IT HAS only taken 35 years and a revolution in technology to do it but someone has completed Ms Pac-Man with a perfect score.
The iconic 1982 game was conquered by a Microsoft-owned AI from Maluuba. The task was taken on by 150 neural networks working as a hive mind to take on the monumental challenge of all 256 levels with a score of 999,990.
It's worth noting, before the pedant police get to us, that the game was completed on an Atari 2600 console, as the arcade version, like its predecessor, had a bug that stopped Level 256 being completed anyway.
We've talked before about how AI is often taught using a risk-reward strategy, and it was this that was employed in allowing the networks to teach themselves how to overcome Inky, Pinky, Binky and Clyde with absolute perfection.
A similar technique was used by Google's DeepMind to defeat human Go champions.
It's actually a very good example of how good AI is, but how far we've got to go. A single network couldn't manage all the intricacies of gameplay, so instead, 150 different AIs were taught different aspects of "grab the power pill", "avoid Clyde", "slide out the side door" (yeah, OK, we were fans), and then a "master AI" was used to coordinate the different techniques.
It's worth noting, for anyone who doesn't know the game as intimately as our arthritic thumbs, that the ghosts all have their own personalities and signature moves - very ahead of its time for 1982 - meaning that simply "avoid the ghost" isn't going to cut it, plus the difficulty of each level means that you need to change tactic as you progress. All of which the different AI agents would need to learn.
The result is that the master controller will defer to the survival imperative over the need to get a pellet, a system the team refers to as "Hybrid Reward Architecture".
Microsoft suggests this could be used by a sales team to prioritise customers according to need, because, you know, let's sap all the fun out of it straight away, yeah?
Microsoft has been buying up AI companies recently, for everything from improving sales data to bolstering anti-virus.
Next up, we hope, how to get past Maria and go to bed in Jet Set Willy. µ
Justine Greening and Greg Clark among those affected
A whole new take on 'cord-cutters'
Surely everyone can get a long?
Report also points to an ARM coprocessor for Touch ID