DeepMind’s Gato Moves Toward Artificial General Intelligence

Alphabet-backed research lab DeepMind last week released a new AI system called Gato that takes a step toward artificial general intelligence, or AGI, technology that enables machines to undertake and learn the same tasks as humans. Described by DeepMind as a “general purpose AI, Gato’s debut coincides with LinkedIn co-founder Reid Hoffman and DeepMind co-founder Mustafa Suleyman securing $225 million to fund Inflection AI, an artificial intelligence company the pair launched earlier this year to simplify communication between humans and computers. Neither investors nor valuation were disclosed.

“Long relegated to the domain of science fiction, it’s been suggested that AGI would bring about systems with the ability to reason, plan, learn, represent knowledge and communicate in natural language. Not every expert is convinced that AGI is a realistic goal — or even possible,” reports TechCrunch, which characterizes Gato a move in that direction.

What puts Gato in the direction of artificial general intelligence is the diversity of the tasks and the training method. Gato was trained at 604 tasks, including conversing, playing Atari video games, captioning images and using a robot arm to stack blocks.

Key to multitask learning “is whether the tasks complement each other” as part of a unified intelligence, instead of separating types of tasks and teaching them to subnetworks, Allen Institute for AI research scientist Jack Hessel tells TechCrunch.

But Gato isn’t perfect. “In captioning pictures, Gato misgenders people. And the system correctly stacks blocks using a real-world robot only 60 percent of the time,” writes TechCrunch, noting that DeepMind claims that Gato can outperform a human expert on only about 450 of the 604 tasks.

TechCrunch reports that University of Alberta assistant professor of computing science Matthew Guzdial thinks “people saying it’s a major step towards AGI are overhyping it somewhat, as we’re still not at human intelligence and likely not to get there soon,” adding that while he finds many smaller, focused models more useful, “there’s definitely benefits to these general models in terms of their performance on tasks outside their training data.”

In March, when Inflection AI was launched, CNBC wrote based on an interview with Suleyman that the company “will look to develop a new suite of technologies that will eventually enable anyone to speak to a computer in plain language,” and the network explains that it’s unclear even “at this stage what the final products will look like, how much they will cost or who they’ll be targeted at.”

DeepMind’s Astounding New ‘Gato’ AI Makes Me Fear Humans Will Never Achieve AGI, TheNextWeb, 5/13/22
DeepMind’s ‘Gato’ Is Mediocre, So Why Did They Build It?, ZDNet, 5/14/22