Current AI systems are hampered by several major limitations, which prevent them from fully reproducing the complexity of human thought, as we discussed in this article. However, this does not prevent the leaders of the sector from continuing their efforts to one day develop THE model that will surpass all others and maybe one day win the IAG trophy (Artificial General Intelligence). A highly advanced form of artificial intelligence that aims to create machines capable of understanding, learning, and performing a variety of intellectual tasks in a human-like manner.
On December 20, a discreet announcement from OpenAI shook the scientific community: their new o3 model achieved an 85% ARC-AGI test score (Algorithmic Reasoning Corpus-Artificial General Intelligence), far surpassing the previous best performance of 55% achieved by other AI systems, and rivaling average human capabilities.
Learning efficiency, a new AI paradigm
The ARC-AGI test differs fundamentally from conventional assessments of artificial intelligence. At the heart of its operation lies the concept of “ sampling efficiency » or the ability of a system to deduce general rules from a minimum number of examples. Where ChatGPT requires millions of samples to establish probabilistic language models, o3 demonstrates a remarkable ability to extrapolate general principles from a few specific cases.
This evaluation confronts the AI with visual puzzles in the form of grids, where the system must identify transformation rules based on just three examples (see below). Concretely, the AI observes simple geometric patterns which transform according to precise logical rules: a square can move along a line, one shape can absorb another, or elements can combine to create new configurations . To succeed, the system must not only understand these transformations, but also extrapolate the underlying logic to apply it to new cases.
This methodology is similar to classic Intelligence Quotient tests, but more specifically assesses the ability to adapt quickly to new situations. The major difference lies in the emphasis placed on the system’s ability to formulate coherent hypotheses from a very limited number of observations. A fundamental skill of natural intelligence that AIs traditionally struggle to reproduce.
The cognitive mechanisms of o3
o3’s technical innovation is based on its ability to identify what researchers call ” weak rules »: subtle and often implicit correlations between data which, although not constituting universal laws, allow the model to make predictions or make decisions. It’s as if o3 was able to detect “clues” in the data, allowing it to better understand and interpret the world around you. François Chollet, AI researcher, proposes an enlightening hypothesis to explain this: o3 would explore different “ thought chains », analyzing various reasoning sequences before selecting the most relevant.
This approach is reminiscent of the methodology of AlphaGo, the system that excelled at the game of Go in 2016, but with a fundamental difference: o3 must generate its own evaluation heuristics for problems he has never encountered before. The system would thus demonstrate a form artificial intuition in the selection of solutions, a bit like our brain organ would do.
The implications of this technological breakthrough have yet to be fully explored. Limited access to the system – reserved for a few laboratories and institutions specializing in AI security – does not yet allow an exhaustive assessment of its real capabilities. The distribution of its performances, the frequency of its failures and successes, as well as the extent of its adaptability will still require analysis to be certain of its potential.
If the capabilities of o3 are confirmed during its public deployment, we could witness a radical transformation of artificial intelligence. This development will require the development of new evaluation criteria for the AGI and in-depth reflection on the modalities of its governance. Such powerful AI could have a profound impact on society, transforming the way we produce, work and live. It is therefore impossible to ignore the ethical and social implications of its development and to put in place sufficiently robust safeguards to avoid abuses. Otherwise, although representing a notable technical advance, its impact on our daily lives could remain limitedespecially in view of the actual costs (not to mention the energy costs) represented by the training of such a model and the changes in regulations.
- OpenAI designed an AI model capable of near-human performance on a test measuring cognitive adaptability.
- This model, called o3, is distinguished by its ability to deduce general rules from very few examples.
- If these results are confirmed, o3 could mark a key step towards truly versatile artificial intelligence.
