Microsoft researchers have created an artificial intelligence-based system that learned how to get the maximum score on the addictive 1980s video game Ms. Pac-Man, using a divide-and-conquer method that could have broad implications for teaching AI agents to do complex tasks that augment human capabilities.
These AIs are relatively simple and single-purpose now, but just remember what computers looked like only a few decades ago.
I wish I could have used something like this in Game of War: Fire Age!
Ah the memories…
Okay, 3 decades ago computers were not simple and not single-purpose. They have greatly increased in speed, formfactor, portability, but fundamentally nothing changed. We are still using the same ideas about motherboards, CPU’s, cache, memory, harddisk.
What did change a lot was software. Multi-tasking became completely normal and we first developed lots of techniques (COM/OLE) to let programs integrate with each other, while now we are isolating/sandboxing them again.
Almost all AI software is now written as a closed block, similar to old standalone exe’s. There are a few shared libraries/frameworks, mostly in the shape of online webservices (Azure Cognitive Services), but even Googles Tensorflow is mostly used as a local shared package.
Multi-purpose, self-writing/adapting software has been an industry “promise” for a long time, but hasn’t really materialized for general purposes