Are we still talking about human intelligence in the age of generative AI?

What are we talking about? ” jeep (For “co-embedding predictive architecture”), relying on a static experimental architecture would be able to learn, reason, imagine, design, and plan actions, marking a real break with existing generative AI paradigms which, it is true, are mostly built from. A relatively basic structure even if it gives the illusion of the opposite.

Rather than focusing on the vast amounts of data and scripts generated or directed by humans, the JEPA model focuses on comparison of representations Abstract images, sounds, or text rather than collecting terabytes of data (as machine learning does). The idea is to understand and then replicate the work of the human brain which, every day, without realizing it, analyzes a lot of information simply by observing the world around it. Through this innovation, the errors inherent in generative AI will be greatly reduced.

when we talk today of generative artificial intelligenceeveryone thought chat Of course or even A poet. JEPA has nothing in common with ChatGPT or any similar conversational language model. With JEPA, many observers claim we have just passed an important milestone. However, should we be excited about this innovation? Is it as disruptive as it sounds? In fact, we consider that JEPA does not constitute a revolution in itself at the technological level. On the other hand, it is a major innovation in use cases that will not be exploited.

What can human intelligence become in the face of this technological tsunami? Is human intelligence threatened by innovations that are supposed to make our lives easier but could also enslave us?

But what do we mean by human intelligence?

Defining intelligence is an issue of certain complexity as one can find in the literature many articles dealing with the topic with as many variables as the authors (or almost). This complexity takes on another dimension when it comes to relating it to human intelligence. Thus, defining intelligence is not easy and remains a controversial topic. According to Paul Guillaume (1878-1962), “To say that someone is intelligent is to make a value judgment.”

We consider that a machine equipped with artificial intelligence, despite its complexity, cannot be considered “intelligent” but rather an interface based on algorithm-based technology that allows it to produce results. Which approaches a form of rationality. Can we ask a machine to explain the logic that led to the presented result? The answer is no.

It only implements the programming that man imagines and creates. the Artificial intelligence applications It will thus allow systems to ensure automatic learning (or machine learning) of situations allowing the development of increasingly sophisticated predictive models according to the amount of data being processed. So, like human intelligence that acquires knowledge and understanding of a situation over time, machine learning relies on inputs, such as training data or knowledge graphs, to understand the situational environment that the AI ​​has to deal with.

Our purpose is not to connect the different types of learning. But depending on the complexity, we can move towards a kind of supervised and unsupervised learning through reinforcement. In general, all systems that use AI rules learn without understanding and decide without understanding. Execute operations in a way Automatic without understanding its meaning, is in no way comparable to the definition of intelligence.

Has artificial intelligence a memory capacity It is critically necessary because of the necessary large amount of data(s) it has to ingest for learning and thus enhance the quality of machine learning. After training variable over time depending on the complexity of the subject to be treated and the volume of available data, the system using artificial intelligence will gain experience and show a certain logic and thus improve the quality of the predictive analytics that will follow. Are we talking about the intelligence of machines? Certainly not, but increasing the intelligence of the individuals who will use this AI.

Humans have an enormous intellectual capacity to generalize their learning In different areas while the machine is only enriched with the elements for which it was programmed. Let’s not forget that the development of these machines and their associated algorithms, in order to be effective, requires a certain amount of human intelligence to design, debug and optimize after trying these machines. Thus, this is a real difference between artificial intelligence that requires logic and humans whose developed intelligence does not purely require logic.

According to Olivier Hudy, intelligence lies in inhibition, that is, in our ability to inhibit intuition in order to adapt to the environment. In other words, being smart means using our brain in a useful and situational way. We share this approach which leads us to consider that an algorithm cannot be considered intelligent from the moment simple logic cannot fully understand the world and its ecological space.

Thus the machine is not superior to the man because he is still the master of the game, as the machine shows a superior capacity for data assimilation. its ability to Data processing He is also faster than a man. for him margins of error weaker. However, the algorithm trains on the provided data mechanically, and without any upstream control it can perform on the correctness of the information. Thus, the machine can also make errors in interpretation, similarity, or even emitting deviations. Only humans are able to correct the inference and interpret the data in some way Logic. How far can his knowledge take him?

Algorithmic programming does not allow the device to adapt to the future. As new concepts or situations emerge in our ecosystem, humans will adapt As always. The machine can only react according to Historical data She knows which will not correspond to the development of the situation. Furthermore, let’s not forget the fact that humans still create or control algorithms. If he makes mistakes, the machine will not detect them. All results given by machines are mathematically interpretable, even if this is sometimes limiting and can take time.

So we cannot speak decently about intelligence. Machines do not have consciousness and cannot adapt to future situations, which is a unique feature of mankind. Throw JabraMeta intends to be an inspiration to human thought. Jabra It is an attempt to reduce the inaccuracies and errors that are still noticeable using traditional generative AI. A step has just been taken. It now remains to be seen over time whether the achievements will align with Meta’s expectations.

Declaring that technology is intelligent, and that this intelligence is similar to our own, cannot but create confusion. We wanted to emphasize in this short article that People create technology, not the other way around. The qualities of the latter are strictly limited by the specifics of the training they have received. This is what the engineers determined. Thus, when we admire recent developments in generative AI, we are aware of the latest advances in human knowledge and its increasing mastery of new technologies. We can give special human traits to machines. These are just improvements on a design worn by particularly brilliant individuals.

Machines are not smart. They cannot explain or justify the algorithms that guide their recommendations. And therefore the limits of the latter are unknown to them. Moreover, machines learn and decide without understanding. The question of the meaning and concrete consequences of their proposals lies beyond the scope of their algorithm. Is it necessary here to recall the Rabelais proverb: Science without conscience is nothing but ruin for the soul »? Knowledge without reflexivity does not allow a person to progress. Finally, they are unable to generalize their learning outcomes to different domains and their ability to adapt to new situations is severely limited.

If machines are not intelligent, current discussions about their future place in our professional environment are biased. Nevertheless, these discussions make it possible to reaffirm the idiosyncrasies of human action that our common intelligence and unique. This intelligence allows us to question the interests and limitations of the reasoning we use. It questions our everyday routines and allows us to transcend the purely ritualistic nature of our behaviour. This intelligence leads us to conscientiously assess the consequences of our decisions. This evaluation guides the balance of our decisionsbetween economic rationality and ethical commitment. Finally, we have a special ability that allows us to find similarities, to build bridges, between new situations and those we have already encountered. This ability allows us to explore new areas and areas of knowledge.

Let’s make sure that it is not always by training an AI with more data that we will reach the level of human intelligence. Yes, machines will be more complex, but this in no way portends a world in which you will dominate. Artificial intelligence is and will remain at the service of human intelligence.

Article written by:

Pascal Montignon – Director of the Digital Research Chair, Data Science AND ARTIFICIAL INTELLIGENCE – OMNES EDUCATION

Eric Brown – Associate Professor – OMNES EDUCATION