I agree. Especially when taking into account the exponential increase in AI computing power, and the fact that AI doesn’t even need real-life examples to learn, and can instead do reinforcement learning based solely on simulations of the real world, there’s no reason to expect otherwise. The difficult step will be getting to there, and even knowing when we are there.
Probably within a year or two after toddler-level, it’ll become superhuman. I would consider that a blink. Heck, I’d consider 5 years a blink, but I don’t think it’ll take that long after toddler-level. The optimist in me thinks it might achieve superhuman just a few months after toddler. It won’t take just 5 minutes, however.
Plus, a truly massive level of interest and funding will come pouring in once we have toddler-level. Let’s just say a few governments will take a keen interest.
It’s a lot of fun to speculate!
Videos of the talk are here:
Q&A: https://youtu.be/VUSrwXZJ5Yg .
Dammit Ben, I have a bunch of work I needed to do on the train journey home, and now I’m going to have to watch these aren’t I? Looking forward to it
The questions from the audience were good and spurred some useful responses from me, I think… it was not a huge crowd but they were smart and knowledgeable, a lot of Josef Urban’s PhD students…
I ended up giving a talk there because my son Zarathustra is doing his PhD at that university in Josef Urban’s group, focused on applications of machine learning to automated theorem proving… I think there are lots of insights from the automated theorem proving world that will help with inference meta-learning in OpenCog (and have been discussing this with Nil Geisweiller and Alexey Potapov and others on the SingularityNET team…). This point comes up here and there in the talk I gave, though it was a high level talk and didn’t dig too much into details…
Awesome! Updated the first post.
Excellent presentation Dr. Goertzel. As always your keen insight and ability to connect diverse threads inspires!
The last I heard, DNA works in complex ways that are difficult to map out.
My guess is that we discover that we are different because of how the genes interact with each other than the genes in of themselves.
Note that pet dogs are around toddler level. They do require constant care and attention…
I think agi will not be born like a baby, it will awaken like an adult. You know how we could measure its state of arousal? The range of data it takes in…