AI: the evolution of precision, or how we all went down the wrong path

4 min read
artificial neurons
Image source: 123RF

In the process of creating any truly new technology, there are two sequential tasks: creating a theoretical concept and implementing the idea in practice. In this article, I will name the main reason why, despite a well-developed theoretical concept, in practice, we are still very far from creating strong AI.

Avoid mistakes at all costs

The first example of the practical implementation of the prototype of AI was in control systems, a device for automatic guidance of anti-aircraft guns based on radar data, created by British mathematicians during the Second World War. The desire to avoid mistakes in such systems was fully justified, but as often happens, the first steps at the start turned out to be decisive for the entire further path.

The evolution of precision

The next major step towards the practical implementation of the concept of AI was the invention by Frank Rosenblatt of a simple but effective AI modeling tool: the perceptron. At the same time, even though many consider the perceptron a variant of a neural network, and therefore an AI system, Rosenblatt himself saw in his invention only a research tool, and not a machine capable of thinking.

In general, systems based on connectionism (and the perceptron was the most successful model in this direction) were extremely close to getting on the right track in creating strong AI. But the desire to achieve accuracy and the fear of an erroneous answer defeated intuition and eventually led to the creation of an error backpropagation system – which finally turned the perceptron into a calculator.

Since the early 1990s, an agent-based method based on the use of intelligent (rational) agents has appeared. According to this concept, intelligence is a complex of mathematical calculations using conditional symbols (a kind of planning), designed to achieve the goals set for an intelligent machine. Symbolic computing is essentially computer algebra or working with mathematical formulas as a sequence of conventional symbols.

Trying to create AI, we dived deeper and deeper into calculations and working with algorithms.

The era of super calculators

A significant increase in computing power (of recent decades) has ended the so-called “Winter in AI Development” period. Modern neural networks, relying on high-performance technical complexes, work with the help of a total enumeration (processing) of information. Due to this, modern AI systems can group data according to very small features, and then sequentially compare these groups with the parameters necessary for analysis.

This approach creates the illusion of learning. For example, the Alpha Go Zero chess program, without even knowing the rules of the game, managed to surpass the leaders of chess skill among people in 21 days of independent work with a database of previous games played.

At the same time, some scientists (Noam Chomsky described this position very clearly) understand that the vector of our development in the creation of AI is moving in the wrong direction precisely because of the complete dependence on statistical methods.

In the language of allegories – whether we like it or not, but trying to create AI, we are still “directing anti-aircraft guns.”

The root of the problem or why it is important to remember that form is not equal to content

The fact that the brain consists of neurons that look like a network does not mean at all that our neocortex works like a technical neural network. I will say more – our brain is not at all afraid to make mistakes and does not seek to show accuracy.

Why?

Because in fact our brain does not work like a calculator and does not do any calculations at all. This is especially clear when we compare the energy balance of the brain during sleep and during complex mental tasks (energy consumption almost does not change, so it is impossible to lose weight from mental work).

Our learning process is about memorizing the exact answer, not working with analytics and parameters. Our intellect juggles with the correct answers known to it in advance, and the gigantic volume and universality of this predetermined data gives rise to the emergence effect – which outwardly looks like deep analytical thinking. That is why each of us has been a naive and inept child for many years. To learn how to do something new, we must repeat long and hard.

The work of our intellect can be compared to a flywheel that slowly spins up during our childhood and does not stop for a minute until death. It is for this reason that we think even in our sleep.

Our brain makes decisions not based on data processing but based on the experience of previous decisions. Emotions and experiences that make us so different from machines are very difficult to describe in the language of algorithms, but easy to form based on a simple choice between “like or dislike”.

Remember our thesis at the beginning of the article – “theoretical model and practical implementation.” We can’t build AI because we’ve built the theoretical model wrong. At the very beginning of the journey, we unfortunately turned the wrong way, forgetting that the form is not equal to the content.

The right way or how to create a strong AI

Not calculations, but the experience of millions of decisions creates intelligence. And don’t let the word experience mislead you, it’s not the experience that is the result of the work of algorithms – it’s just knowing the answer (not necessarily the right one).

A small child does not analyze the parameters of the parents’ behavior – he simply repeats and remembers what and how to do. Therefore, AI can become intellectually like us only with the help of one process – continuous streaming learning (and in fact copying) in direct contact with a living human brain.

Therefore, the brain-computer interface is something without which it is impossible to create a strong AI. But the software block for the new individual AI will be quite unusual.

If several accelerating bypass technologies can be applied, it is likely that the first working prototype of individual artificial intelligence will appear quite soon. In any case, this is exactly what I am currently working hard on. Stay tuned for more…

Previous articleWhy ChatGPT is not a threat to Google Search
Next articleWT2 Edge: A very convenient AI-powered translator earbud
I am a doctor and an independent researcher. For almost 30 years I have been creating the concept of individual artificial intelligence based on the integration of the human brain and a computer system into a single functional complex united using a scanning-type brain-computer interface. The concept of individual artificial intelligence developed by me is based on the hypothesis of the existence of a dual system of initiation of nerve impulses in the synapses of the human neocortex and the dynamic concept of quantum spin in a new relativistic or high-speed model of our three-dimensional space. To be honest, I've been doing this all my life. This is not just a new invention or a new scientific idea. In fact, this is a new reality that is already on our doorstep and will soon change the life of every person.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.