The Work-equivalent AGI
What would it mean to achieve AGI in the context of all-remote knowledge work
The explosive growth of AI over the past 6-12 months has brought the issue of the Artificial General Intelligence (AGI) back to the forefront of everyone’s concern. From top researchers, to a host of doomer armchair pundits, the issue of when AI systems will achieve some kind of human-parity in terms of their general cognitive skills has resurfaced with the vengeance. This is, of course, an old question, one that has been on the mind of everyone who works in this general field ever since it became a standalone intellectual discipline. For the longest time Turing test served as the definitive test of this issue, but it could arguably be said that with the recent advancement with intelligent chatbot we are now in the post-Turing regime. However, most of us will concede that we are still not at the point where we can confidently say that AGI is among us.
So what would make one conclude that we have reached AGI now? I don’t think we’ll get a single conclusive definition before AGI actually arrives (circular reasoning, I know), but one place that I think AGI will immediately make the single biggest impact will be in the area of knowledge jobs. AI systems are already able to do a lot of tasks that entry-level workers have been doing for a while. Some compared these AI systems to interns - they can get a lot of simple tasks done, but still require monitoring and supervision. So I would like to propose a work-equivalent AGI definition:
A work-equivalent AGI is an AGI that you can not distinguish from a general work colleague that you interact with on a regular basis.
In other words, say you are a knowledge worker and you work remotely for some company. Most of the time your interact with your colleagues, or if you are a manager with subordinates, using a variety of tools: email, Slack, Jira, GitHub, etc. You communicate, set agenda and require various deliverables, most of which are in some form of digital content. One day there is a new teammate called Bob. Bob seems affable and competent, and you start working on a variety of projects with him, in addition to your other colleagues. And you do this for a few weeks, and in that span you have no reason to doubt that Bob is anything but another human coworker. It doesn’t mean that Bob is necessarily perfect with everything he does, but he will certainly get a passing grade on his annual performance review. Once we get AGIs that can be like this Bob, that’s when I think we can meaningfully start talking about the real arrival of AGI for all practical purposes.