Trend Tuesday: Automating Iterative Improvements for Devilishly Hard Problems
Art by @basilonmypizza: https://lnkd.in/eF8FkWzN - https://basilhefti.ch/
Has it happened to you that your navigation system proposed a route which you ignored for a shortcut you had in mind? Sometimes you were right. Sometimes the system had seen a constraint you had not.
Beyond the private errands, optimizing routes is known as the traveling salesman problem: given a number of points to drive to, the challenge is to find the route with the shortest overall travelled distance or - better yet - travel time. These are very practical problems because time spent in travel is often less productive than time spent at the destination. In other words, cutting travel distance and/or time to the bare minimum makes economic sense (it creates value).
The traveling salesman problem has been known for a long time. Solving small instances is manageable. Trouble starts in real life: customers might have different priorities. Interventions different urgencies. And driving there with the wrong equipment or know-how also does also no good. Once you add these constraints, the problem becomes really hard.
That changes when the autoresearch approach is used which we discussed previously (https://lnkd.in/eJttigqK). Google Deepmind had a similar idea: use an iterative approach of automated code optimizations to find better traveling routes (called AlphaEvolve). Yes, the approach relies heavily on AI. But the result is an improved, practical routing heuristic. Applied to routing decisions, the improvement reported in the case is 10%. In a recent project we did, we observed comparable improvements.
We are quite familiar with iterations. The logic is familiar from evolution: variation, selection, retention. And craftsmen refine their product many times until it reaches the required maturity.
In this way, AI helps us find solutions to devilishly hard problems - be it in material science (think batteries), medicine (think cancer treatment), or - like here - logistics. And while AI is the means to find the solution, it is not necessarily part of the final implementation.
Where does AI create more value in your organization: in the final interface, or in the process of finding a better solution?
• Art: https://lnkd.in/eF8FkWzN https://basilhefti.ch/
