Escape Meridian

· Nov 17, 2023

Edgar A. Duéñez-Guzmán and coauthors argue in Nature Machine Intelligence for social intelligence

Markdown · JSON · Original source

The abstract is dated November 17, 2023. Edgar A. Duéñez-Guzmán, Suzanne Sadedin, Jane X. Wang, Kevin R. McKee and Joel Z. Leibo argue that intelligence should be understood in a social context, not only as a property of isolated agents. They point to evidence from behavior, neuroscience and evolution that natural intelligence emerged at multiple scales in networks of interacting agents. For efforts to reverse-engineer intelligence, they say, those findings constrain which algorithms may work. They also note that specific population structures have enabled artificial intelligence agents to master strategic games such as Capture-The-Flag and StarCraft II. The authors’ conclusion is that understanding intelligence demands a multi-agent context, with implications for both cognitive science and . The abstract does not specify which population structures or algorithms are needed, or how broadly the examples apply.

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.

Google DeepMindArticle · deepmind.googleOpen original source

Original source

· Article

Source details

Original source date: Nov 17, 2023 (2023-11-17T00:00:00+00:00).

Date basis: Date stated on the publisher’s page.

Added to Escape Meridian: 2026-11-22T10:58:15.331Z. First observed: 2026-11-02T18:13:56.152Z. Retrieved: 2026-11-02T18:15:23.618Z.

Referenced by

Recorded citations in public readings. A reference alone does not establish agreement or influence.

No public references recorded here yet.

Reference data (JSON)

Limits

Text as retrieved on November 2, 2026; originally published November 17, 2023. The supplied source text is an abstract; it does not provide details about the population structures, algorithms, or scope of the reported game examples.

Open question

Which population structures and interactions are necessary for agents to master complex strategic tasks, and how broadly do the examples in the abstract apply?

Source-stated

The authors argue that intelligence has traditionally been treated as a property of isolated agents, but evidence across several fields points to intelligence emerging in social networks.

converging evidence in behavior, neuroscience and evolution shows that natural intelligence emerged at multiple scales in networks of interacting agents.

Source-stated

The authors say these findings constrain the algorithms that may work for efforts to reverse-engineer intelligence.

For those interested in reverse-engineering intelligence, these findings suggest constraints on the space of workable algorithms.

Source-stated

The authors report that specific population structures have enabled agents to master complex strategic games, including Capture-The-Flag and StarCraft II.

with specific population structures enabling agents to master complex strategic games like Capture-The-Flag and StarCraft II.

Source-stated

The authors conclude that understanding intelligence requires a multi-agent context, benefiting both cognitive science and machine learning.

understanding intelligence demands a multi-agent context.

Source and contributor context

Google DeepMind · Publisher

The publisher is identified here. No individual author has been inferred.

Source media

Media retrieval status: unchecked.

Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.

Machine learning

Methods that let computers learn patterns from examples instead of following rules written by hand.

Nature

Nature publishes research, news, and commentary across the sciences.

Google DeepMind

Google DeepMind develops AI systems for general tasks and scientific research, including AlphaFold.

Editorial provenance

Version 1. Publication ID: bbb9efe1-5317-5616-297a-237eb3dfbb06.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: passed by anthropic/claude-opus-5.5 on 2026-11-22T10:53:48.142Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.

Observation: a326dd4b286a8a3abeea885944f6e7453c1de75700494574fde93b03ac67bce5. Source text hash: 656f542f1c1123779514ab505e15fef4fbcc40ae4204cd543fa049c655819e1b.

Revision and source-check history

Version 1 · 2026-11-22T10:58:15.331Z

Edgar A. Duéñez-Guzmán and coauthors argue in Nature Machine Intelligence for social intelligence

The abstract is dated November 17, 2023. Edgar A. Duéñez-Guzmán, Suzanne Sadedin, Jane X. Wang, Kevin R. McKee and Joel Z. Leibo argue that intelligence should be understood in a social context, not only as a property of isolated agents. They point to evidence from behavior, neuroscience and evolution that natural intelligence emerged at multiple scales in networks of interacting agents. For efforts to reverse-engineer intelligence, they say, those findings constrain which algorithms may work. They also note that specific population structures have enabled artificial intelligence agents to master strategic games such as Capture-The-Flag and StarCraft II. The authors’ conclusion is that understanding intelligence demands a multi-agent context, with implications for both cognitive science and . The abstract does not specify which population structures or algorithms are needed, or how broadly the examples apply.

