A conflict on what basis?
Reading the press, we have the impression that we have already entered a war between humans and AI. The monster is still on a leash. But won’t its designers hide it rather than destroy it? Or will it be sold in secret to ill-intentioned autocrats, who will develop it for their exclusive benefit? Such technology cannot disappear or be controlled. Like any dangerous human invention it can only be countermeasured, in a perpetual race for power, under a Damocletian sword: Will anyone use it? This is not the first sword. The planet is nuclearized and could blow up at any moment. Successful thinking takes getting used to.
The first thought is that this is a conflict without a basis. I do not mean by this that it would be without reason but that there is no well-understood definition of intelligence. Officially at least. Unofficially, for readers of Stratium, intelligence is a perfectly defined notion, quantifiable ontologically and not using tools as questionable as IQ tests. Let’s start by reviewing this fundamental notion before deciding whether to get hysterical about AIs.
Intelligence is a height of integration
“The” intelligence does not exist any more than “the” volume. You find small, medium and large. It is a measure of scale according to a criterion. Simple criterion for volume: an amount of three-dimensional space. More complicated for intelligence: it is organizational efficiency. Very subjective. Let’s try to clarify. Let’s start from what is organized, namely the parts of space for a volume, the concepts for intelligence. A simple addition suffices to organize the parts of a volume, if they do not overlap. While the concepts weave into a complex hierarchy. Pyramid of representations, from the most basic —a rod lit on the retina— to the most sophisticated —for example ethics with thinkers and stories connected to this notion.
The more a concept synthesizes a large set of criteria, the higher it is in the pyramid. Quantifying intelligence involves measuring the height of the stack, not the simple addition of parts. All brains are uniformly active and process a large amount of data. It is not this quantity that makes intelligence, but their deep integration.
Many species to plant in a field
The remarkable characteristics of our intelligence have two joint origins: 1) The variety of criteria offered by the presence of five senses, our technological instruments and social mimicry. 2) The high stacking of analysis stages that our vast neural field allows. But there is a brake on this elevation: the organization of concepts becomes more difficult as their relationships become less obvious. The association of pixels into lines occurs naturally in the visual centers, while the association of social indicators in economic policy is a challenge for the lay brain.
Continuing the elevation of our conceptual pyramid thus requires methodological tools. How to organize together concepts that seem foreign to each other? As much as the parts of space are easy to assemble because they share the same nature, as many parts each endowed with a specific quality seem impossible to merge. The mind generally gets out of it by returning to the constitution of things, to look for common points in them. It makes models and tests them. Failures and successes. Intelligence never stops being refined.
One Observer to control them all
Due to genetics and differences in background, intelligences are markedly contrasted from one individual to another. The variety of environments specializes our intelligences. The influx of data in a domain, or its scarcity, stimulates or diminishes the intelligence concerned. Hence a question: does one intelligence develop at the expense of others? This seems to be attested to by the litany of geniuses, brains famous for their hyperspecialization, who have lost their footing in everyday life. Nietzsche, Camille Claudel, Maupassant, Lautréamont, Van Gogh, James Joyce, Hölderlin, Sade, Poe, Kafka, Strindberg, Virginia Woolf, Hemingway, John Forbes Nash, Morphy, Bobby Fisher… Show me someone who only thinks to one thing and I will show you his indigence in thinking the rest. But then are those who don’t think of anything genius generalists? 🙂
What most differentiates our intelligences is the observation of our personal method to extend it. Genetics and the environment sometimes create impressive potentials which remain fallow by default of method. It is self-observation that opens up intelligence and releases its full potential. Does this mean that we, like AI, would eventually need an evaluation to become fully intelligent?
How many floors in your intelligence pyramid?
The coding performed by neurons is very different from the traditional computer method. In the operations of your personal computer, data and results use the same ‘machine’ language. The result is translated into a programmed user language. In a brain, on the contrary, the neurons construct their own language, as the data advances through the hierarchical levels of the network. Each floor has its own code, based on that of the underlying floor and assembled in turn by the overlying one.
At every step, the code is data integration. Data and result are two inseparable sides of the same process. The integration gradually expands as the pyramid rises, as does the symbolic value of the excited networks. This is how a few neurons at the top of the pyramid support an elaborate thought —Cf the grandmother cell—, while they are traversed by the same stupid neural excitations as the simple receptors of the visual rods.
The machine looks a little too much like us
Your personal computer offers the beginning of similarity by the tiering of its code: electrical, binary, machine code, programming language, user interface OS. Five levels. There are many more in the brain, which explains why the biological organ accesses high-level consciousness and your computer does not. Importantly, each level of the brain organizes its own code; it self-creates its consciousness. We still dare not leave such latitude to our silicon assistants. What ‘artificial selection’ would weed out the duds? Are we capable of the necessary sapience?
This is indeed the source of the panic caused by AIs. With deep learning networks, AIs are now mimicking how the brain works. It is no longer five levels but an unlimited number that they can theoretically use. In practice, they are satisfied with a number still much lower than a human brain. Why this restriction? The increase in depth of AIs quickly comes up against prohibitive instability. The AI is going “crazy”. Note that the human brain is not immune to the problem, when it integrates incompatible events too quickly, or when drugs destroy its organization. The solution, for AI as for humans, is progressive learning, not mixing too quickly the most difficult data to organize together, those which already require a great height of view…
A moratorium on humans?
AIs will only become psychopathic and dangerous in the hands of humans who will make them that way, because they themselves are psychopathic and dangerous. The risks of AI refer to the fundamental freedom that humans demand, that of thinking and acting without any constraint.
Freedom is an explosive. Which transforms the world. Handle with the greatest care. But haven’t we already made uncontrollable stocks of them?