Demis Hassabis, CEO of Google Deep Mind assures that Gemini, the artificial intelligence system they are developing, will be even more revolutionary than ChatGPT. This division of Google is dedicated to work in areas such as artificial intelligence, robotics or neurosciences.
The development of Gemini could be the search engine company’s answer to the competition in the race for artificial intelligence. After the appearance of ChatGPT, Bard, Google’s AI, was launched and began to incorporate tools of this type in its applications. However, today the popularity of ChatGPT is still very high and Bard has not yet finished installing.
What are the features of Gemini, Google’s AI system?
The interesting thing about Gemini is that it is based on AlphaGo techniques. It is an artificial intelligence program that defeated a champion player of Go, the popular board game, in 2016.
According to Hassabis, Gemini will be more capable than ChatGPT, as it will combine AlphaGo techniques to give it new capabilities. Some of them will be the ability to plan and the ability to solve problems. Gemini is a language model very similar to GPT-4 which is the model on which ChatGPT is based. According to Hassabis, Gemini is a combination of the strengths of AlphaGo-like systems with the capabilities of large model language systems.
One thing to note about AlphaGo is that it is based on a technique called reinforcement learning. Such learning includes solving difficult problems and making decisions in board games (such as Go) or video games. Learning consists of making repeated attempts, receiving feedback on your performance. Another learning method used is called tree search. Through this method, the AI can explore and remember possible moves on the game boards. Unlike OpenAI AI, Gemini can learn from the environment and not just from text.
Reinforcement learning is a technique that Deep Mind has pioneered and has vast experience in. This will allow Gemini to have interesting capabilities for an artificial intelligence that have not been seen so far.
What are the risks of a development like Google’s Gemini?
The development of a type of artificial intelligence such as Gemini also entails dealing with as yet unknown risks. In response to calls from experts for a pause in AI development, Hassabis argues that the potential benefits of AI mean that the technology should be further developed.
Deep Mind has been studying the potential risks of artificial intelligence, and there is even an in-house AI security group. In addition, Hassabis participated in the declaration signed by several IA experts, in which they warn that this technology may one day pose a risk such as a pandemic or nuclear war.