Much is said about the capacity of artificial intelligence, with a bias that it is unlikely to surpass human abilities. In a recent article, IGN Brasil mentioned, for example, that AI systems are far from writing like a human.
On the other hand, as robots evolve, evidence shows that they have already managed to reach a level that reproduces a crucial ability of human intelligence: compositional generalization.
We are talking about the ability to learn the meaning of words and then apply them to other concepts by association.
This is an important capability of human brain. It is from this that we are able to abstract ideas and recognize objects by their shape, even if they are different, have different colors or are made of other materials.
Image: Reproduction
A surprising discovery
Cognitive scientists Jerry Fodor and Zenon Pylyshyn reported in the past that artificial neural networks would be capable of making these connections.
Since the 1980s, there has not been such an important advance in this area as now, with the increase in artificial intelligence.
Researchers from New York University and Pompeu Fabra, in Spain, have been focused for a few years on developing techniques to improve this capacity.
A study recently published in the scientific journal Nature shows that systems like the ChatGPTlaunched by OpenAI at the end of 2022, are capable of making compositional generalizations based on the so-called “Meta-learning for compositionality”.
Practice Tests
Tests carried out by scientists have shown that Artificial Intelligence is not only capable of matching human capacity, but also surpassing it in terms of cognition.
The testing process was based on practice, not learning. Basically, the system received a word and responded to the request to apply it in another context.
An example of this is the word “speak”, which can be associated with different situations, such as speaking a lot, speaking little, speaking softly or loudly. With evolution, the systems were also able to develop contexts such as “talking nonsense”, “talking nonsense”, “talking nonsense” and others.
The elaboration occurred in both the literal and figurative sense, which significantly expands the scope of the language of these tools.
Confirming this capacity for association appears to be a significant attraction, especially for the programming area.
Systems will be able to understand and respond to a wide variety of commands, even the most complex ones.
Support our work ❤️
If you enjoyed this article, consider leaving a tip to help us keep publishing great content.
