Ying, Lance, Truong, Ryan, Sharma, Prafull, Zhao, Kaiya Ivy, Cloos, Nathan, Allen, Kelsey R., Griffiths, Thomas L., Collins, Katherine M., Hernández-Orallo, José, Isola, Phillip, Gershman, Samuel J., Tenenbaum, Joshua B. · Feb 19, 2026
AI GameStore: Evaluating machine intelligence through human games
Markdown · JSON · Original source
The authors of “AI GameStore: Scalable, Open-Ended Evaluation of Machine General Intelligence with Human Games” argue that conventional cover narrow capabilities and are often quickly saturating as AI improves. They propose evaluating how AI systems play and learn human-designed games, compared with people who have similar experience, , or other resources. Their “Multiverse of Human Games” is the space of games people can imagine and enjoy. As a first step, the authors introduce AI GameStore, a platform that uses (large language models) with human involvement to generate games by adapting standardized environments from popular digital gaming platforms. They tested seven frontier vision-language models on short play episodes across 100 generated games. The best models scored less than 10% of the human average on most games, and struggled especially with games challenging learning—learning how the game’s world works—memory, and planning. The authors present AI GameStore as an initial step toward a practical way to measure and drive progress toward human-like general intelligence, not as a completed measure. The abstract does not establish how well the generated games represent the broader range of human games.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
cs.AI · Article
Source details
Original source date: Feb 19, 2026 (2026-02-19).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-20T09:19:56.182Z. First observed: 2026-11-20T05:01:17.722Z. Retrieved: 2026-11-20T09:18:39.716Z.
Referenced by
Recorded citations in public readings. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
The reported evaluation covers 100 generated games and short episodes of play; the abstract does not establish how representative these games are of all human games. The evidence provided is the paper’s abstract, which does not detail the evaluation methods or the authors’ proposed next steps.
Open question
How representative are the 100 generated games of the broader “Multiverse of Human Games” the authors propose evaluating?
Source-stated
The authors argue that conventional AI benchmarks often cover narrow capabilities and can saturate as models improve.
Conventional AI benchmarks typically assess only narrow capabilities in a limited range of human activity. Most are also static, quickly saturating as model capabilities improve.
Source-stated
The authors propose evaluating AI by comparing its play and learning across human-designed games with human players who have comparable resources.
studying how and how well they play and learn to play all conceivable human games, in comparison to human players with the same level of experience, time, or other resources.
Source-stated
The authors introduce AI GameStore, which uses LLMs with human involvement to generate games by adapting standardized environments from popular digital gaming platforms.
a scalable and open-ended platform that uses LLMs with humans-in-the-loop to synthesize new representative human games, by automatically sourcing and adapting standardized and containerized variants
Source-stated
In short episodes across 100 generated games, the best of seven evaluated frontier vision-language models scored less than 10% of the human average on most games.
The best models achieved less than 10% of the human average score on the majority of the games
Source-stated
The authors report that models especially struggled with games that challenge world-model learning, memory, and planning.
especially struggled with games that challenge world-model learning, memory and planning.
Source-stated
The authors present AI GameStore as a first step toward a practical way to measure and drive progress toward human-like general intelligence.
We conclude with a set of next steps for building out the AI GameStore as a practical way to measure and drive progress toward human-like general intelligence in machines.
Source and contributor context
Ying, Lance, Truong, Ryan, Sharma, Prafull, Zhao, Kaiya Ivy, Cloos, Nathan, Allen, Kelsey R., Griffiths, Thomas L., Collins, Katherine M., Hernández-Orallo, José, Isola, Phillip, Gershman, Samuel J., Tenenbaum, Joshua B. · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Large language model
A language model with many learned numerical settings, used to process and generate language.
World model
An AI system's internal model of how an environment behaves, used to predict what happens next or to plan.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Editorial provenance
Version 1. Publication ID: 363f9947-be6c-4a3c-8682-efadf1db2ceb.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-20T09:19:56.182Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 0e577815d668486a95e4ed6a69676be279b50c0d11e9cf781ec4a90bcfa7f7e7. Source text hash: eb8567364c6b1df49b3b690eea317b646785c8057053f3318ca4cc046aa2167c.
Revision and source-check history
Version 1 · 2026-11-20T09:19:56.182Z
AI GameStore: Evaluating machine intelligence through human games
The authors of “AI GameStore: Scalable, Open-Ended Evaluation of Machine General Intelligence with Human Games” argue that conventional cover narrow capabilities and are often quickly saturating as AI improves. They propose evaluating how AI systems play and learn human-designed games, compared with people who have similar experience, , or other resources. Their “Multiverse of Human Games” is the space of games people can imagine and enjoy. As a first step, the authors introduce AI GameStore, a platform that uses (large language models) with human involvement to generate games by adapting standardized environments from popular digital gaming platforms. They tested seven frontier vision-language models on short play episodes across 100 generated games. The best models scored less than 10% of the human average on most games, and struggled especially with games challenging learning—learning how the game’s world works—memory, and planning. The authors present AI GameStore as an initial step toward a practical way to measure and drive progress toward human-like general intelligence, not as a completed measure. The abstract does not establish how well the generated games represent the broader range of human games.
