Intelligent agents are prone to irrational action. Biases and decoys, for example, have been shown to mislead all sorts of organisms, from the simplest intelligences of slime molds to supposedly-rational humans. Sometimes groups of animals are better at making good decisions than their constituent individuals. This project will study the emergence and contours of such collective intelligence.

Rationality is variously defined across disciplines as an ability to make decisions that strategically maximize one’s outcomes or conform to a process of reasoning. Reasoning, in turn, is a process in which options are considered, and decisions are made for coherent reasons. Such reasons typically stem from logic, cognitive maps for spatial navigation, or attention to high-order causes of social or non-social events—making them relatively accurate and generalizable. For millennia, rationality and reasoning have been central to debates about what constitutes intelligence. And, critically, they have almost always been assumed to be the sole province of individual brains, and in particular of individual human brains.

This proposal challenges these fundamental conceptions. The goal is to advance a falsifiable framework for mapping rationality and reasoning across the nonhuman biological and machine worlds and for identifying rationality and reasoning among individuals and collectives. To do so, the project marshalls theory and practice from across the computational, cognitive, and evolutionary sciences. It unites experimental psychology’s focus on individual decision-making processes and collective behavior’s focus on collective outcomes to define a new domain of research into comparative collective psychology.

Specifically, the team will pursue four broad aims, through six work packages.

The four aims are:

1. To determine whether collectives of organisms can produce reasoning-like processes or rational outcomes not available to individuals
2. To determine which factors make or break the emergent capacities of collectives
3. To investigate whether these patterns and processes extend to collectives at other levels of biological organization
4. To test the reasoning capacities of individuals beyond humans. These aims are addressed through experimental and computational work across various organisms (including pigeons, fish, chimpanzees, humans, and even individual cells and whole planaria), intelligences, and decision-making spaces.

The six work packages include:

1. An exploration of group dynamics in the relatively-simple decision making of fish. Observing individual fish, shoals, and a mix of fish with a robotic “stooge” through a video monitoring system, differences in problem solving with varying parameters will be studied. 
2. A study of great apes. Chimps are known to deploy groups in many situations including hunting, aggression, etc. In particular, certain hunting techniques may indicate a predictive, logical reasoning process. However, a well-established logic test seems to indicate that chimps cannot master the kind of discriminatory thought that human children show starting at around age four. This package will create a team-based version of this test and explore the capacity of groups to solve it.
3. An extension of the lead researcher’s previous work with homing pigeons. That work has already uncovered aspects of collective intelligence and the impact of the introduction of naïve individuals to collectives. Here the team will use GPS trackers and computer mapping to observe the changing dynamics of route optimization as individuals are first allowed to learn variations from a set route as individuals and as groups. The goal is to determine whether novel shortcuts, not experienced by any one individual, emerge.
4. Work built upon previous pioneering studies showing that flatworms can successfully complete growth when badly deformed by chemical stress. That work shows the overall collection of cells solving a problem but has not probed the way in which individual cells respond. With this work package, the team asks whether the cells or the collective are better at responding and how such processes result from very basic information sharing and group decision making.
5. A virtual environment of computerized agents abstracted from the questions examined in the above explorations. These agents are given the capacity to gain information about their virtual world and build a picture of the likelihood of a given action resulting in a given outcome. This is combined with social information regarding the decisions and actions of others (but without insight into the “mind” of the other agent). Agents must then impute the motivations and calculations of one another and act according to their predictions both of the world and of other agents’ actions. Through varying the parameters of the simulation, the team will identify thresholds of information transfer, fidelity of prediction, etc. that yield good collective decision making. This provides a stripped-down, mathematically-analyzable perspective of the behavior seen in the experimental subjects above.
6. An experiment focused on individual intelligence. The team argues that there is insufficient evidence of non-human individual reasoning and seeks to apply their definitions to a set of experiments to explore whether non-human primates can individually establish logical predictions in a variety of contexts. From a variant of the Y-shaped tube test (using actors in gorilla suits and humans), to baffling continuity experiments involving rewards (grapes), to videos of dominance displays between dyads, the group will seek to infer the internal reasoning of the study subjects. To do this, they will deploy novel non-invasive eye tracking techniques that can reveal anticipatory looking, indicating expectations of outcomes.

The project is expected to yield disruptive theoretical papers, many empirical journal articles, and a range of methodological, analytical, and conceptual resources for the broader scientific community that will foster engagement from complementary disciplines and materials for directly communicating findings to the public. The knowledge generated by this framework will be instrumental in understanding human and nonhuman behavior and determining how best to harness the power of the collective in the service of human and animal flourishing.