In November 2023, a mysterious artificial intelligence project by OpenAI was the subject of many rumors and speculation. According to the Reuters news agency, several of the company’s researchers wrote a letter to the board of directors warning them of a powerful artificial intelligence (AI) development. In their view, it could threaten humanity.
What is it all about?
It’s a project involving extremely advanced artificial intelligence algorithms, secretly developed by the company that created ChatGPT, OpenAI. The project is called Q*, or Q-Star. There are still many mysteries surrounding it, and details about it have not yet been made public. According to Reuters, Q-Star marks OpenAI’s attempt to cross the border into general artificial intelligence.
OpenAI defines general artificial intelligence as “autonomous systems that outperform humans in the most cost-effective tasks”. This term refers to a system capable of performing a wide variety of tasks with a level of intelligence comparable to that of humans. In theory, this could allow it to escape all human control. This is also known as artificial superintelligence, or strong AI.
This concept of artificial superintelligence has long remained a distant prospect, even an utopian dream. However, with the rumors that have surrounded the OpenAI project in recent months, this form of AI, capable of acting autonomously in imitation of human intelligence, could be nearer than we thought.
This kind of strong AI would enable the automatization of complex cognitive tasks. It could help solve complex social problems by analyzing large quantities of data to suggest innovative solutions. For instance, artificial superintelligence could contribute to better management of humanitarian crises or the fight against climate change.
An AI that surpasses human cognitive abilities?
This mysterious project seems to be focusing on advanced problem-solving methods, particularly in mathematics. By mastering mathematics, artificial intelligence comes very close to human intelligence. It could then decide to stop obeying humans, or even rebel against them…
The Q-Star project is a large language model (LLM). LLMs are AI models that generate text. They generate text by predicting the next word, and answers to the same question can vary. However, the Q-Star project is said to be able to solve simple mathematical problems that were not part of its training corpus. Some researchers see this as a crucial step towards the realization of general artificial intelligence. This is a major first in the field of AI, because in principle, LLMs are based on probabilistic operations. These models are not good at mathematics when only one correct answer is possible.
According to Andrew Rogoyski of the Institute for People-Centred AI at the University of Surrey, an LLM capable of solving mathematics problems would indeed be a breakthrough. He says that “the intrinsic ability of LLMs to do mathematics is a big step forward, enabling AI to offer a whole new range of analytical capabilities“.
How does it work?
Q-Star is based on three new training techniques.
The first is “thought tree reasoning“. This allows an AI model to produce several reasoning paths to arrive at a solution. The model then evaluates these paths to determine which leads to the best possible answer.
The second is reinforcement learning. According to the French Data Protection Authority (Commission nationale de l’informatique et des libertés – CNIL), this is a machine learning process that consists in learning the actions to be performed, based on experiments, so as to optimize a quantitative reward over time. This method improves the model’s performance without the need to continuously create new data during training.
The third technique is inspired by the success of AlphaGo, where an AI improves by playing Go against itself. It involves neural networks and self-learning. Open AI is said to be using similar approaches to improve its LLMs.
These methods could provide Q-Star with exceptional cognitive capabilities, representing a major advance in the field of AI.
Should we really fear artificial intelligence?
Finally, this Q-Star project is nothing other than a type of artificial intelligence that can solve simple mathematical problems. For the moment, the complete automatization of artificial intelligence is not for the immediate future. Machines still need people. AI is a long way from being able to imitate human intelligence. Algorithms do not have free will. They only do the tasks for which they have been programmed. Algorithms can only make decisions based on simple, frequent, and unambiguous cases. They do not have the capacity to reflect and always need human supervision.
Artificial intelligence that surpasses human intelligence remains science fiction. Humans are still in control and will be for a very long time. So there’s no need to be afraid of this new Q-Star project.
Julie FREIERMUTH
M2 Cyberjustice – Promotion 2023/2024
Sources :
https://www.silicon.fr/q-star-le-projet-qui-aurait-seme-la-zizanie-chez-openai-473712.html
https://www.lebigdata.fr/q-star-tout-savoir
https://www.forbes.fr/business/q-vers-une-intelligence-artificielle-generale-chez-openai/
https://medium.com/@InnovateForge/what-is-project-q-star-162c4b8767b5
