Our scientific strategy begins with a preliminary, behavior-centric definition of intelligence.
Behavior is intelligent if it is adaptable, general, cost-effective, goal-directed, and can be performed in the real world.
Behavior is adaptable if it achieves its goal under varying initial conditions. Behavior is general if it can achieve a variety of similar, structurally related goals. Behavior is cost-effective if it balances cost (physical, computational, metabolic, etc.) and benefit. Behavior is goal-directed if it contributes to achieving a goal (survival, mating, etc.). These four criteria are not binary, but can be satisfied to a varying degree, with a higher degree of fulfillment being associated with higher degree of intelligence.
By real world we mean that behavior must be adaptable, general, and cost-effective in the face of real-world physics, as opposed to (1) a “gamified” world (as in chess, Go, Atari games), (2) an engineered world (factory floors with part feeders, etc.), or (3) a simulated world (with partial physics, assumptions, and specific boundary conditions).
Intelligent behavior can be performed by a single agent or by a collective. SCIoI will investigate individual intelligence and social intelligence of intelligent individuals interacting with each other, as well as collective intelligence, where intelligence is attributed to the collective rather than to its constituents.
SCIoI assumes that the key to intelligent behavior is a reduction of the combinatorial complexity associated with high-dimensional state spaces and a reduction of the detrimental effects of uncertainty. These reductions are achieved by leveraging priors (also referred to as bias, prior information, knowledge, heuristics, etc.). The most effective priors pertain to integrated intelligent behavior, i.e. the relationship between functional components of intelligence, not to the functional components themselves. This links our understanding of intelligence back to SCIoI’s point of departure: the non-decomposability of intelligence. Priors can be computational (explicit computation) or embodied (morphological computation) .
Our preliminary definition emphasizes behavior. It enumerates properties of behavior that can be tested and quantified. The focus on behavior is a first step toward reflecting non-decomposability in our scientific strategy, as we do not refer to any specific functional components. What matters is the resulting behavior, irrespective of the functional components (and their interactions) required to generate that behavior. As a result, the definition includes the entire behavioral phenomenon of intelligence, irrespective of possibly diverging views about its functional components.
Our preliminary definition is also a first step to avoid the pitfalls of Moravec’s paradox, as we require behavior to be general and to be performed in the real world. This excludes the demonstration of narrow skills, such as playing board games.
It is helpful to compare our definition to a somewhat aged, yet still widely cited and accepted working definition of intelligence from the literature :
[Intelligence] […] involves the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly, and learn from experience. It is not merely book learning, a narrow academic skill, or test-taking smarts. Rather it reflects a broader and deeper capability for comprehending our surroundings - “catching on”, “making sense” of things, or “figuring out” what to do.
This working definition provides a non-exhaustive list of skills that are indications of intelligence. It also lists some skills that are not sufficient. Interestingly, it already reflects an understanding of intelligence that excludes a chess-playing computer from being intelligent. Still, it does not provide a clear criterion, but captures an intuition about intelligence.
 Mainstream science on intelligence: An editorial with 52 signatories, history and bibliography - Gottfredson, Linda. Intelligence Vol. 24-1, pp. 13-23, 1997