We are now into week 3 of the AI bubble, a week I have been dreading because I have to say something intelligent about AGI. I don't work for a frontier AI lab or an academic organization that has insight into the frontier labs. From a distance, I have no way of knowing whether AGI is two days away, two decades away or two centuries away. If that wasn't bad enough, I find AGI doomerism - a la Gary Marcus - uninteresting and I find AGI eschatology (both of the singularity and the end of days kind) uninteresting.
Maybe it's because I find AGI uninteresting.
There, I said it. Seems strange to say that the prized goal of the most prized technology of all time is uninteresting, but I am going to stick to my assessment.
OK champ.
Forget about its intellectual appeal. What about its civilization threatening possibilities? Shouldn't everyone be interested in a technology that could soon turn us into paperclips?
Geoffrey Hinton, an AI pioneer, argues there is a 10-20% chance that the technology will end in human extinction. A former colleague, Yoshua Bengio, puts the risk at the high end of that range. Nate Soares and Eliezer Yudkowsky, two of hundreds of people working in AI who signed an open letter in 2023 warning of its perils, will soon publish a book about superintelligence entitled "If Anyone Builds It, Everyone Dies". In private, grandees from big AI labs express similar qualms, albeit not always so apocalyptically.
If a wandering asteroid had a one in five chance of hitting the earth, I would be dropping everything else and helping us figure out how to shift the odds in our favor.
Why not AGI?
I have many reasons for not getting excited about AGI, but at its core the problem I have with it is: it's too anthropocentric; it assumes humans are the measure of all things. I guess it's a big deal when computers can write better novels than we do, but at the end of the day, I am interested in:
What can humans do (with or without AI) that we haven't done before.
What new spaces for imagination and exploration does AI open up for us?
Anyways, AGI is in the news a lot and will continue to do so for a while. What does it even mean?
Today's Daily Planet starts with the following definition:
AGI is an AI that can match or exceed the cognitive versatility and proficiency of a well-educated adult.
Want to learn more? Read the paper!