An anonymous reader quotes a report by The Independent: Artificial intelligence at the human level is close to being finally achieved, says a leading researcher from Google’s DeepMind AI division. Dr Nanda de Freitas said the game was “over” in a long-standing drive to implement artificial general intelligence (AGI) after DeepMind introduced an artificial intelligence system capable of performing a wide range of complex tasks, from composing blocks to writing poetry. The new DeepMind Gato intelligence, described as a “general agent,” should simply be expanded to create an AI capable of competing with human intelligence, Dr de Freitas said.
Responding to an article written in The Next Web, which claimed that “people will never reach AGI”, the head of research DeepMind wrote that in his opinion, such a result is inevitable. “Now it all depends on the scale! The game is over!” he wrote on Twitter. “We want these models to be bigger, safer, more efficient in calculations, faster in sampling, smarter memory, more modalities, innovative data, online / offline … Solving these problems is what will give AGI ».
Asked by machine learning researcher Alex Dimikas how far he thought AI Gato was to pass the real Turing test, a measure of computer intelligence that requires a person to be unable to distinguish a machine from another, Dr de Freitas said: far away. […] Responding to additional questions from artificial intelligence researchers on Twitter, Dr de Freitas said that “safety is paramount” in developing the AGI. “This is probably the biggest problem we face,” he wrote. “Everyone should think about it. The lack of enough variety also worries me a lot.” DeepMind describes Gato in a blog post: “The agent we call Gato works as a multimodal, multitasking and multi-mode general policy. The same network with the same weights can play Atari, subtitle images, chat, fold blocks with a real robot hand and more, making context-based decisions output text, joint torques, button presses or other tokens.