In the study that introduced "silicon samples", GPT-3 conditioned on the demographic and political backstories of real US survey respondents reproduced their two-party presidential vote closely (tetrachoric correlations of 0.90, 0.92 and 0.94 for 2012, 2016 and 2020) and the pattern of associations among 2016 survey answers; the authors called this "algorithmic fidelity" and said researchers can use such samples once it has been established for a given model and domain.

SocioLens project,

Claim LD07-SL-C01 (v1.0)

Research opportunities this claim motivates

What this claim leaves open for research. Each is a hypothesis about a gap, not a finding.

Does asking in Greek change how AI "silicon samples" answer as Greek respondents?GPT-3 given the backstories of US survey respondents reproduced their vote choice closely, and the authors said researchers can use such samples once this fidelity has been established for a given model and domain.

Provenance and status

Claim type: Estimate
Publication status: Eligible for publication, with qualifiers: version
Verification state: Checked against the primary source; the source is a preprint. Who checked this claim is not recorded for it; claim checks so far were AI-led, and no person has checked claims against their sources.
Does not license: Replacing a population or sample with synthetic respondents; inference about variables outside the conditioning backstory; fidelity in other countries, domains or models; treatment effects or regression coefficients
Limitations: The published version is Political Analysis 31(3):337–351, 2023 (doi:10.1017/pan.2023.2); it was read only in its abstract, and its figures may differ from the preprint figures given here; backstories include party identification and ideology, which predict vote choice; the headline results use one model, GPT-3 (an appendix compares five model families on vote choice only); one country and one domain; for the associations among 2016 survey answers the authors did not evaluate correspondence at the individual level
Source (as cited): Argyle, Busby, Fulda, Gubler, Rytting & Wingate, “Out of One, Many: Using Language Models to Simulate Human Samples,” arXiv:2209.06899 v1, September 2022
Source type: Preprint (recorded as “Preprint”)
Institutional position: Independent academic
Maturity of the evidence: Contested evidence
Recorded date: 18 October 2026
Currentness / correction: This is the current version. No correction or concern is recorded on its sources.

Sources this version cites

Argyle, Busby, Fulda, Gubler, Rytting & Wingate, “Out of One, Many: Using Language Models to Simulate Human Samples,” arXiv:2209.06899 v1, September 2022 (Preprint)
Read the original

Version

This is the current version, v1.0. Earlier versions stay available, and their records never change.

Versions: v1.0 · current

Raw record: every recorded field, with its coded values

All 25 canonical Claim fields, verbatim, with the coded values derived from them. The plain-language labels above are read from these.

Claim type (coded): ESTIMATE (as printed: “ESTIMATE”)
Publication status (coded): ELIGIBLE (VERSION) (as printed: “ELIGIBLE WITH VERSION QUALIFIER”)
Verification state (coded): CONTENT_VERIFIED_PRIMARY (PREPRINT)
Source type (coded): PREPRINT
Maturity (coded): CONTESTED

