A New AI Based On Insect Behavior

While my research in AI has most recently been in Causal AI, something has been on the back burner. This is an AI-based on an ant colony.

While colony optimization has been successfully applied to networking, the idea of an organism made up of organisms has not been used in AI.

For example, an ant colony consists of four types: queen, male, soldier, and worker. Each colony has one queen who mates with a male once. Most of the colony consists of worker ants who feed the related masses. Soldier ants guard the colony from foreign invaders, be they non-related ants or predators.

While the colony functions like a living organism, it consists of creatures that learn from each other and are themselves living insects. In terms of a software version of these insects, they are more complex than the individual neuron-type models of AI infancy. However, the program to make a software ant is easier for a human programmer to conceptualize.

When talking about general AI and even with the standard deep learning models, we see a significant need for guardrails or safeties that prevent catastrophes. For example, while an intelligent chat application can give erroneous answers, an ant cannot decide to become vegetarian.

Software ants can be created with multiple types of roles based on what the designers want for the colony. Each software ant is not limited to a specific geographic area and could be dispersed across the globe or decentralized. Since cross-colonization is not allowed, these digital ants can only learn from other ants within their family group, with ultimate direction coming from a pre-built program.

The most significant applications that come to mind revolve around complex problem-solving. Let me know what you think, and please share any research you are aware of in this area.