Considering another huge breakthrough by Google’s DeepMind AI team, we may be one step closer to human-level AI or artificial general intelligence.
DeepMind’s new Gato artificial intelligence system is one of the first multitasking AIs of its sort. It may use the same network to perform various tasks, such as playing Atari video games, stacking blocks with a robot arm, chatting with people, pressing buttons, and so on.
Nando de Freitas, DeepMind’s Research Director, claims that Gato, aka “generalist agent’ needs to ‘scale up’ to reach the next level of intelligence and become inseparable from a human.
”Gato AI needs to just be scaled up to create an AI,” Dr de Freitas said.
Even though most AI systems excel at specialised tasks like facial recognition or even road navigation, they lack the “general” intelligence that sets humans apart.
Humans may immediately respond to various situations by drawing on their prior experience or intuition. A baby, for example, may see a dog once and then recognise all other dogs as dogs, regardless of their form or shape.
However, machine vision AI must be trained hundreds of thousands of photos of dogs to recognise what a dog is. The ability to ‘generalise’ in this way is a crucial component of true artificial intelligence that passes the Turing test, which claims that a person cannot differentiate an AI from a human.
DeepMind’s Gato project aims to get us closer to this goal, but not everyone is convinced. DeepMind’s research director responded to these concerns by stating that he believes such an eventuality is unavoidable.
“It’s all about scale now! The Game is Over!” he wrote on Twitter.
“It’s all about making these models bigger, safer, compute efficient, faster at sampling, smarter memory, more modalities, innovative data, on/offline… Solving these challenges is what will deliver AGI.”
Yet, some AI experts have cautioned that the emergence of AGI could end in humanity’s destruction.
“A superintelligent system that surpasses biological intelligence could see humans replaced as the dominant life form on Earth,” Oxford University Professor Nick Bostrom warned.
One of the most disturbing questions about an AI system smarter than humans is how difficult it would be to turn it off.
On Twitter, Dr. de Freitas answered more inquiries from AI researchers by noting, “When building AGI, safety is paramount.” He wrote, “It’s perhaps the biggest challenge we face.”
“Everyone should be thinking about it. Lack of enough diversity also worries me a lot.”
While a coherent, ‘general’ AI may never be achieved, systems like Google’s Gato make AI much more powerful and capable and could play a significant role in the fate of future bots.