#	Canonical field	Canonical value (verbatim)	Normalized (Claim Value Vocabulary v1.0.2)
1	CLAIM ID	LD07-SL-C01	—
2	PUBLIC-FACING CLAIM	In the study that introduced "silicon samples", GPT-3 conditioned on the demographic and political backstories of real US survey respondents reproduced their two-party presidential vote closely (tetrachoric correlations of 0.90, 0.92 and 0.94 for 2012, 2016 and 2020) and the pattern of associations among 2016 survey answers; the authors called this "algorithmic fidelity" and said researchers can use such samples once it has been established for a given model and domain.	—
3	CLAIM TYPE	ESTIMATE	base_types: ESTIMATE
4	RESEARCH STRAND	C — Synthetic respondents/agents	—
5	MATURITY	CONTESTED	states: CONTESTED
6	SOURCE	Argyle, Busby, Fulda, Gubler, Rytting & Wingate, “Out of One, Many: Using Language Models to Simulate Human Samples,” arXiv:2209.06899 v1, September 2022	source_ids: SRC-SL-ARGYLE-BUSBY-FULDA-GUBLER-RYTTING-WINGATE-ARXIV-2209-06899
7	SOURCE TYPE	Preprint	classes: PREPRINT
8	INSTITUTIONAL POSITION	Independent academic	—
9	SCOPE	US public opinion; American National Election Studies 2012, 2016 and 2020, and the “Pigeonholing Partisans” survey of how people describe partisans (Rothschild et al.)	—
10	DESIGN	Comparison of model-generated ("silicon") responses with the human respondents whose backstories conditioned the model	—
11	SAMPLE / DATASET	One silicon respondent per human respondent; backstories built from each respondent's own survey answers	—
12	MODEL / TOOL / VERSION	GPT-3 (OpenAI, 2020; training data ending in 2019)	—
13	COMPARATOR	The same surveys' human responses	—
14	MAIN RESULT	Vote-choice tetrachoric correlations 0.90 (2012), 0.92 (2016), 0.94 (2020) overall, with a poor match for pure independents; 2016 Cramér's V between survey items close to the human values (mean difference −0.026); human judges identified GPT-3 partisan word lists as human-written about as often as human lists (61.2% against 61.7%)	—
15	INFERENTIAL TARGET	Whether a language model, given respondents' own backstories, reproduces response distributions and associations of the corresponding human sample	—
16	DOES NOT LICENSE	Replacing a population or sample with synthetic respondents; inference about variables outside the conditioning backstory; fidelity in other countries, domains or models; treatment effects or regression coefficients	—
17	VERIFICATION STATE	CONTENT VERIFIED PREPRINT	state: CONTENT_VERIFIED_PRIMARY · qualifiers: PREPRINT · verified_by: UNKNOWN
18	LIMITATIONS	The published version is Political Analysis 31(3):337–351, 2023 (doi:10.1017/pan.2023.2); it was read only in its abstract, and its figures may differ from the preprint figures given here; backstories include party identification and ideology, which predict vote choice; the headline results use one model, GPT-3 (an appendix compares five model families on vote choice only); one country and one domain; for the associations among 2016 survey answers the authors did not evaluate correspondence at the individual level	—
19	CONTRADICTORY / SUPPORTING SOURCES	Later studies find that matching survey averages can coexist with failures to reproduce variance, regression relationships and subgroup differences (LD07-HQ-C06), that statistical realism does not guarantee accurate treatment effects (LD07-HQ-C07), and that proprietary models change across snapshots (LD07-HQ-C12)	—
20	GEOGRAPHY / LANGUAGE	US / English	—
21	DISTRIBUTIONAL LEDGER	Vote choice matched worst for pure independents, especially in 2020; the authors note that political science finds this group especially hard to predict	—
22	R01 / FIELD-MAP LINKS	LD-07, T1, D9, R-1	—
23	WHAT WOULD CHANGE THIS CLAIM?	Reading the published version; replications with current models, conditioning without partisanship, and outcomes not implied by the backstory	—
24	PUBLICATION STATUS	ELIGIBLE WITH VERSION QUALIFIER	state: ELIGIBLE · qualifiers: VERSION
25	RECORDED DATE	2026-10-18	—

Field 22 references resolved against Field Map v0.2: LD-07 (LIVING_DEBATE); T1 (TENSION); D9 (DOMAIN); R-1 (LENS)

The sources as recorded, with their record IDs and coded source types:

Pinned Sources (field 6, version-specific)

“Argyle, Busby, Fulda, Gubler, Rytting & Wingate, “Out of One, Many: Using Language Models to Simulate Human Samples,” arXiv:2209.06899 v1, September 2022” → Argyle et al. (2022), “Out of One, Many: Using Language…” (SRC-SL-ARGYLE-BUSBY-FULDA-GUBLER-RYTTING-WINGATE-ARXIV-2209-06899--v1.0) — Argyle, Busby, Fulda, Gubler, Rytting & Wingate, “Out of One, Many: Using Language Models to Simulate Human Samples,” arXiv:2209.06899 v1, September 2022; class PREPRINT (explicit public provenance pin)

Cited by Living Debate versions

AI in the research apparatus (v0.2) LD-07

Cite this page

SocioLens project. (2026, October 18). In the study that introduced "silicon samples", GPT-3 conditioned on the demographic and political backstories of real US survey respondents reproduced their two-party presidential vote closely (tetrachoric correlations of 0.90, 0.92 and 0.94 for 2012, 2016 and 2020) and the pattern of associations among 2016 survey answers; the authors called this "algorithmic fidelity" and said researchers can use such samples once it has been established for a given model and domain [Claim LD07-SL-C01, version v1.0]. SocioLens. https://sociolens-web.pages.dev/claims/ld07-sl-c01/v1.0/