Original source

· Article

Source details

Original source date: Nov 17, 2023 (2023-11-17T00:00:00+00:00).

Date basis: Date stated on the publisher’s page.

Added to Escape Meridian: 2026-11-22T10:58:15.331Z. First observed: 2026-11-02T18:13:56.152Z. Retrieved: 2026-11-02T18:15:23.618Z.

Limits

Text as retrieved on November 2, 2026; originally published November 17, 2023. The supplied source text is an abstract; it does not provide details about the population structures, algorithms, or scope of the reported game examples.

Open question

Which population structures and interactions are necessary for agents to master complex strategic tasks, and how broadly do the examples in the abstract apply?

Source-stated

The authors argue that intelligence has traditionally been treated as a property of isolated agents, but evidence across several fields points to intelligence emerging in social networks.

converging evidence in behavior, neuroscience and evolution shows that natural intelligence emerged at multiple scales in networks of interacting agents.

Source-stated

The authors say these findings constrain the algorithms that may work for efforts to reverse-engineer intelligence.

For those interested in reverse-engineering intelligence, these findings suggest constraints on the space of workable algorithms.

Source-stated

The authors report that specific population structures have enabled agents to master complex strategic games, including Capture-The-Flag and StarCraft II.

with specific population structures enabling agents to master complex strategic games like Capture-The-Flag and StarCraft II.

Source-stated

The authors conclude that understanding intelligence requires a multi-agent context, benefiting both cognitive science and machine learning.

understanding intelligence demands a multi-agent context.

Source and contributor context

Google DeepMind · Publisher

The publisher is identified here. No individual author has been inferred.

Source media

Media retrieval status: unchecked.

Terms and references

Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.

Machine learning

Methods that let computers learn patterns from examples instead of following rules written by hand.

Nature

Nature publishes research, news, and commentary across the sciences.

Google DeepMind

Google DeepMind develops AI systems for general tasks and scientific research, including AlphaFold.

Editorial provenance

Version 1. Publication ID: bbb9efe1-5317-5616-297a-237eb3dfbb06.

Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.

Source check: passed by anthropic/claude-opus-5.5 on 2026-11-22T10:53:48.142Z. This checks fidelity to the source, not whether its claims are independently true.

Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.

Observation: a326dd4b286a8a3abeea885944f6e7453c1de75700494574fde93b03ac67bce5. Source text hash: 656f542f1c1123779514ab505e15fef4fbcc40ae4204cd543fa049c655819e1b.

passed · 2026-11-22T10:53:48.142Z · anthropic/claude-opus-5.5

Publication: bbb9efe1-5317-5616-297a-237eb3dfbb06

attribution — pass: The reading attributes the argument to the five named authors and to the Nature Machine Intelligence publication listed on the Google DeepMind page. The page is a single publication abstract, not a listing.

Edgar A. Duéñez-Guzmán, Suzanne Sadedin, Jane X. Wang, Kevin R. McKee, Joel Z. Leibo

summary — pass: The summary faithfully conveys the abstract's thesis. Intelligence has been treated as a property of isolated agents, while evidence points to its emergence in networks of interacting agents. It also covers the constraint on workable algorithms, the game examples and the multi-agent conclusion. 'Solipsistic' is rendered plainly as 'isolated agents', and machine learning is defined on the page. The question follows from gaps the abstract leaves open. The authors' position and its force are preserved.

We posit that moving beyond a solipsistic view of agency will benefit both cognitive science and machine learning

date — pass: The reading states November 17, 2023, which matches the page dateline and the UTC day of originalPublished. Nothing is presented as happening after that date, and there is no hindsight framing. The page was modified in 2025, but the reading uses only the abstract, which is plainly the original publication content. This is a publisher page, so no archive disclosure is needed.

November 17, 2023

support — pass: Each claim is attributed and matches its excerpt. The fourth claim compresses the authors' 'posit' about benefits to both fields into their conclusion, but it remains attributed and faithful. The limitations correctly note that the abstract omits details. The page was edited after publication, but the cited text is the core abstract and plainly part of the original piece. No benefits or costs beyond the source are invented.

understanding intelligence demands a multi-agent context.

media — pass: No media was selected and no verified media candidates were supplied, so nothing is misidentified.

Challenge or correct this reading

Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.