Microsoft’s AI boss and Sam Altman disagree on what it takes to get to AGI

It depends on your definition of AGI, right? AGI isn’t the singularity. The singularity is an exponentially recursive self-improving system that very rapidly accelerates far beyond anything that might look like human intelligence. 

To me, AGI is a general-purpose learning system that can perform well across all human-level training environments. So, knowledge work, by the way, that includes physical labor. A lot of my skepticism has to do with the progress and the complexity of getting things done in robotics. But yes, I can well imagine that we have a system that can learn — without a great deal of handcrafted prior prompting — to perform well in a very wide range of environments. I think that is not necessarily going to be AGI, nor does that lead to the singularity, but it means that most human knowledge work in the next five to 10 years could likely be performed by one of the AI systems that we develop. And I think the reason why I shy away from the language around singularity or artificial superintelligence is because I think they’re very different things.

The challenge with AGI is that it’s become so dramatized that we sort of end up not focusing on the specific capabilities of what the system can do. And that’s what I care about with respect to building AI companions, getting them to be useful to you as a human, work for you as a human, be on your side, in your corner, and on your team. That’s my motivation and that’s what I have control and influence over to try and create systems that are accountable and useful to humans rather than pursuing the theoretical super intelligence quest.