What Is Natural Intelligence?

What Prions, Thermos Bottles, and Deep Neural Networks Have in Common

Joe: You know what the greatest example of artificial intelligence is?

Moe: No, tell me.

Joe: It is the thermos bottle.

Moe: Yeah, How's that?

Joe: You put icy stuff in it and it keeps it cold, you put boiling stuff in it and it keeps it hot. The amazing thing is: how does it know?

I recently searched my library of Artificial Intelligence (AI) and cognitive science books for a good definition of intelligence. Most of the sources did not bother to define intelligence. The ones that did displayed a level of anthropocentrism and vagueness that shocked me. In 1994, the Wall Street Journal asked 52 experts to define intelligence. This is what the committee came up with:

Intelligence is a very general mental capability that 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.

Besides a vagueness designed by a committee to offend nobody, you know it is a dubious definition if it includes idioms placed in quotes. How does one know when their robot is "making sense" of things or “catching on”? This viewpoint is anthropocentric; it discounts the intelligence of animals that cannot engage in reading and test-taking.

I would love to know what these 52 experts mean when they claim that “Intelligence is a very general mental capability…”. Does that mean that one algorithm accounts for all cognition? Or does it mean that our cognition can solve any problem? You’d think we would have figured out intelligence by now if it was a single algorithm. Computers use unique algorithms to beat us at chess, so the latter interpretation is incorrect. Natural intelligence is the opposite of 'general'; it is a kluge of many interconnected pieces that evolved over several billion years.

Artificial General Intelligence, or AGI, is a term used to describe AI that works more like humans. Since human cognition is a natural kluge, you will never find me referring to AGI. I prefer the term Natural Intelligence or NI. The ‘natural’ in NI refers to nature's algorithms and mechanisms, not a particular implementation.

Sara Shettleworth’s book, Cognition, Evolution, and Behaviors (2), gave me the best definition of cognition:

Cognition refers to the mechanisms by which animals acquire, process, store, and act on information about the environment. These include perception, learning, memory, and decision making.

Her definition embraces a biocentric view of cognition to the exclusion of everything that is not an animal. However, her definition is really just a list of attributes. It does not answer why cognition exists, nor does it provide any clue of how cognition works. So, I have created my own definition of Natural Intelligence (NI): 

Natural intelligence is an autonomous agent’s strategy for survival. It adapts to change through (1) the evolution of genetically determined behaviors and (2) learning. Learning is a continuous process that involves using sensed data and memory to make plans and improve outcomes.

This definition captures a wide spectrum of intelligent process and behavior. It assumes a biocentric view of intelligence. Rather than focusing on human intelligence as an exemplar of intelligence, it applies equally to all intelligence. Before we can hope to understand ourselves, we need to first extract shared features of NI from those features unique to humans. Only then we can bootstrap an understanding of human intelligence from simpler model organisms.

An intelligent agent refers to anything that conforms to this definition of NI. This includes bacteria, honey bees, mice, humans, some plants, some robots, and perhaps extraterrestrial life. Based on our definition of NI, things that qualify as intelligent agents—or not—might surprise you.

Thermos Bottles

This post begins with a joke about AI and thermos bottles. Thermos bottles are not naturally intelligent, as they cannot sense or follow a plan, despite being a great example of AI. That a thermos bottle keeps hot liquids hot (you could call this memory) is a consequence of its properties of insulation, not a decision or plan. Likewise, compasses are not intelligent because they ‘know’ how to point north. Both “behaviors” are consequences of physical properties.

Viruses and Prions

Viruses are not alive. They are complex assemblies of proteins, nucleic acids, lipids, and carbohydrates, but they remain inert until they enter a living cell. Without living cells, viruses cannot multiply.

Prions are infectious proteins and are also not alive. Like viruses, they require a living host to multiply, but unlike viruses, prions have no RNA. They cause neurodegenerative diseases such as Mad Cow disease. Scientists have implicated prions in Alzheimer’s and Parkinson’s disease. Unlike other infectious diseases, prions can remain active after boiling and disinfectant.

