The astonishing ability of the new AI tools to increase productivity relies on being able to ‘learn’ by mining pools of data, then detecting and projecting responses based on statistical outcomes of that mining.
The next step, Generative AI, Generative Artificial Intelligence, is the point at which the artificial systems can reason, much as we do. This happens by making ‘neurological’ connections between apparently disconnected data, depth of domain knowledge and experience, breadth of more general knowledge that provides a ‘thinking canvas’ and context. These add up to instinctive responses we sometimes describe as pure ‘gut feel’.
There is however, a middle point.
‘Deep mind’ is a research unit now owned by Google. Their models evolved as AlphaGo and subsequently AlphaZero. These models cracked the barrier that seemed uncrackable, the ‘4-minute mile’ of computing. By beating the best humans at the complex game of Go, it demonstrated the ability of an algorithm to replicate in some form, the neural networks we have in our brains. In short, it can learn from its own experience, not reliant on outside data.
Crossing this Rubicon opens whole new territories to be explored.
It is in effect a ‘rolling probability’ calculation, each step using an estimation of the outcome of the previous calculation to deliver an adjusted outcome, in an ongoing process.
This is how we learn: from our experience.
As a kid I remember my younger cousin crawling towards a campfire surrounded by rocks. The immediate response of most was to grab him to prevent him getting burnt. However, my aunt stopped us, pointing out he would not be badly hurt by the mildly heated rocks surrounding the fire. However, when he touched a heated rock, it would create a memory-response loop that connected the fire to a modest hurt, thus ensuring he would automatically adjust his behaviour, and not go near another fire.
That incident stuck in my memory, and it reflects the way these AI tools are evolving rapidly towards ‘thinking’.
The dystopian view is that such developments over a few decades will see the machines take over. I prefer to think that we humans will find a way, as we always have, to overcome such threats. I guess my great grandchildren might know the right answer.
The header was created with help from DALL-E in about 3 minutes using a short series of prompts.
E&OE: A few hours after posting this post, I stumbled across this post on Medium that might bring forward the passing of the Turing test by a machine back into my lifetime. It records the evolution and current state of Googles 1.5 Pro tool, claiming it is to Current ChatGPT4 what a Model T is to a Ferrari.
The pace of change is astonishing, logarithmic, which makes it hard to comprehend by normal people..
.
