Aging has few qualities to recommend it. There are two things that concern me most: the first is working with people who are younger than your oldest pair of trousers; the second is having to stop using Monty Python gags because literally no one under the age of 30 has the faintest idea what you’re talking about.
So it was with not one ounce of regret that I used a line from the famous (no it really was, long ago…) Argument Clinic sketch for the title of this set of ramblings. You see arguing is fantastically complicated. It is. No it isn’t. No really it is. Very.
Last week, it all came flooding back – we’ve been collecting video interviews with the great and the good in the world of AI and we had the privilege to interview a guy called Chris Reed who is a Professor of Computer Science and Philosophy. Chris does Argument Mining… but not like this:
Man: I came here for a good argument!
Other man: No you didn't, you came here for an argument!
Man: An argument isn't just contradiction.
Other man: It can be.
Man: No it can't! An argument is a connected series of statements intended to establish a proposition.
Other man: No it isn't!
Man: Yes it is.
Etc., etc.
Fortunately, my colleague Laura van Beers was the interviewer, or we might not have progressed much further. What was interesting about Chris and his work – apart from the fact that the man is wonderfully articulate, an expert in his field and a thoroughly decent chap – is that it really is the new frontier in the world of Natural Language Processing and AI.
But, and here’s where the subject gets really interesting to me, it is also desperately interdisciplinary. Combine computing power, linguistics, psychology, environment, facial expressions, irony, sarcasm and just about everything else we have evolved to understand, and you start to see just what a massive challenge arguing presents to computers.
Let’s start at the beginning: want to stop the Terminator in his tracks? Use a pronoun. At the risk of being patronising, a pronoun is a word that refers either to the participants in the sentence (e.g. I, you) or to someone or something mentioned elsewhere in the sentence (e.g. she, it, this). The example Chris gives in this excellent article on the BBC is:
"I like Amy Winehouse. Play something by her," the software/terminator would be unable to work out that by "her" you mean "Amy Winehouse".
As Chris puts it, this is “hardly the stuff of robot-apocalypse nightmares”, but arguing is what people do best – we do it all the time. I sometimes do it with myself. (I rarely win.)
Alexa, Siri, and all the little voice bots out there are only making baby steps. They are good at understanding simple intents like “Alexa, play Back to Black please”, but what of the discourse used in court, government or the often diametrically opposed but well-argued articles in the Guardian vs The Daily Mail or Fox News vs CNN?
‘Fake news, fake news!’ I hear you cry. Well, yes, lots of arguments made by clever people can sound very plausible and yet be mistaken or, worse, deliberately misleading.
The most upsetting thing about Brexit was the arguments from the likes of smarty pants Michael Gove (minor public school, 2:1 English, Oxford) and even cleverer clown-in-chief, Boris Johnson (Eton, 2:1 Classics, Oxford). Their claims about money for the NHS and ludicrous comments about fishing quotas and ‘ruddy Brussels’ flew in the face of any kind of deeper analysis. But both learnt a thing or two about argumentation at their respective public schools, Oxford Union positions and their roles as journalists and (to their own surprise and, I hope, shame) they won the day.
The work that Prof Chris is doing at Dundee is a brave new frontier. Taking a look at millennia’s-worth of argumentation techniques from Plato to Wittgenstein gives us the foundation to unpick complex arguments and see if we can train computers to understand them. If we can, that might help us stop fake news, arrive at better conflict resolutions and maybe make the world a more balanced and thoughtful place.
The technical challenges are enormous. In the interview, Chris described an explosion of research. Ten years ago, he reckons, there were fewer than five published papers in this field. Over the last five years, literally hundreds of researchers have begun trying to work out how to train a computer to have a decent argument – or at least explain the truth behind the claims, (and no, Michael Gove, the fishing industry collapsed, not because of Brussels red tape, but because we ate all the ruddy fish. Ask your dad – he called it right).
The interview will be published shortly and take us deeper into this fascinating field so all I need to do is close off with another obscure quote:
Man: I'd like to have an argument, please.
Receptionist: Certainly sir. Have you been here before?
Man: No, this is my first time.
Receptionist: I see. Well, do you want to have the full argument, or were you thinking of taking a course?
I think most of us need the full course.
Ref: For those with young trousers, watch this and reflect on how things were funnier when I was young: https://youtu.be/wdoGVgj1MtY