Are viruses and prions naturally intelligent? Both seem hellbent on survival. Viruses mutate their RNA strands, producing virus variants that adapt to changing conditions. However, viruses and prions do not sense or choose (implement a plan) anymore than an acid “senses” proteins or “decides” to dissolve a protein. Also, they are not autonomous — they require a living host. So, by my definition, they are not naturally intelligent. They are clever and dangerous molecules, but they are not naturally intelligent.

Plants

Several plants show simple reflexive behaviors. A reflexive behavior is like throwing a switch and a light comes on. There is no memory, no brain, no adaptation beyond evolved genetics. I used to believe that plants were solely reflexive and so lacked natural intelligence. But no longer.

In 1871, Samuel Butler wrote:

There is a kind of plant that eats organic food with its flowers: when a fly settles upon the blossom, the petals close upon it and hold it fast till the plant has absorbed the insect into its system; but they will close on nothing but what is good to eat; of a drop of rain or a piece of stick they will take no notice. Curious! that so unconscious a thing should have such a keen eye to its own interest. If this is unconsciousness, where is the use of consciousness?

The response of Butler’s Venus Flytrap (Dionaea muscipula) to a fly is a brainless, innate response. The Flytrap has small hairs inside its maw-like leaves. When a fly (or stick) activates at least two of those hair sensors within 20 seconds, the leaves will close. That Dionaea remembers when its hairs were last touched shows memory that drives an adaptive action. Dionaea is naturally intelligent by my definition (3)

The Immune System

The function of the immune system is to protect us from disease and infection. Natural immunity works by first recognizing foreign cells (sensing). It then creates antibodies (memory and action) that counteract or kill targeted viruses and bacteria in the future.

I disqualify the immune system as a naturally intelligent agent because it lacks autonomy. It depends on the rest of the body to provide creation, nutrition and waste removal. Also, its purpose is not to survive but to sacrifice itself in service to the body it protects. At any rate, the immune system is a pretty remarkable adaptive sub-system.

Deep Neural Networks (DNNs)

A DNN cannot qualify as naturally intelligent for several reasons. First, it fails as NI unless it can revise its own memory to improve future outcomes. Most DNNs deployed today do not. They are fancy classifiers and are certainly useful, but they rarely learn from mistakes once deployed. A DNN configured for reinforcement learning might qualify on this score, but it would represent a very primitive example.

The second reason that DNNs fail as NI is because they are not autonomous agents. Mind you, autonomy is a spectrum that ranges from interdependence to wholly independent. But DNNs entirely depend on the intervention of intelligent designers. As Dr. Alison Gopnik, Berkeley psychology professor, says, “We call it ‘artificial intelligence,’ but a better name might be ‘extracting statistical patterns from large data sets’”. DNNs cannot recognize anything without the data set collected by an intelligent agent. ChatGPT-4 is a generative DNN from OpenAI that has passed the Uniform Bar Examination in the top 10th percentile. As impressive as that it, it is possible because a human fed answers to it beforehand. ChatGPT-4 is autocomplete-on-steroids; it cannot apply the concepts of law to a unique case in a creative or original way.

DNNs also fail our definition of natural intelligence because they do not sense or adapt to their environment continuously. Most animals sense their environment several times each second and adapt. Some (honey bees) sense and adapt at a higher rate than others (humans).

Finally, DNNs lack a plan to act on. They are very good at doing one thing: playing chess, recognizing speech, or driving a car. But are they ready to raise a child or manage a workforce? I don't think so.

---

1 - Downloaded from http://www1.udel.edu/educ/gottfredson/reprints/1994WSJmainstream.pdf on 12/18/2021.

2 - Shettleworth, Sara J. Cognition, evolution, and behavior. Oxford university press, 2009.

3 - Calvo Garzón, P., & Keijzer, F. (2011). Plants: Adaptive behavior, root-brains, and minimal cognition. Adaptive Behavior, 19(3), 155–171. https://doi.org/10.1177/1059712311409446

Chamovitz, Daniel (2017). What a plant knows : a field guide to the senses (First revised paperback ed.). ISBN 9780374537128.