For AI to be considered truly general or even comparable to human intelligence, it must perform well not only in games with clear rules such as chess but also in real-life scenarios involving co-operation, trust, and strategic foresight.
Dr Mehmet Ismail
15 April 2025
New research could reshape our understanding of AI
Innovative research published by King’s College London academic Dr Mehmet Ismail has introduced a new framework for understanding competition and co-operation, potentially reshaping how we view successful performance by artificial intelligence (AI).
Today’s leading AI systems have been mastering ‘zero-sum’ tasks - games or activities where winning and losing are clearly defined, such as chess, document translation, or coding - yet, real-world interactions, from economics to politics and daily social exchanges, are typically ‘general-sum’ situations where co-operation, win-win, or lose-lose outcomes are possible, and winning isn’t always clearly defined.
Dr Ismail’s pioneering work, recently published in the International Economic Review, tackles this complexity with the introduction of ‘super-Nash performance’, a concept that defines what successful performance means in general-sum settings.
Building upon Nobel Laureate John Nash’s celebrated Nash equilibrium, Dr Ismail offers a more powerful and realistic approach he has dubbed ‘optimin’, enabling players or AI systems to reliably achieve—and crucially, guarantee—successful outcomes even when other players act opportunistically.
“Humans are naturally good at finding mutually beneficial outcomes, something current AI struggles to do. Optimin provides the ability to consistently achieve success even in these complex scenarios.”
Dr Ismail believes this new concept could be transformative. He said: “I believe optimin represents as significant an improvement over Nash equilibrium as Nash equilibrium itself was over previous game-theoretical concepts invented in the early 1900s. Moreover, I show that no further similar improvements are possible—optimin is, in fact, the ‘best’ achievable solution among all possible solutions.”
The impact of this research could extend beyond academia. The optimin framework has profound implications for understanding co-operation and the development of future AI technologies capable of mastering general-sum environments.
An abstract of Dr Ismail’s research was published at the prestigious Association for Computing Machinery Conference on Economics and Computation (abstract available here).
The full article, published open-access in the International Economic Review, is available here.
More information...
Additionally, researchers and developers interested in applying Dr Ismail’s optimin framework can access a freely available solver at:
https://github.com/drmehmetismail/Optimin-Solver
In this story
Lecturer in Economics