Insane Algorithms

I don’t think we realise just how often things that happen to us are as a result of decisions taken by artificial ‘intelligence’ algorithms running on someone else’s computer. I had a classic example the other week when a grocery and household goods delivery, ordered online from our local supermarket, arrived. Three of the items we ordered were out of stock, so the store’s computer made substitutions. The first substitution two tiny pots of chicken gravy substituted for one large packet of fresh chicken stock. Well... You could argue that they were similar and that stock is used to make gravy. The second substitution was of a similar ilk. We asked for a packet of resealable plastic freezer bags. What we got was very large freezer bags that weren’t resealable. Again just on the edge of what a not very bright human might have chosen.

The classic case, though was the third item. My wife wanted one of those plastic claw hair clips to hold her hair back. It turned out that they’d run out of plastic claw hair clips, so what did they offer in stead? A packet of EMERY BOARDS! For those of you not familiar with emery boards, think in terms of disposable nail files. I have no idea how it got to that substitution, but it’s not the first time it’s come up with such bizarre substitutions.

Actually, I’m less worried about those sort of things than I am about more subtle errors that are not immediately obvious. I don’t think people realize just how much decision making is under the control of computer algorithms. To give just one example, 85 percent of all trading in the foreign exchange markets is conducted solely by algorithms.

And it’s not just single algorithms, it’s also the results of multiple algorithms feeding off, and competing with, one another – resulting in problems like books being priced at millions of dollars. The problem is that these are the things that are obvious, and we can see them, correct them, have a laugh and feel superior to a ‘mere’ computer. We had a good laugh with the delivery driver over the emery boards. He reckoned that he’s seem even more bizarre substitutions in the past.

And it’s not just the algorithms – part of the problem often lies in the data sets they are trained on, which contain unrecognised biases. If you train an algorithm with a biased dataset, then you will get an algorithm that gives biased decisions or biased recommendations. And how do you figure out whether a data is biased (always assuming you know what data set was used, of course)? Please don’t suggest using an algorithm to check it!

Still, all is not doom and gloom, I can offer one bit of consolation. It seems that a recent survey of 2,830 AI start up companies revealed that 40% of them have no AI. That’s right, two in five had no evidence of using or developing machine learning tech. It’s all bluff to get their hands on venture capitalist funding.
https://www.sciencealert.com/algorithms-are-already-making-decisions-for-us-and-some-are-utter-madness
https://www.theregister.co.uk/2019/03/05/eu_startups_no_ai/

Alan Lenton
17 Marcy 2019