If you have a smart phone or even a not-very-smart-phone, then you will probably be familiar with the predictive text function that can be activated for text messaging and the like. Often as not, it is the default setting and you are obliged to wade your way through menus and sub-menus to turn it off. But who wouldn't want an AI program that helps when you misspell or mistype words, right? Everybody loves a helper.

Except when it not a helper. And that seems to be most of the time with my predictive text function, which offers up absurd or plain incorrect alternatives and is impervious to learning from my most frequent mistakes. Because I am two finger typist and a one-finger messager, I often get simply words around the wrong way, if yuo get ym dritf. But the program never learns from these simple errors.

Worse, it often inserts preposterous guesses, such as 'lesbian' for 'I've been' and is very poor on verb agreement. How many times have I had 'has' rendered as 'had' and much worse besides.

Much is made of an AI revolution and the threat that might bring to organic life, should the former develop some level of 'digital sentience.' But I think the problems that are inherent in predictive text programs augurs well for us. Just imagine a text from a central AI unit to commandeer all smart phones going horribly wrong (for them). The self-destruct button will always be a possibility if instructions turn to gibberish.