Machine-to-Human Communication: Nobody Cares

Ticketless illegals trapped inside tram

I continue to have bad experiences with the machines that purport to talk to me in everyday life. Recently I took one of the new trams to the Edinburgh airport. The computer-controlled doors closed and the tram moved off. As it glided away, a smooth prerecorded voice told us: “Please note that tickets must be purchased, or cards validated, before boarding the tram.” A bit late for that! Couldn’t the system have been programmed to supply that crucial information while the doors were open, before the hapless and illegal non-ticket-holders were trapped inside?

And talking of automatically closing doors, in the elevator in the building where I work, when the doors are about to close a voice announcement says “Doors closing.” If an extra person slips in before the doors move, the door-blockage sensor spots them, and the voice says “Doors opening.” (They’re not opening, because they’re already open, so the voice is completely wrong.) And then the doors close again, but this time without a “doors closing” warning!

What’s the point of an audible warning to the blind if it is sometimes not given, leaving them oblivious to whatever is supposed to be the danger of gently closing doors? (The doors stop closing instantly anyway if they detect anything between them, even a piece of paper. There is actually no earthly reason to warn anybody, let alone blind people who are well able to detect the sound of an electrically operated sliding elevator door and step out of its path.)

It’s a bug in the program: The voice is supposed to issue a spoken warning whenever the doors are closing, but in the last-minute-interruption condition the operation fails. And nobody at the manufacturer’s noticed, or nobody cared. (They also didn’t care that at the third floor, the synthesized voice says “Level droon!” — the speech chip only needs seven integer names, but there’s a bug in one of them.)

Minor glitches, you might say. Even trivial. But they are indicative. They involve ineptitude in attempts to get machines to communicate with humans in spoken English. And the one place where you really need machines to show a little bit of intelligence is in communications with human users.

It’s all very well to say that algorithms and hardware for searching and sorting and data analysis are wonderful. They are. But those tasks do not call for intelligent interaction with us. They involve machines doing work of exactly the sort that should be delegated to machines because they are good at it. What I’m talking about is saying the right thing at the right time in simple “Hello — What-do-you-want — Here-it-is — Goodbye” interactions.

I have mentioned this before, as regular readers will recall. The commenters weren’t all sympathetic. Some stubbornly defended the machines (they did their job well enough) or the programmers (some bosses actually tell you not to bother ensuring that the device says “You have 2 new messages” but not “You have 1 new messages”).

Some people utterly missed the point I was making about logical communication and told me piously that I was ignoring the needs of blind people (“I find this article and its premise rather condescending and ignorant of other people’s needs” said one person coldly).

It must be my fault: I’m not explaining clearly enough what my worry is.

Basically, I’m worried that machine-to-human interface programming is being done so superficially, sloppily, and incompetently that it looks as if the industry has decided this is good enough: that nobody needs anything better.

It must get better, surely. One day these machines without which we cannot live have to be upgraded to a level where they can understand what we say in our native languages, and tell us things sensibly in those languages. That is going to demand attention to the pragmatics of conversation (no more telling us things that in the present context we couldn’t possibly need to know, or giving us information that we couldn’t possibly use), and some elementary grammar.

Some in-car GPS systems in Germany make voice announcements mentioning streets and roads without even getting the bare essentials of the gender classification system right, e.g. distinguishing nouns like Weg (“way”, masculine) from nouns like Strasse (“street”, feminine), and so on. Nobody bothered, probably because nobody’s boss thought they should.

All of the examples in my earlier post and this one are meant to illustrate the same thing: my worry that machines do a miserable job of talking to us because of bad and sloppy programming, and the programs are that way because nobody in charge really cares.