An algorithm is a process or set of rules that is followed to make calculations or solve problems, especially by a computer, says Wyn Drabble Photo / 123rf
An algorithm is a process or set of rules that is followed to make calculations or solve problems, especially by a computer, says Wyn Drabble Photo / 123rf
OPINION
We all know that Big Brother and others are watching us, but some watchers appear to be getting a few things wrong. Perhaps their algorithms need servicing (or at least an oil change).
An algorithm is a process or set of rules that is followed to make calculations orsolve problems, especially by a computer. After watching (spying on) our online purchases and the like, algorithms work out our likes and dislikes so they can send us unsolicited advertising for things we might buy. It is personalised and targeted and constant.
The word “algorithm” comes through French, Old French, Medieval Latin and Arabic, and initially meant “word that is a bit funny to spell”.
In my case, Spotify’s algorithms seem to be working effectively so they’ve clearly been thoroughly serviced. I’m constantly surprised by how their suggestions match my tastes. Even when they recommend singers or bands I’ve never heard of, I almost always like them — despite the spooky background feeling that I’m constantly being watched.
Other operations are not so successful. A New Zealand auction site I use reasonably regularly — let’s call it Trade Me because that is its name — either has faulty algorithms or their algorithms are desperately in need of a service.
For argument’s sake, I’m listing an imaginary object (not its real name) that I once purchased for $5000. Because it is still in brand-new condition, I decide to list it for $4000. The helpful advice pops up as I am typing the details that “items like yours have sold on Trade Me for $295 to $465″. Pardon me?
Sometimes Trade Me sends me auctions for items I might be interested in, that they have based on “my previous purchases”. The suggested items are invariably things I would never even consider parting with money for, or items in which I am not even remotely interested.
Trade Me: Based on your past purchases, we thought you might be interested in this auction: Macrame wall hanging. Hand-made. Some beads missing. $5. Shipping extra.
Wyn Drabble. Photo / NZME
I’m afraid their algorithms have missed the mark by a long way. I definitely would not buy a macrame wall hanging with beads missing. I would definitely not buy a macrame wall hanging in any state of repair.
Another online seller has missed the mark with Mrs D as well. They’ve sent her recommendations for reading that they claim are based on her purchases/reading habits so far, but the algorithms are very often way off the mark. “I’ve never read a (insert genre here) book in my life!” she might assert.
In the general world of business, predictive algorithms can make serious mistakes. They could, for example, find out the characteristics that make real estate desirable in Queenstown and make a jolly good job of it. But apply the same findings to the Auckland real estate market and they might do a very bad job.
So how can Spotify and Trade Me have such different findings? One possibility is that Spotify is entering far more exact data and crunching the numbers more carefully; the results certainly suggest that.
Trade Me, on the other hand, needs to up its data game or even revert to the old-fashioned method of allowing human judgment to do the work. Human decisions can be very flawed too, but on current evidence they might be better than their algorithms. In a scientific digital world it should, of course, be the other way round.
Trade Me (interrupting): Based on your comments in this newspaper column, our algorithms have determined that you might be interested in the following auction: Spotify subscription, mint condition, no algorithm issues. $1 reserve. Buyer must pick up.
