Why AI does need biology after all

Why AI does need biology
after all

02

Neuroscientists look down on the top-down approach to AI development and "are "in the tank" for "theirs", which is the second, bottom-up approach. But such a "study" faces a fundamental challenge — the brain not being a wing, how can we "take a peek" at the way it works? This gives rise to the problems of the primary and the secondary, of cause and effect, of emergentism, i.e., inconsistency between the properties of the parts and those of the whole.

In reality, the nervous system uses dozens or hundreds of different chemical signals between neurons. In addition to wired signal transmission, the nervous system also uses volume signaling.

07

06

Signals between neurons are chemical rather than electrical in nature. Chemical signal coding significantly expands the amount of transmitted information and makes the variability of possible structure, architecture, and methods quite different from what current AI has to offer.

We should study the way these very "neural networks and evolutionary computations that simulate intelligent behavior based on biological elements" manifest themselves rather than the way some unclearly working parts of the "whole" are organized.

03

Understanding how the information part of the synapse signal is encoded poses some challenges. According to what neuroscience currently knows, the neuron and its synapses are molecular machines that transmit and integrate chemical signals.

05

04

What does neuroscience say about the brain? The brain is a superdense system permeated with a network of connections. 100,000 to 1 million connections per neuron. The signal is transmitted from the body of the neuron to the synapse along the axon using the electrical structure and chemically between synapses using mediators and their combinations in synapses or, more specifically, synaptic vesicles. Mediators and their combinations form the information component (information processing "toolbox") of the brain.

08

Perhaps the brain's task is NOT to accumulate information (unnecessary facts) and transmit it instantly, like a computer, but to MAKE decisions by PROCESSING (consolidating, structuring) reality through primary PERCEPTION mechanisms.

Already at the single nerve cell level, such a phenomenon as heterochemistry gives the brain system unexpected — emergent — properties. In a biological neuron, the quality of "synaptic weights" — that is, weight considered as a synapse coefficient, is not essential. A living nerve cell is not a SUMMATOR of weights (unlike neurons in neural networks).

09

01

Currently, there exist two main approaches to the development of AI systems:

top-down, semiotic — creating expert systems, knowledge bases, and logic inference systems that emulate thinking, reasoning, speech, emotions, etc.
bottom-up, biological — studying neural networks and evolutionary computations that simulate intelligent behavior based on biological elements, as well as creating the corresponding computing systems, such as neuro- and biocomputers.

These principles serve as a basis for a perfectly working engineering solution. But claiming that AI copies the brain, intelligence, and thinking or, as Hinton says, "overtakes it" is erroneous.

Therefore, emergentism is possible at all levels of brain organization where heterochemistry is represented. The variety of incoming element values implies a great variability in the resulting pattern (structure) characterized with fundamentally new (emergent) properties.

10