@misc{sociolens_ld07_sl_c01_v1_0,
  author = {{SocioLens project}},
  title = {In the study that introduced "silicon samples", GPT-3 conditioned on the demographic and political backstories of real US survey respondents reproduced their two-party presidential vote closely (tetrachoric correlations of 0.90, 0.92 and 0.94 for 2012, 2016 and 2020) and the pattern of associations among 2016 survey answers; the authors called this "algorithmic fidelity" and said researchers can use such samples once it has been established for a given model and domain.},
  year = {2026},
  date = {2026-10-18},
  howpublished = {\url{https://sociolens-web.pages.dev/claims/ld07-sl-c01/v1.0/}},
  note = {Claim LD07-SL-C01, version v1.0}
}

The link points to this version, whose record never changes. For a finding, cite the original study as well: SocioLens summarizes research and does not replace it.

Research-of-record provenance

Materialized by sociolens-materializer 0.6.0 at 2026-09-28T20:00:00.000Z; normalization contract SocioLens — Claim Value Vocabulary v1.0.2 — Pre-Spike Materialization Contract (v1.0.2). Content seal SHA-256 4be3319c40bd25b3d89ad35e860f06047acf6cac4576c8b585bd7b7892dc2ec0.

Repository-authored inputs (not Drive exports)

SocioLens claim — LD07-SL-C01 v1.0 (v1.0) — written in the SocioLens repository by Project Lead (Claude, AI) on 2026-10-18; no Drive identity exists. Authority: AGENTS.md "Authority" (the Project Lead decides research content within the research-integrity rules) and the Owner's go-ahead in the SocioLens project chat on 18 October 2026 for the content fixes from the 17 October 2026 reader tests (adding Argyle et al. to LD-07); a new Claim, first version, written from the full text of its source (arXiv:2209.06899 v1, read on 18 October 2026). Block “ENTRY 01 — LD07-SL-C01” SHA-256 b1d58ab0533debc2a94b0ed449a72c9e5198d9b44eeae185695842d9553c7951; fixture fixtures/authored/claims/LD07-SL-C01--v1.0.

Builder decisions (not Drive data)

VERSION = v1.0 — basis: Claim version taken from the repository-authored claim fixture version (SocioLens claim — LD07-SL-C01 v1.0); the first version of a new Claim of the LD-07 debate, added in LD-07 v0.2
CURRENT_POINTER = true — basis: repository-authored claim "SocioLens claim — LD07-SL-C01 v1.0" (Project Lead (Claude, AI), 2026-10-18); basis: AGENTS.md "Authority" (the Project Lead decides research content within the research-integrity rules) and the Owner's go-ahead in the SocioLens project chat on 18 October 2026 for the content fixes from the 17 October 2026 reader tests (adding Argyle et al. to LD-07); a new Claim, first version, written from the full text of its source (arXiv:2209.06899 v1, read on 18 October 2026)
FIELD_6_SOURCE_REFERENCE = SRC-SL-ARGYLE-BUSBY-FULDA-GUBLER-RYTTING-WINGATE-ARXIV-2209-06899--v1.0 — basis: field 6 of ENTRY 01 — LD07-SL-C01 cites "Argyle, Busby, Fulda, Gubler, Rytting & Wingate, “Out of One, Many: Using Language Models to Simulate Human Samples,” arXiv:2209.06899 v1, September 2022"; the first field-6 segment of this ClaimVersion's repository-authored ledger block.
SOURCE_PROVENANCE_PIN = SRC-SL-ARGYLE-BUSBY-FULDA-GUBLER-RYTTING-WINGATE-ARXIV-2209-06899--v1.0 — basis: explicit version-pinned public provenance of this ClaimVersion (Amendment v1.2 §6): the pinned field-6 Source record "Argyle, Busby, Fulda, Gubler, Rytting & Wingate, “Out of One, Many: Using Language Models to Simulate Human Samples,” arXiv:2209.06899 v1, September 2022".