Turing Tests and Train Trackers
In his classic paper “Computing Machinery and Intelligence” Alan Turing side-steps the question of whether or not computers can think by describing a simple experiment. He asks us to consider a discussion with a human and a computer pretending to be a human conducted via a text interface. If we can’t tell which is which, the computer passes the test.
Today I had a chat with a computer using an audio interface when I phoned up national rail enquiries for timetable information. Actually I spoke to two computers: a female on the main switchboard who, having categorised my enquiry, passed me on to one of her colleagues at train tracker. This colleague never pretended to be a human, but he did have a lively voice and he dealt with me politely:
“Thanks … I’m sorry, I missed that … Was that seven in the morning or seven in the evening? Say ‘morning’ or ‘evening’ … Right … OK then …”
Actually, I’ve spoken to him before, and he seemed to remember this. Maybe I should make a note of his direct number.
Nice but Dim
My only complaint is that, I’m sorry to have to say this, but he was a bit thick. Didn’t catch my drift, if you know what I’m saying. Couldn’t stop apologising. Kept repeating what I’d just told him.
Most of the time I access train information far more easily using a real computer anyway.
Just plain stupid
Network rail have been using robots to announce train information in their stations for a long while now. They have a female and a male announcer. (I wonder what their names are?) Most of the time these unfortunate robots are left to do the dirty work of apologising just how late services are running.
If a train is running a couple of minutes late:
“I’m sorry to announce that the 07:20 to London Paddington is running 2 minutes late.”
Most passengers would be delighted to get on a train which was only a couple of minutes late, but we soon learn the robot was actually preparing us for worse news:
“I’m very sorry to announce … 10 minutes late.”
which soon becomes:
“I’m extremely sorry to announce … 30 minutes late.”
When things go really wrong at a busy station like Bristol Temple Meads the whole station rings with a cacophony of hollow apologies1.
Thank goodness for the actual staff on the actual trains when they actually arrive who do a decent job of explaining what the real problem was.
An alternative approach
In the middle of the last century Turing predicted computers would perform well in his test in fifty years’ time (a few years before now, that is). Whether or not they could actually think he believed a meaningless question, but he did say:
“Nevertheless I believe that at the end of the century the use of words and general educated opinion will have altered so much that one will be able to speak of machines thinking without expecting to be contradicted.”
Thinking? It’s certainly common to speak of computers as being stupid or annoying. National rail’s robots aren’t trying to fool us into thinking they’re real people, but they are trying to present a more human interface and for me it’s the repeated apologies that grate. Is it, perhaps, that we’re afraid of computers taking over, so we make them behave like obsequious slaves? Or is it just poor interface design2.
Here’s a better idea. The computer I use can speak in a number of voices: male, female and novelty. My personal favourite is the “hysterical” voice, which mixes speech with uncontrolled giggling. Here’s how a hysterical delayed train announcement would sound.
An improvement, I think.
1 There are parallels with my day job here. When developing software it’s all too easy to concentrate on smooth operation; on what’s meant to happen when the user does what they’re supposed to, when their inputs are valid and the system has plenty of memory, just as the train announcement system probably works just fine when trains run on time. We may well attract a few users with such software, but if, when things go wrong, those same users feel out of control, or lose their data, or simply get irritated, then it’s going to be hard to regain their goodwill.
2 Yes, it is poor interface design. So often when dealing with a computer we have to repeat ourselves or spell things out: yes, I really do want to delete that file (but provide me an undo button when things go wrong); and yes, if I can’t edit the file just open me a read-only version for goodness sake (a quick note in the status bar will let me know what’s going on). Computers should understand that sometimes it’s easier to ask forgiveness than permission.
Feedback
-
Turing probably assumed that the differentiator between human and computer would be stupidity. If you're implying that the differentiator is actually potliteness, then I'd agree with you
I book train tickets by phone quite a lot, and I find the robots to (pretend to) be polite, interested and cheerful. The humans make it clear that they couldn't care less. Recognition accuracy is no higher with humans, especially now that they are rarely native English speakers.
Of course I don't really care whether the agent cares about my plight, but I do want a reasonably pleasant interaction. I think voice user interfaces have higher acceptance in the UK than either the US or Europe, and I shouldn't be surprised if this is the reason why.
[Disclaimer: I am a speech tech developer - although more R than D]
-
Turing probably assumed that the differentiator between human and computer would be stupidity. If you're implying that the differentiator is actually potliteness, then I'd agree with you
I don't think the train tracker robot is actually trying to pass the Turing Test, so the whole Turing thing is a bit of a red herring. That said, to pass the Turing test, a computer would have to pretend to be "stupid" -- to have trouble with chess problems and maths, for example.
I book train tickets by phone quite a lot, and I find the robots to (pretend to) be polite, interested and cheerful. The humans make it clear that they couldn't care less. Recognition accuracy is no higher with humans, especially now that they are rarely native English speakers.
I've not used the phone to book tickets in a while and didn't know robots were manning the line now. I rarely found it especially pleasant dealing with the human ticket sales agents in the past.
I certainly found the train tracker robot to be polite, interested and cheerful. He also has the opportunity to recognise me and adapt himself to my voice etc. I also discovered I could rudely interrupt him when I knew the answer he needed before he'd completed his question.
-
We call that barge-in.