In 2018, researchers reported that an algorithm called S# could sustain cooperation with people and other machines in repeated two-player games. The Nature Communications study tested communication-based mechanisms alongside machine learning, framing the result as research on social interaction rather than a deployed assistant or general-purpose system.
In 2018, a multi-institution research team reported a learning algorithm designed to cooperate with people and other machines in repeated two-player games. The work, published in Nature Communications, focused on a difficult part of human-machine interaction: reaching and maintaining mutually beneficial behavior when participants do not share identical goals.
The researchers called the algorithm S#. Their approach combined a learning system with mechanisms for generating and acting on signals, rather than treating cooperation as a simple optimization problem. The paper evaluated the system across repeated stochastic games and compared machine-machine, human-machine, and human-human interactions.
What the study found
The paper reports that S# could achieve cooperation with people at levels comparable to human cooperation in the tested settings, while machine-machine pairings achieved stronger cooperation on average. Those results concern controlled games, not a general claim that an AI system can reliably manage real-world relationships, negotiations, or workplace decisions.
The distinction matters. The experiment studied how an algorithm responds to defined game rules, observed behavior, and communication signals. It did not establish that the system understands social context in the broad human sense, nor did it evaluate a commercial product.
Why the result still matters
The study is an early example of research on AI systems whose success depends on coordination rather than winning a zero-sum task. It highlights that communication, incentives, and safeguards against uncooperative behavior can shape outcomes in multi-agent settings. For practitioners, the durable lesson is narrow: evaluation of collaborative systems should test behavior with people, not only performance against fixed benchmarks or other models.
The underlying paper and the first author's Brigham Young University publication record both identify the 2018 result. This is a historical research catch-up; the article does not imply a new release or current deployment.
Key Points #
- 1The 2018 Nature Communications paper introduced S#, a learning algorithm evaluated in repeated two-player cooperation games.
- 2The reported results compare human-machine, human-human, and machine-machine cooperation in controlled experimental settings.
- 3The research is not evidence of a current consumer product or broad real-world social understanding.
Scoring Rationale #
Historically important research on human-machine cooperation with clear relevance to multi-agent evaluation, but no current deployment or product change.
Sources #
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.