THE’artificial intelligence is making great strides, but how is it really different from ours? human intelligence? Surprisingly, it is not clear how different the heuristics of our mind are from the statistical functioning of generative AI models. The artificial neural structure and that of our brain present many similarities and some decisive difference. What undoubtedly differentiates us is the provision of a body selected by natural evolution, driven to reproduce and survive in a physical world in which there are complex challenges that require the senses to be addressed.
It is often said that artificial intelligenceunlike the human one, is capable of performing complex calculations and analyze huge amounts of data. In fact, this is also a characteristic (and for now above all) of the human brain. Of course, if we try to compete on mathematical calculations, we will not get very far, but think of the calculations that our brain must perform to catch a ball in flight from baseball: while trying to figure out the trajectory of the ball to know where it will arrive he devotes himself to very complex things adjustments of every part of the body involved in the movement which will help you be in the right place at the right time. To do this, it is not enough to “guess” where the ball will land, but our brain needs monitor every change which happens second after second, to adjust the prediction and recalibrate nerve signals and muscle movements. A very significant calculation.
During the 1950s and the following decades, there was the common opinion from the experts of artificial intelligence that it would have been very easy to reproduce with a machine everything that had to do with the body and with movementwhile a lot more difficult it would have been imitate human reasoning. In fact, if you think about it, science fiction very easily imagined a near future world in which humans would coexist with humanoid robotssometimes not very intelligent but skilled in movements. On the other hand, the real challenge for researchers seemed to be the creation of “thinking machines”, that they could deceive their interlocutors into believing they were talking to a human, or that they could prove virtuous in what makes humans “special”: chess, go, writing poetry or novels, that is, in reasoning and in the creativity. The Moravec’s paradox reverses the perspective, and highlights this contradiction according to which, when put to the test, the movement and sensation skills human are the most difficult human skills to simulate and reproduce artificially, while high-level activities (typically mental skills associated with reasoning) are better formalized through algorithms, and require less computing power.
We generally associate intelligence with skill in the arts, sciences, chess, or mathematics. We rarely exclaim, “That’s a smart person” when we watch a female diver win a gold medal at the Olympics. Interestingly, however, the underlying neural mechanism that makes a person a great mathematician or a great swimmer is pretty much the same: they both need specialize a large amount of neurons in one specific task. When we dedicate ourselves with passion and dedication to an activity, our brain applies its calculation capabilities to the activity undertaken, lighting up areas of the brain which you dedicate to logical reasoning in the first case, to perception and movement in the other. What happens next is relatively similar: training specific skills the populations of neurons that are activated begin to specialize more and more, they recruit brain matter further to increase the total computing capacity and connections increase between neurons. Here intelligence, seen from this perspective, becomes a vague, elusive concept. The constants of almost every extraordinary ability are constancy And dedication.
For AIs it’s a little different, their consistency depends on theavailable energywhich is usually continuous and regular, and dedication is just a matter of timeIt’s not about motivation, it’s not about effort, it’s just algorithms that carry forward logical sequences which, at their end, activate other logical sequences, until the desired output is generated.
To put it very simply, the artificial neural networks They are particular types of mathematical models inspiredin their function and structure, to neurons in our brain. It is the same architecture that we find under generative AI models such as ChatGPT, Claude and Llama, the most popular software at the moment. These architectures are made up of knotslike the cell nuclei of our neurons, and from links between the nodes, just like the axons and dendrites that connect our brain cells together. The similarities don’t end there here: both artificial neurons and brain neurons have a activation threshold which, if exceeded by the incoming electric charge, generates an outgoing electric discharge. The result is that only signals that exceed a certain threshold they manage to spread across networksFrom here on the analogies become less close, especially with regard to one of the fundamental characteristics of our brain, which makes it flexible and adaptable to the various problems of the real world: the neuroplasticityThe connections between our neurons are dynamiccan undergo a process of pruning when little used, or can create new ones when a certain population of neurons close to each other is particularly active during the same period of time (a process called synaptogenesis). It goes without saying that, without a external interventionone artificial neural network cannot change its structure. However, computer engineers have found a way to ensure that, despite the inability to change its structure, the network can learn from feedback and the databases on which it is trained. In fact, nodes can change their weight, or the numerical value that conditions their strength, so that a greater node strength will correspond to a greater influence of the node on subsequent nodes. It is a bit like there was an information traffic manager that decided, depending on how correct the output is, which roads should become highways and which should be limited to country lanes.
The reality is that AI philosophers disagree in affirming that the two intelligences are truly different in their basic functioning, perhaps precisely because define intelligence it’s already a philosophical problem unsolvedor because we are missing the passage that connects the “dance of the neurons” at the demonstrations of the mind. There are a couple of things we can be sure of, however: we are equipped with a organic body which provides an infinite amount of different types of feedback to our brain, as many feedbacks as there are senses (also counting balance, interoception – the perception of internal organs – and proprioception – the perception of one’s own body-) and that this work of processing the external world filtered by our perception is a very complex result of natural selectionwhich has modified our sensory organs, has shaped our body and, consequently, the possibilities of action and perception. We do not have an engineer behind us who rationally decides the architecture of our hardware or our software, but we “self-construct” ourselves by following the commands written in our genetic heritagewhich occasionally mixes through sexual reproduction or changes due to transcription errors or mutations. This means that Our intelligence is shaped by the selection of billions of years of life on Earthin which it has been and continues to be put to the test by the dangers present on our planet. The strategy of our genes and our development has been to select a flexible organwhich can adapt to most contexts to be able to challenge any problem that arises. Artificial intelligence is not able to do this for now, and resigns itself to performing (very well) only the tasks for which it was designed.
