Principles of Intelligence

All Principles

Multi-timescale computation

Adaptability to environmental change is a hallmark of intelligence. A mechanistic prerequisite for such adaptation is the presence of multiple time scales of computation to detect, track, and act upon changes in the environment. Changes occur across timescales, and so, the timescales of computation must match them.

Intelligent beings adapt to changes in the environment, but also to variations in the timescales at which the environment changes. A natural example of this is given by the way vision works in an intelligent system, and by how the system controls selective attention and moving gaze in dynamic scenes. The eye operates on a very fast timescale. Its quick movements help our brain get a general picture of what is currently contained in the scene, and this happens within a very short timescale. On a longer timescale, however, the brain also keeps track of the scenes or parts thereof it has already looked at previously and uses memory to avoid going back to the same salient or noticeable thing over and over. The brain also keeps track of the changes in the environment, adapting to new situations and making use of different scales of time to efficiently analyze the scene.

Another example of this is brought to us by the findings from SCIoI project 41, which describes the peculiar behavior of a fish collective living in a Mexican pond. When a bird approaches, the intelligent system, in this case the swarm itself, goes into a state of excitability (known as criticality) which brings the fish to collectively dive down, forming a very quick wave that startles the predator and normally discourages the attack. However, after this initial quick startle effect, the wave continues in a different form for minutes, discouraging the bird from further attacks. The swarm thus proves to be able to operate on different timescales as an adaptation to its environment.

Sulphur mollies respond to the kingfisher’s attack by forming a first quick wave followed by a set of waves, presumably to disorient the bird

What would a non-intelligent system do instead?

For better understanding, it often makes sense to compare the intelligent systems described above with a possible non-intelligent counterpart. For example, an engineering system such as an ordinary security camera could be used as a good comparison for the human vision system. The camera continuously records every part of the scene on a single timescale (e.g., recording 30 frames per second), without differentiating between motions, whether or not suspicious, and moments of stillness. It takes intelligent interpretation, by a human or a computational model, to analyze the scene and discern the parts that need attention. In our human visual system, this interpretation is based on the multiple timescales described above.
(As a minor addition to this point: examples of an intelligent cameras do exist. Event cameras, for instance –see SCIoI project 36– are built to only record “events,” i.e. the moments of movement that require attention)

A more in-depth look

In the simplest case, at least two timescales—a fast and a slow one—must be present. In order to be able to detect changes in the environment, the sampling frequency of the fast timescale should be significantly higher than the environmental rate of change. Integration of the environmental dynamics then provides (quasi-stationary) statistics on the environmental signal. It allows the detection of changes and adaptation to them on a slower timescale specific to the ecological niche. The slower time scale should be slow enough to filter out random variations, but also fast enough, i.e., in the order of the relevant rate of change of environment, to allow for successful adaptation.

Due to the fundamental nature of this principle, it is a necessary precondition for intelligent behavior in non-stationary environments with abundant examples in biology. They include the fast scale of individual neuronal firing and the slow time scale of spike-time dependent synaptic plasticity, or the presence of a whole hierarchy of time scales for information processing in the brain, and in animal behavior. It has also been identified as a core challenge for the reinforcement learning of artificial agents in non-stationary environments.

Related SCIoI projects

Project 1: Object-level scene descriptions and attention in visual search
Project 27: Speed-accuracy tradeoffs in collective estimation
Project 36: Active tracking using bioinspired event-based vision
Project 57: Internal and External Visual Information Sampling Using Eye-Movements and Overt Attention in Dynamic Scenes

Research

An overview of our scientific work

See our Research Projects