Schloss Dagstuhl by Monet by Dall-E |

At Dagstuhl earlier this month, I hung out for a little bit with the participants of the other seminar, Knowledge Graphs and their Role in the Knowledge Engineering of the 21st Century. Knowledge graphs are what you would expect them to be, nodes are objects like "Berlin" and "Germany" with directed edges with labels like "capital". Think of having knowledge graphs of hundreds of millions of nodes and how that could help answer queries about the world. These secondary workshops are shorter and focus on creating a new vision, in this case how to maximize the importance of knowledge graphs in an increasing ML-focused world.

Perhaps we need such a visioning seminar for complexity. While we often get lost in the mathematical questions and techniques in our field, computational complexity is designed to understand the difficulty of solving various problems. Machine learning and advances in optimization should be changing that conversation. If you imagine a world where P = NP (and I did exactly that in chapter 2 of my 2013 book) much of what you consider is starting to happen anyway. ML does fail to break cryptography but then again, isn't this the best of all possible worlds?

Look at what Scott Aaronson said back in 2006.

If P=NP, then the world would be a profoundly different place than we usually assume it to be. There would be no special value in “creative leaps,” no fundamental gap between solving a problem and recognizing the solution once it’s found. Everyone who could appreciate a symphony would be Mozart; everyone who could follow a step-by-step argument would be Gauss; everyone who could recognize a good investment strategy would be Warren Buffett.

If I can be a Monet, can Mozart be far behind? ML trading by some hedge funds are beating Warren Buffett but remember if everyone trades perfectly, no one beats the average. Gauss is going to be trickier but it's coming. There's a reason Scott is spending a year at OpenAI to understand "what, if anything, can computational complexity contribute to a principled understanding of how to get an AI to do what we want and not do what we don’t want".