Being afraid of a parrot?

ChatGPT is a remarkable technical achievement and if somebody is afraid: it’s a consequence of being used to life without imagination and innovation. It is not so much about “intelligence” but more about “big data” and automatization of everything. The emphasize is on “automatization”. There is a lot of repetition and routine in our life which can easily be delegated to machines. So: will we be able to increase our productivity by automating our lives?

In its heart life or existence on earth is a search problem. The search space is unimaginably large and dynamic. Forecasts of how the world will look like in 1 year or how innovations will develop are as reliable as a weather forecast. Tools like ChatGPT and successors are bound to this physical reality, too. Hence their technology is no larger threat than any other human being born today.

The singularity when it gets interesting: if the combined search power of machines like ChatGPT surpass the search power of humanity. Then there is a chance that they may find new knowledge faster than we do and outsmart us.

Not so fast: human beings are still many decades ahead of surviving physical reality. ChatGPT is all virtual. It has no idea how to maintain itself, how to multiply, how to get the chips and energy it needs. Therefore, all the talk about the “danger” of these tools is marketing hyperbole. The underlying big technical breakthrough of large language models is their super-large dimensionality. From a conceptual perspective the underlying algorithms are around for a long time already. But only now do we have achieved just the required dimensionality to make them useful in practice. Quantity sometimes creates another quality.

The next steps I expect: to make any progress in the physical world we need high-dimensional robots. 4-axis or 6-axis robots are kind of stone-age quality. We need 1000-axis plus 1000-sensor robots. Why is dimensionality so important? Because the real world is super-high dimensional and useful tools must have equivalent dimensionality. In the real world there are thousands and thousands of exceptions of rules of rules. ChatGPT and its vast parameter space seems to possess sufficient dimensionality in the field of language. Now we must transfer this capability to physical interactions, too.