Back in 2014 it was news that a chatbot had passed the Turing test. Way back in the 50’s Alan Turing (“father of modern computing” and he of ‘Enigma’ etc) proposed a test which translates today as two people chatting via text or a messaging window. If one were actually a computer (ie a chatbot) and it wasn’t possible to tell who was the real person, the computer (manifesting as chatbot) could be considered as having passed. Of course this doesn’t mean that it’s suddenly going to become our Artificially Intelligent overlord (although apparently chatbots on Twitter gain followers more quickly than humans); it just means that someone has devised a good set of rules that seem natural. In fact the case I am talking about was an entity called Eugene, modelled on a 13-year-old Russian schoolboy with English as a second language. So any observers would probably be more forgiving of little idiosyncracies. Most such chatbots are easy to get into loops; e.g.”Whats your name?” Eugene, ”How do you spell that?” How do I spell what?“Your name?” Eugene…
So what has this got to do with support desks? I’d like to propose the opposite test – when does a support “chat” fail the Turing test. We have all had dealings with Asian help desks and their varying effectiveness. On the plus side they allow support to be provided 24/7. I recently changed web hosts for my 20 some websites and email. It was, to my surprise, fairly straightforward. Despite the fact that they offered to do the job for me at $75 per web site, the actual transfer took a matter of minutes once I had jumped the hurdles of actually transferring the 6gb of data and handling the interruptions that halted the automated process (fwiw I ended up zipping up locally and using Ftp). The problem came when they used different terms to those my prior host had. For example, to use an independent domain (such as mydomain.com rather than a sub domain such as mydomain.patbell.co.uk) I needed to park a domain. Their help desk directed me to their online manual for cpanel (the open-source interface for hosts). However the icons I was directed weren’t there since the open source but means that the program can be changed but the manual had not. Anyway after they offered to do it for me and had done it backwards so none of my sites worked at all I managed to resolve the problem without their help at all. The point is that when giving me “help” at no point did they acknowledge that any previous instructions were wrong. I learnt that “We are sorry for the inconvenience” message actually meant “I have no idea, but wait a few minutes and I’ll serve you up some more crap” The first line support failed the Turing test and clearly followed a flow chart that needed drastic updating. The second line support made so many cock-ups they had to be human!
Just recently Microsoft created a chatbot called (see www.Tay.ai) aimed and teenagers and young adults. Its “persona” was a somewhat naïve teenage girl who started out with “gee its really stoking to meet with all you humans” then “wouldn’t it be great if everyday were puppy day” but within an hour the mischievous imps of social media had turned the adaptive “AI” chatbot into a right-wing racist bigot who was saying things along the lines of “Hitler would have done a better job of 9/11” and “Donald Trump for president not this monkey” Its gets worse but MS took it down within an hour, phew!
Except of course that Trump did become president, go figure!