The Diverse Intelligences (DI) community has revealed remarkable capacities in many natural and artificial systems often referred to as “complex adaptive systems” (CAS). CAS are systems with diverse parts, whose interactions generate adaptive behavior. CAS exhibit intelligence at many spatial scales. For example, institutions are composed of humans, who are composed of tissues, which are composed of cells, etc. Intelligence also appears on a range of time scales — intelligent processes can be ephemeral, but they can also transcend the lifetime of their parts.

Rather than viewing intelligence as a property of a specific thing (a behaving organism, an artificial neural network, etc.), new research posits that intelligence is a property of the dynamical processes that allow a system to master its circumstances by solving a problem or completing a task. If true, which processes emerge — and how effective they are — will depend on the context. 

At UCLA, Jacob Foster, a founding co-Director of the Diverse Intelligences Summer Institute, leads a research project looking at two questions. One is around the idea of “compositionality”, i.e., what about the parts of a CAS make them easier to combine? The other concerns what structures of the subparts — or “mechanism design” — are best; i.e., how does the system put parts together so that their collective behavior produces desired effects? The goal of this research is to develop a new Framework for explaining intelligent behavior in diverse CAS. The supporting institution, the Santa Fe Institute, is the preeminent center of study of such systems. Team members bring a wide variety of disciplinary skills.

Human experience provides ample evidence of groups and organizations assembling only to produce massively unsatisfactory outcomes. But the fact that multi-level intelligences often do succeed (most notably in biological organisms like us) suggests that there must be underlying design principles for successful wholes. Moreover, because these relationships are in constant flux and each subprocess is itself dynamic, there must not only be a sort of blueprint but also an ongoing flow of control and feedback up and down the levels of complexity.

The team’s central idea is that the reciprocal flux of control up and down a system leads to building command structures that optimally deploy lower-levels of intelligence. In other words, they hold that well-adapted systems, including human institutions, novel AI, and many more, require a constant balancing of the flexibility provided by individual freedom with the efficiency of strong overall guidance. Top-down organization, heavy-handed central control, must balance with bottom-up flowering of novel and original structures.

The research will examine how semi-independent, adaptive processes achieve necessary coordination to function as an intelligent, higher-level whole. The team will address this question along three parallel work packages:

1. Experiments considering communities of humans that compete and cooperate to solve problems. 
2. Research drilling down how humans (or human-like AI) cope with tasks, and what their neural subprocesses are doing to contribute. 
3. An examination of non-human biology on the simplest level (“xenobots”) to ascertain how individual and collective problem-solving overlaps or collides.

Extensive outreach and community-building is planned. With this new Framework, the team hopes to inspire considerable new research; new architectures for artificial and collective intelligence, especially those that draw insight from natural intelligence in designing AI; and to encourage designs of organizations, institutions, and AI-human collaborations that are constitutive of, rather than corrosive to, human flourishing.