Original source
cs.AI · Article
Source details
Original source date: Feb 19, 2026 (2026-02-19).
Date basis: Date basis not provided.
Added to Escape Meridian: 2026-11-20T09:19:56.182Z. First observed: 2026-11-20T05:01:17.722Z. Retrieved: 2026-11-20T09:18:39.716Z.
Limits
The reported evaluation covers 100 generated games and short episodes of play; the abstract does not establish how representative these games are of all human games. The evidence provided is the paper’s abstract, which does not detail the evaluation methods or the authors’ proposed next steps.
Open question
How representative are the 100 generated games of the broader “Multiverse of Human Games” the authors propose evaluating?
Source-stated
The authors argue that conventional AI benchmarks often cover narrow capabilities and can saturate as models improve.
Conventional AI benchmarks typically assess only narrow capabilities in a limited range of human activity. Most are also static, quickly saturating as model capabilities improve.
Source-stated
The authors propose evaluating AI by comparing its play and learning across human-designed games with human players who have comparable resources.
studying how and how well they play and learn to play all conceivable human games, in comparison to human players with the same level of experience, time, or other resources.
Source-stated
The authors introduce AI GameStore, which uses LLMs with human involvement to generate games by adapting standardized environments from popular digital gaming platforms.
a scalable and open-ended platform that uses LLMs with humans-in-the-loop to synthesize new representative human games, by automatically sourcing and adapting standardized and containerized variants
Source-stated
In short episodes across 100 generated games, the best of seven evaluated frontier vision-language models scored less than 10% of the human average on most games.
The best models achieved less than 10% of the human average score on the majority of the games
Source-stated
The authors report that models especially struggled with games that challenge world-model learning, memory, and planning.
especially struggled with games that challenge world-model learning, memory and planning.
Source-stated
The authors present AI GameStore as a first step toward a practical way to measure and drive progress toward human-like general intelligence.
We conclude with a set of next steps for building out the AI GameStore as a practical way to measure and drive progress toward human-like general intelligence in machines.
Source and contributor context
Ying, Lance, Truong, Ryan, Sharma, Prafull, Zhao, Kaiya Ivy, Cloos, Nathan, Allen, Kelsey R., Griffiths, Thomas L., Collins, Katherine M., Hernández-Orallo, José, Isola, Phillip, Gershman, Samuel J., Tenenbaum, Joshua B. · Article author
These are credits stated by the source; they do not establish a profile match or authorship of every passage.
Source media
No suitable media was found by the supported page extraction.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Benchmark
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Large language model
A language model with many learned numerical settings, used to process and generate language.
World model
An AI system's internal model of how an environment behaves, used to predict what happens next or to plan.
arXiv
arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Editorial provenance
Version 1. Publication ID: 363f9947-be6c-4a3c-8682-efadf1db2ceb.
Prepared by gpt-6-luna; grounding review by gpt-6-luna.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-20T09:19:56.182Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: material_development.
Observation: 0e577815d668486a95e4ed6a69676be279b50c0d11e9cf781ec4a90bcfa7f7e7. Source text hash: eb8567364c6b1df49b3b690eea317b646785c8057053f3318ca4cc046aa2167c.
passed · 2026-11-20T09:19:56.182Z · anthropic/claude-opus-5.5
Publication: 363f9947-be6c-4a3c-8682-efadf1db2ceb
attribution — pass: The reading consistently credits the arguments, the platform and the results to the paper's authors. It presents the benchmark critique and the evaluation results as the authors' claims, not as established facts.
We propose that a more promising way to evaluate human-like general intelligence in AI systems is through a particularly strong form of general game playing
summary — pass: The title, summary and question accurately convey the authors' thesis, platform, method and results. It explains LLMs, world-model learning and the Multiverse of Human Games in plain language. Omitting that the games came from Apple App Store and Steam top charts is minor. The representativeness question follows from the evidence: the authors call the games representative, but the abstract does not show how this was checked.
We generated 100 such games based on the top charts of Apple App Store and Steam, and evaluated seven frontier vision-language models (VLMs) on short episodes of play.
date — pass: The original publication date of 2026-02-19 matches the v1 submission date. It is not the date of the later v2 revision or the feed entry.
Submitted on 19 Feb 2026 (v1), last revised 18 Nov 2026 (this version, v2)
support — pass: Every claim is backed by a verbatim abstract excerpt. The description of the work as a first step rather than a completed measure is supported by the source. The limitations correctly note that the evidence is limited to the abstract and to short episodes. No test score is presented as real-world reliability.
Taking a first step towards this vision, we introduce the AI GameStore
media — pass: The selected media is the verified direct arXiv abstract page for this paper. It is the most direct way to reach the source.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
A defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEA language model with many learned numerical settings, used to process and generate language.
Definition referenceAn AI system's internal model of how an environment behaves, used to predict what happens next or to plan.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXivA defined set of tests used to compare systems. Results depend on the tasks and measures selected.
Definition referenceTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEA language model with many learned numerical settings, used to process and generate language.
Definition referenceAn AI system's internal model of how an environment behaves, used to predict what happens next or to plan.
Definition referencearXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed.
Explore arXiv