The question of general artificial intelligence (GAI) has been on everyone’s lips for several years, but what if we have actually already reached this stage without really realizing it? Through recent analyses and reflections paving the way for new definitions of intelligence, we dive into the intricacies of a technology that could very well already be equipping our current conversational systems, without our knowledge.
Redefining artificial intelligence
The definition of GAI is not a simple exercise, as it varies according to experts and perspectives. In an opinion piece published in the scientific magazine Nature, researchers from the University of California propose the idea that artificial intelligence should include humans within its scope. Consequently, they contradict the notion of perfect or universal intelligence, which could not apply to any species, including humanity. This opens the door to debates about large language models (LLMs) and their ability to achieve performance comparable to experts in various disciplines.
The achievements of the Turing test
The famous Turing test, designed by Alan Turing in the 1950s, has long been a benchmark for evaluating machine intelligence. Researchers note that, under current conditions, systems such as ChatGPT pass this test remarkably effectively, to the point that some users perceive them as human. Far from being a mere tool, this intelligence could present itself as a reflection of some of our own mental capabilities, leading to a reevaluation of our understanding of intelligence.
The limits of human perception
According to studies conducted, it appears that our perception of intelligence may be too centered on humans and their own capabilities. Researchers defending the idea that we have changed our conception of intelligence highlight that even LLMs, which seem to play the role of a “stochastic parrot,” are capable of solving complex problems and transferring skills. This contradicts the idea that these intelligences would be mere imitators devoid of true competence.
The challenges of embodied intelligence
The question of embodying intelligence in physical devices is also a point of debate. Despite the absence of a body, language models demonstrate their ability to solve logical issues, erasing the argument that an intelligence must necessarily be anchored in a physical form. With advances in robotics and the emergence of hardware AI, these aspects are increasingly less of an obstacle to the evolution of artificial intelligence.
The hallucinations of artificial intelligences
The hallucinations of chatbots, these false memories or cognitive biases that can alter the reliability of responses, are another facet to explore. Although progress has been made in reducing these occurrences, the persistence of this phenomenon raises questions about the truthfulness and adaptability of these systems. For example, OpenAI reported that even with advanced models like GPT-5, about one in ten responses contains hallucinations. This highlights the complexity of the challenges that AI designers face.
Towards a new understanding of intelligence
Ultimately, researchers propose a reconsideration of our vision of intelligence. If we have already reached a point where artificial intelligence acts autonomously and coherently in various contexts, does it not invite us to expand our understanding of what it means to be “intelligent”? In this sense, the voices advocating for a shift in perspective emphasize that our anthropocentrism could obscure the recognition of forms of intelligence emerging beyond our usual reach. This could also explain why certain personalities, such as Mark Zuckerberg, place greater emphasis on the notion of superintelligence.
To understand the magnitude of these developments, it is appropriate to explore other sources of information and reflections, such as those available on related themes: the advantages of local artificial intelligence compared to giants like ChatGPT, the issues of Ripple’s blockchain facing security questions, or even the analysis of investments made and their actual impact on companies three years after the emergence of ChatGPT.
