NA: I want to move on later to talk about AI as a creative tool, but first we should talk about chatbots. I listened to a podcast recently about Howard Schneider, the inventor of self-checkout machines. He was an emergency psychiatric doctor and he became obsessed with how the brain works and what happens when it stops working. He was also a tinkerer and programmer, so the two interests combined. This was in the late 1980s when ATMs had taken off and he saw the check-out as the next frontier. Some of it was relatively simple – scanning barcodes and installing weighing scales. But the tricky question was how to deal with unpredictable human behaviour. The podcast talks about ‘unexpected item in the bagging area’ and how they spent ages working on a phrase that wouldn’t sound like an accusation.
Without getting into amateur psychiatry, it’s hard not to think of autism when you read about Schneider – he has this obsession with reinventing human interactions, but removing the human. And there’s an obvious parallel with chatbots. Right at the point where a customer wants to interact with them directly, brands are desperate to insert machines instead.
It sounds strange and dysfunctional. But I also like the idea of robots representing brands, because they might be a better fit than humans, who are often terrible at representing brands – like when that British Gas tweeter recently tried to pay respect to David Bowie and messed it up. A bot wouldn’t do that. What do you make of chatbots and conversational UI?
RD: Well, I’d like to start with another story. We went to Disneyland in California a few years ago and my son was at an age where he liked doing things over and over again. So we went on the Jungle Cruise several times. It’s a trip on a boat and some animatronic lions and hippos pop up around you while the ‘captain’ does some comedy business. I wasn’t looking forward to the second trip until the captain did a completely different set of gags – and they were different each time we went on it.
It made me realise that Disney has very easy access to quite a rare natural resource: lots and lots of high-quality out-of-work writers. Disney would never allow a machine to say ‘unexpected item in the bagging area’. They’d throw a decent writer at the problem. Companies that ‘automate processes’ typically have no idea how to do that.
Chatbots are machines for distributing, across time and space, the decisions a writer would make
And that’s why so many conversational interfaces suck – they’re not written by writers. Chatbots are machines for distributing, across time and space, the decisions a writer would make. Bad writers make them awkward and creepy; good writers make them useful and fun. And I suspect brand and advertising writers will be good at this because they’re used to evoking the exaggerated, bulbous, cartoon-like personalities of brands. Humans are inclined to see everything that has two circles and a line on the front as some sort of little creature. Similarly, we treat everything that behaves in a vaguely human way as a person. Chatbots can tap into that and good writers can use it to create usefully differentiated personalities.
Plus, it seems like it’d be enormous fun. Wouldn’t you like to pour your skills into a piece of software that does the writing for you, rather than have to crank it all out by hand?
NA: Yes, I would. It feels like someone will do something great with chatbots. I thought Siri had lots of nice writing touches, but things haven’t moved on much since then. There are so many potential personality types – glum, sarcastic, excited. But a single personality type seems to be developing for all personal assistants – Alexa, Siri, Cortana – and they’re all female. It’s interesting how there are so many fictional forerunners to chatbots. Holly in Red Dwarf; KITT in Knight Rider; HAL 9000; the ship’s computer on Star Trek. It’s like everyone assumes the future is spoken UI, but I wonder how many of us really want that – even Star Trek has Scotty getting wound up by chatbots.
I also wanted to talk about AI as part of the creative process. A year ago, you created the Twitter account @taglin3r and it’s been tweeting corporate taglines on the hour ever since. I’ve done a similar thing with a physical tool – the Slogan Cube – which is good for three-word lines. As an aside, I realised recently that ‘tagliner’ is an anagram of ‘triangle’. Given the trend for three-word lines, and the numeral 3 in @taglin3r, I felt like I’d stumbled on a tagline leyline.
You’ve written before about how @taglin3r could be converted into a creative tool to help in brainstorms – feed in key words and generate lots of options. And it’s not just taglines – we’ve seen bot-generated poems from Google, Stanford University and the University of Massachusetts; screenplays, such as the sci-fi short, Sunspring; songs and just about everything else. Is it all just a novelty – and not even a very new novelty – or is it something to get excited about?
RD: Well, it’s exciting me because I think we’re about to get some brilliant new tools to play with. @taglin3r was supposed to demonstrate the way that quite a lot of apparent ‘creativity’ is more mechanical than we suppose – it’s actually pretty constrained by industry convention and corporate vocabularies.
But it inadvertently illustrates two other things. Firstly, that an algorithm will often make more startling and original choices than a human (because it doesn’t know lots of the unstated rules that humans don’t realise they’re following); and secondly, that a lot of the real value in creativity is in judgement not generation. It may simply be more efficient and fun to get a machine to generate a thousand choices and then get a human to pick out the nuggets. I’m obviously not alone in spotting this and I suspect people are working on better tools than @taglin3r. Designers aren’t far off having software plug-ins that will use AI to make their artwork match the style of any piece of stimulus you feed in.
I’m hoping that’s around the corner for writers, too. It’ll make everyday hack work a lot easier but what’s genuinely exciting is what’ll happen when we take those tools, turn them up to eleven and start generating all sorts of feedback and distortion. Then we’ll see something properly new.
Below: A bot generated song
NA: We started off talking about spoken UI, but the area that interests me most is still text. Maybe it’s my bias as a writer, but text is an amazingly flexible technology for sharing and mining information – more searchable than voice, less intrusive as a UI. Even now, many of us choose to have conversations in text messages that would logically be quicker over the phone, but we just find it more comfortable and less effort that way. And text also absorbs new technologies easily – we’re already used to autocomplete, spellcheck and basic creative tools like domain name generators. Now there are lots of growing areas like voice-to-text, live writing and tone analysers that use deep learning to filter out hate speech.
Maybe speech and text will merge more closely in future years, but I imagine text will still be the main place where information lives. Star Trek has everyone walking round talking to computers, but it could be more a case of people sitting at home texting each other and steering the Enterprise remotely, while reading a killer experimental immersive novel.
Anyway, I won’t ask you to do futurology, but to wrap things up it would be good to hear what areas interest you most right now and if you have any ideas for further reading?
RD: Well, personally, I’m not ready to give up on speech yet, but I think speech radio might be a more fruitful model to pursue than chat. Speech is brilliant as an ambient, broadcast channel, gently whispering stuff into the edge of your consciousness, rather than demanding constant interaction and 100% attention like a speech UI. Or like text.
I can imagine something AI-generated that drones away in the background like cricket commentary, occasionally snagging your attention when it mentions your name or the fact that your train is delayed. I’m looking for that. For links, there’s always new and crazy stuff at CreativeAI. On Medium, Ross Goodwin has a great piece on how to actually do computational creativity; Paul Ford has written about how bots are as old as the hills; and Jonathan Albright’s article on robotically-generated fake YouTube news should scare the bejeesus out of you. And if I tell you about my slightly stunted experiment at funnyAI.science then I’ll feel obliged to try and do more with it.
The post Nick Asbury and Russell Davies discuss the impact of automation on writing appeared first on Creative Review.