Artificial intelligence sometimes seems remarkably clever. Still, we shouldn't forget what AI does at its core.
It doesn't understand the world the way humans do. It processes vast amounts of data and calculates from it which answers are most likely to fit. Put simply: AI doesn't search for truth, it searches for patterns.
To make AI better, more and more data is fed in and the models are optimized. The AI model keeps adjusting until its answers contain fewer and fewer errors.
In research, this relationship is often shown by plotting a model's errors against the computing power applied. The more computation, the smaller the error. You can see this in the graph above as well.
The dashed line represents a kind of AI scaling law. It describes how, with more computing power, AI makes fewer errors, yet the number of errors never drops below the dashed line.
AI can acquire an enormous amount of knowledge. If you know all the data and can think logically, you shouldn't really make any mistakes. That's why, to me, this dashed line is more than just a technical trend line. I see it as the comprehension horizon.
Patent Cockpit's customers don't work at common understanding. They stand at the frontline of their technologies. Often the whole point is to do something that isn't yet obvious, something that exceeds existing knowledge or has already been dismissed. That requires a genuine understanding of the situation. This is exactly where innovation arises.
AI can support this work. It can help to search faster, structure better, and decide on a more solid basis. But it doesn't replace actual progress.
Because progress doesn't come from the average. Progress happens where people and companies go beyond what can already be calculated.
< Back to all blog posts