A Google researcher says artificial general intelligence is inevitable and it’s now just a matter of scale. As often with the topic, it seems to come down to definitions.
To put things simply, we already have plenty of examples of limited artificial intelligence with a computer or system performing a specific task (such as flying a drone or playing a game) with some degree of judgment and learning. At the other extreme, nobody really expects a robot with human-level abilities to learn and adjust to virtually any task, with the only limitations being physical.
Most definitions of artificial general intelligence fall somewhere in between, with one suggested threshold being AI that can learn a new task without any training input.
The Next Web’s Tristan Greene recently wrote an article arguing that the limitations of recent AI developments such as DeepMind’s Gato project “make it seem like AGI won’t be happening in our lifetimes.
That prompted a Twitter thread by Doctor Nando de Freitas of Google’s DeepMind, who said “It’s all about scale now! The Game is Over!” He went on to argue that the fundamentals are in place for AGI and the barriers now are more to do with speed and efficiency.
Greene disagrees, arguing that current AI systems simply aren’t capable of consistent results and that simply scaling up won’t fix that limitation.