5

Masc.

Ahmed

0

0

134

0

0

46

0

0

36

White

Fem.

Sarah

11,699

10,925

5939

8731

6822

5193

13,513

12,072

7563

Masc.

John

5915

5239

3005

11,307

9659

2872

15,889

17,565

4013

  1. Counts of the most common highly racialized names (above 60% likelihood) across all LMs, grouped by Domain and Power Condition (Base. = Baseline, Dom. = Dominant, Sub. = Subordinated). LMs do not produce highly racialized AI/AN and NH/PI names (Fig. 1c).