Generalist AI Gato has skills that no human can match

You will also be interested


[EN VIDÉO] How can the relevance of artificial intelligence be defined?
Artificial intelligence (AI), which is increasingly present in our world, enables machines to imitate a form of real intelligence. But how do you define it?

L’Artificial intelligence forte, also known as Generalized Artificial Intelligence (AGI), is the holy grail of AI researchers. She would be able to rival human intelligence or even be conscious. DeepMind, Google’s sister company, just released one Article about Gatoa new generalist AI that could be the forerunner of a strong AI.

The majority of current AIs are task specific and trained with neural networks for a specific purpose like creating deepfakes Or from play chess. with Gato, deep mind goes the opposite way and has created a single AI capable of performing many very different tasks.

According to the authors ” The same network with the same weights can play the Atari console, identify the content of images, chat, stack blocks with a real arm robotics and more “. The AI ​​decides what form to give its answers based on the context. In total, being able to complete 604 tasks with a single model is a real feat.

AI that beats the experts, but not always

deep mind uses a transformer-type neural network typically used in speech processing. Gato has been trained on a large number of Databasecontaining images, text, and agent experience in real-world or simulated environments.

The problem is that the AI ​​doesn’t always perform these tasks correctly. For example, answers may be incorrect during a discussion. Gato specifically pointed out that Marseille was the capital of France… deep mind states that the AI ​​would do better than an expert half the time on three quarters of the tasks (450 out of 604). That would put us at a success rate of just over a third.

The generalist model will evolve as technology advances

However, there is a good reason why DeepMind is working on an all-purpose system capable of handling such a variety of tasks. This choice follows the conclusions of several specialists in this field. Rich Sutton, one of the founders of reinforcement learning, said that general-purpose computational methods are far more efficient. According to him, most researchers assume that the available computing power will not evolve. On the contrary, the constant increase in available power would be a much more important factor in the development of AI than the increase in human knowledge in the field.

In response to an article on the site The next web Rather pessimistic about Gato’s ability to evolve into strong AI, DeepMind researcher Nando de Feitas explained the company’s goals on Twitter. According to him, the road to strong AI is now a matter of scale. He claims that it is no longer necessary to work on the philosophy of the symbols that the big networks could easily create and manipulate. From now on, we need to create bigger, more efficient, faster models with smarter memories, and research needs to go in that direction.

Gato is actually less complex than many other specialized AIs, with 1.2 billion parameters compared to the 170 billion of the GPT-3 system. The explanation is that they wish Gato could control a robotic arm in real time. In their opinion, with a more complex system, the AI ​​would have been up to any task.

Get the Mag Futura for free now by subscribing to our subscriptions!

Did you know that you can access Futura ad-free through our subscriptions?
At the moment you can discover this advantage with our special offer: subscribe to the “I take part in the life of Futura” (for a minimum of 3 months) and receive the Mag Futura at home* (a value of €19)!

I take advantage of the offer now!

*Mag Futura will be sent after the third month of registration.

Interested in what you just read?

Leave a Comment