AI and nature - what we can learn from ants, octopuses and mushrooms

Today's AI development is strongly influenced by the human imagination. Ever since the 1950's SciFi novels, the ultimate goal seems to be the human-like robot, holding fluent conversations. Oh, and helping in the household of course.

The Jetsons, anyone?

Based on this human-centric view, our own cognitive abilities, such as logical thinking, problem solving or language comprehension, are often the focus.

But what if AI development would seek inspiration beyond the human intelligence, but elsewhere in nature? In nature, there are numerous fascinating examples of intelligence that differs greatly from ours. Maybe it's just arrogance not to look?

So let's totally try, but before we can take a look, let us find a suitable definition for "intelligence". It's quite a futile exercise, but maybe you can accept the following definition for the scope of this blog entry:

Intelligence is the ability to solve problems, even those seen for the first time.

Now let's look at examples of nunhuman intelligence, which show strong problem solving skills, but can't play Go or capture invoices.

Nonhuman intelligence - Examples from the animal and plant world

Swarm Intelligence

The classical example of swarm intelligence are ant colonies. The demonstrate collective problem solving in which individual ants only follow simple rules, but master complex tasks such as foraging or building nests or even bridges through their cooperation. However, the individual ant isn't a brainiac per se (sorry, ant!)

The message? If you have the wrong focus, and look at just one (daft) node of a larger network, you might arrive at the wrong conclusions. Such a decentralized approach to problem solving can be used in AI development, to achieve flexibility and very resilient applications.

While ants act together to form a single "brain", octopi have both....

Distributed brains in Octopuses

Octopuses (octopi?) have a distributed neural network. Each of their tentacles is able to not only collect vast amounts of sensory input, but also capable of processing it and acting upon.

This means, that the "central" brain has a much lower load, and only receives summaries of what happened.

This decentralized intelligence allows them to find surprisingly creative solutions to problems, such as opening screw caps, using tools or even recognizing faces.

Fungal networks (mycelia)

The mycelia of fungi act like an underground communications and transport network. It is capable to efficiently distribute nutrients and even signals, usually in a forest ecosystem. Fungi react to changes and can flexibly adjust their resource allocation - all without a central brain.

Did you know? Even though their structure seems very simple, fungi are capable to solve mazes!

Pros and Cons of using Natural Intelligence as an Example

The idea of imitating natural intelligence seems to offer clear advantages.

• Efficiency: Decentralized systems could be more resilient and adaptable, as we've seen in the exampes of swarm intelligence and fungi networks.
• Creativity: Looking outside the "box" of human-like intelligence allows to develop innovative solutions that go far beyond human ways of thinking.
• Resource conservation: Natural systems are often extremely efficient in their resource consumption, which could encourage the development of energy and resource-efficient AIs. For example, an adult human brain uses ~20Wh of power, while a small data center can easily consume 20kWh.

However, there are also challenges:

• Loss of control: Decentralized systems are more difficult to predict and control. This could become a problem in safety-critical applications. For the nerds out there...our AI systems could show nondeterministic behaviour, which even its creators would not understand. Check out my previous artical on "magic" if you want to hear my thoughts about possible consequences of such a development.
• Ethical considerations: Nonhuman intelligence act instinctively (thus: fast), but without any ethical considerations. Let's just run away from this lion, even if he's a nice guy. This "simplified" approach could be problematic if AI is to make decisions in human contexts....the classic example is the selfdriving car, which must "decide" to crash or not.
• Specialization: Natural intelligence is often specialized in certain tasks. The AI systems we're trying to build today however are intended to be generic. We are looking for systems which can move between different areas of application....which is, well.... human.

Conclusion: The future of AI - A balance between humans and nature

Nature seems to offer numerous exciting examples for AI development that we can learn from, if we would only take a look.

Swarm intelligence, decentralized brains or the efficiency of something simple like a fungal network show that intelligence has many faces, not all of them human. An approach to AI development that is based both on human and natural intelligence could achieve an optimal balance between (energy) efficiency, creativity and ethical aspects.

And how about you? Do you feel our current approach to AI development is good as-is, or shall we rather have ChatGPKraken? I'm curious to hear your human and nonhuman views :-)