The Long, Uncertain Road to Artificial General Intelligence

Last month, DeepMind, a subsidiary of technology giant Alphabet, set Silicon Valley abuzz when it announced Gato, perhaps the most versatile artificial intelligence model in existence. Billed as a “generalist agent,” Gato can perform over 600 different tasks. It can drive a robot, caption images, identify objects in pictures, and more. It is probably the most advanced AI system on the planet that isn’t dedicated to a singular function. And, to some computing experts, it is evidence that the industry is on the verge of reaching a long-awaited, much-hyped milestone: Artificial General Intelligence.

Unlike ordinary AI, Artificial General Intelligence wouldn’t require giant troves of data to learn a task. Whereas ordinary artificial intelligence has to be pre-trained or programmed to solve a specific set of problems, a general intelligence can learn through intuition and experience.

An AGI would in theory be capable of learning anything that a human can, if given the same access to information. Basically, if you put an AGI on a chip and then put that chip into a robot, the robot could learn to play tennis the same way you or I do: by swinging a racket around and getting a feel for the game. That doesn’t necessarily mean the robot would be sentient or capable of cognition. It wouldn’t have thoughts or emotions, it’d just be really good at learning to do new tasks without human aid.

This would be huge for humanity. Think about everything you could accomplish if you had a machine with the intellectual capacity of a human and the loyalty of a trusted canine companion — a machine that could be physically adapted to suit any purpose. That’s the promise of AGI. It’s C-3PO without the emotions, Lt. Commander Data without the curiosity, and Rosey the Robot without the personality. In the hands of the right developers, it could epitomize the idea of human-centered AI.

But how close, really, is the dream of AGI? And does Gato actually move us closer to it?

For a certain group of scientists and developers (I’ll call this group the “Scaling-Uber-Alles” crowd, adopting a term coined by world-renowned AI expert Gary Marcus) Gato and similar systems based on transformer models of deep learning have already given us the blueprint for building AGI. Essentially, these transformers use humongous databases and billions or trillions of adjustable parameters to predict what will happen next in a sequence.

The Scaling-Uber-Alles crowd, which includes notable names such as OpenAI’s Ilya Sutskever and the University of Texas at Austin’s Alex Dimakis, believes that transformers will inevitably lead to AGI; all that remains is to make them bigger and faster.