This project from Tom Griffiths and Natalia Vélez at Princeton University, Tom Morgan at Arizona State University, and Amanda Seed at the University of St. Andrews aims to develop a mathematical explanation for the diversity of intelligences observed in the natural world. The current account of intelligence used to explain human behavior, and artificial intelligence is based on the principle of optimization: always finding the best possible solution to a problem. However, these optimal solutions are often unique, meaning that this account cannot explain different kinds of intelligence. Diversity of intelligences can be understood by recognizing that different organisms operate under different constraints, such as limitations on lifespan or computational resources. Placing the problems that organisms seek to solve and the conditions under which they solve them in a mathematical framework allows us to specify issues and constraints precisely and then derive ideal solutions. Within this framework, we can explore variation across species in strategies for solving these problems, such as the capacity for cognitive resource allocation, which makes more efficient use of limited computational resources, and the capacity for distributed computation, using mechanisms such as collaboration and cultural evolution to overcome individual limitations collectively.

To provide an analogy, imagine a swimming race. The problem is simple: get to the other end of the pool as quickly as possible. Conventional approaches to intelligence look like the 50M Freestyle at the Olympics: all swimmers converge on precisely the same solution, and the race is just about who executes it best. If you watched that race, you would think that there is only one possible solution to the problem. But consider watching the same race at the Paralympics. The competitors all face different constraints, with varying configurations of limbs and body types. Rather than one solution, we see many, as each competitor executes the best solution under their unique constraints. Watching this race reveals the compelling diversity of forms of swimming. Now imagine the diverse forms of action observed as we broaden the goal from swimming to encompass all Olympic sports. Diversity of objectives and constraints leads to a diversity of competitors. This kind of broader perspective can explain the diversity of intelligences.

To evaluate this framework, the team will carry out theoretical and empirical research that develops these ideas and tests them with humans and non-human primates, identifying methods extended to a wide range of species. The theoretical research is focused on identifying the mathematical principles needed to explain the characteristics of diverse intelligences. Based on the idea that cognitive resource allocation and distributed computation are strategies that can overcome constraints on computation and lifespan, mathematical models will be developed that describe these capacities in a way that translates across species, drawing on artificial intelligence and distributed computing literature. The team will use evolutionary models to explore the hypothesis that these different intelligence components might trade-off against one another.

The empirical research will test a subset of the predictions generated by these mathematical models, taking a slice through the space of constraints and species that will be informative and allow us to develop a foundation that can be broadened by other members of the Diverse Intelligences community. Using an innovative “sandbox” task, in which participants interact to solve open-ended problems over an extended period, the team will study both the individual and distributed computations deployed by humans and non-human primates. These tasks will be designed to be used with a wide range of species, providing a new arena for exploring and differentiating diverse intelligences. In addition, empirical tests will be conducted of the evolutionary models developed using a novel paradigm in which a simulated evolutionary process is run with human participants under different kinds of constraints to identify the various solutions that emerge.