ChatGPT - in the Zone of Awe

This is a framework I used to explain the evolution of AI in my book. It's not a scientific framework, just a useful conceptual construct. I looked at the way we perceive AI in how it can do 'human activities' as well as how it can do things that humans can't. Examples of the former would be: recognize handwriting, decipher expressions, or drive a car, and examples of the latter would be: place millions of transistors on a chip, or solve for protein folding problems. On the other axis is the issue of whether the AI learns in a supervised or unsupervised manner. Supervised learning is usually with a clear objective in mind, and involves labelled data. It's used typically for classification, for example. Unsupervised learning involves unlabelled data and the AI is left to discover the hidden patterns. Clustering is one of the outcomes of unsupervised learning. The picture below is over simplified, but the point I'm getting at is the need to avoid seeing human performance as the only arbiter for AI. It's just one benchmark.

Talking of humans, I often compare AI to a child. Possessing an incredible competence, but as yet untrained and relatively early in its development. And just as we wouldn't judge a human being based on what they can do at age 10, we should not judge AI at 15 years, or ChatGPT at 2 months. That's why there are 4 zones here in this diagram.?

The zone of condescension is the classic 'even my 8 year old can do that' zone, such as recognizing traffic signals. We often forget what a feat of science it is to get to this point. The zone of indulgence is even more complex, but still within the realm of human performance - such as recognising a cyclist on the road, including their hand signals, and likelihood of swerving. The zone of pride gets us to an area which humans can't reach but is something we can train the AI to do. Relatively easier technologically, this would include 360 degree vision which can be sorted by cameras placed around the car. The Zone of Awe is where it gets truly interesting because it gets us into complex areas where where humans can't go - for example safe high speed driving with multiple vehicles on the road. The zone of awe is actually unbounded and who knows where it might take us!

Language based AI models started in the zone of condescension but ChatGPT has leapfrogged into the zone of awe and will keep driving upwards and outwards. And even though it can't do maths (as of now) or gets some facts wrong, have you thought about what it might do when it's 5 years old instead of the current 3 months?