This post is also available in .
As kids, we often treat our toys as if they were intelligent. A teddy “wants” to be tucked in, a toy car “tries” to win. Later, some of us reserve intelligence only for humans and animals, while some extend it to plants and trees. Some think robots and chatbots are intelligent, and some consider forests, companies, and cities to be intelligent things. Some think the entire planet (“Gaia”) is a super-intelligent being.
An intelligent system can simply be defined as a system that wants to be somewhere, or in other words, it shows goal-directed behaviour. For instance, a white-blood-cell wants to find and digest a harmful bacterium. However, when it comes to measuring intelligence in a system, we rely on a cognitive-science-coloured lens. We assume that intelligence looks like human cognition that involves neurons and brains that have memory, can perceive, and can predict things.
These measurements are good at defining intelligence in cognitive beings, much better than the trivial IQ-tests. However, outside the cognitive world (white blood cells, active particles, plants, etc) they start to collapse.
Collective intelligence
A more basic assumption about intelligence is that all intelligence is collective. This means intelligence is a property that emerges when many components that themselves are intelligent combine with each other in a particular way. Now, this may sound circular, like taking a loan to pay off another loan, but the scaling of the goal is the key here. Recent research states that when components that themselves have goals in the world they live in combine with each other, such that their combination aims to achieve another, larger goal that they themselves cannot sense. For example, none of the cells in the hand “knows” that the hand must have four fingers and a thumb, yet upon cell division, the simple cells whose aim is to maintain their internal homeostasis stop dividing when four fingers and a thumb are developed. Thus, the individual goals of the components get scaled into a bigger goal that they themselves don’t know, but their combination achieves.
Diverse intelligence
Once we are comfortable with the “collective intelligence” assumption, we can let go of our cognitive-science flavoured assumption that the units/components of an intelligent system need to be neurons or sub-parts of a biological brain. The individuals themselves are replaceable as long as their combination achieves the larger goal. This is the common thesis of the upcoming field of “Diverse-Intelligence.” This line of research aims at defining intelligence in a general way, beyond the cognitive abilities of the intelligent system. For instance, intelligent systems not only move to a favoured position in a problem space but also keep remapping the space itself based on the input they get from their environment. Now this can be seen not only in human brains but also in large-language models in the space where they represent the words.
Although we have come a long way from trivial IQ tests, we are still far away from converging on a universal functional-definition of intelligence. Maybe the key is to keep relaxing our assumptions that are based on our brains and embrace intelligence as a complex emergent property of many interacting parts across different scales.
Author: Siddharth
Buddy and Editor: Helena
Translator and Editor Translation: Rick
About The Author
