Wednesday, June 28, 2017

Unexplained Phobia

Scale and complexity apart this article trying to whip up some AI phobia, recalls issues that have always been the around with poorly written code. Sure, the guy who comes up with the algorithm for the machine to teach and train itself may no longer be able to understand the results being produced by his code. The way the program works and the results its produces are clearly very hard to debug. This is no surprise.

Without naming names here, enterprise software products that have been around for decades have no lack of unsolvable bugs with more being added every release. The problem of untested or un-testable code (driven by bad design) is hardly new to AI but the consequences have grown considerably. But for those of us who understand the basics of programming it is hard to understand why the AI driven car run amok is any different from the various calamities we have faced in our professional lives when random shit happens with code with serious consequences for business and customers they serve.

There was the time when the e-commerce website was pricing stuff at $0 at midnight and some folks wised up to it and started to place large orders. I was grateful not be the on that fire-drill and only support those who were. Now imagine there was an AI programmed to be rewarded for largest number of orders being placed, it may have learned from the transaction pattern here to set all prices to $0 so the orders would flow in. The losses could have easily ballooned when bad code gets such a shot in the arm. This is like teaching your precocious four year old to read Green Eggs And Ham and then making them in charge of the MFA program in university.