- Peter: I am opening for the PS Waverley, which is passing downstream.
- Keith: Learn to forgive yourself
- Peter: I am closing after the PS Waverley has passed downstream.
- Keith: Everything happens so much
- Peter: BONG BONG BONG BONG BONG BONG BONG BONG BONG BONG
- Keith: Your perfectly manicured hands are folded serenely in your lap. You re totally prepared. . .or are you?
We like robots, and we inhabit a world in which robots are real, and we communicate with them freely and on a daily basis. This morning we are going to talk about how some of these robots entertain us; some help us; and how we have accidentally employed a team of robots at Safari Books Online.
To be clear, when we talk about “robots” we don’t mean the fictional robots of cinema and literature like HAL or Marvin (much as we love them both).
Instead we have given up on the things that vacuum your house, and mean robots made out of code, almost always without a physical manifestation, who just have a simple job to do day in, day out, communicating in a human world. Communicating with these robots is like “overhearing machines talking about what they’re doing” – their interior monologues, status reports, and chats with other robots.
Twitter, being all about status updates, has always been a great breeding ground for robots.
In 2008 Tom Armitage built a robot to update the world on the status of Tower Bridge in London and more recently Tom Coates did something similar for his house, which he has wired with sensors connected to the internet.
Both of these robots personify data as jovial, polite robots. But robots don’t always have to be jovial or polite.
In fact, one of my favourite robots is Stealth Mountain, whose job it is to be heroically, epically pedantic.
All it does is sit on top of Twitter and wait for people to mis-spell “peak” when they say “sneak peek”. It does this hundreds of times a day, to the endless amusement of thousands.
When early scifi authors talked about robots in the future – they wren’t really thinking about the world’s most massively scalable, real-time pedant. But we find this amusing.
So – we like robots with human characteristics; but sometimes robots confuse us. Like a lot of people we were taken with the idea of @horse_ebooks, a robot who appears to tweet incongruous, decontextualised passages from ebooks, with no apparent motive or sense.
This gets a bit meta, but when @horse_ebooks was exposed as a human masquerading as a robot, did we feel cheated? Or had we had a sneak peek into the future? People were happier with a poorly implemented spambot than they were with a performance art piece. Which is weird.
As well as some robots, Safari Books Online also employs 150 humans. We have five offices across the US and many other people spread across the country and the world working from home. Unlike some companies, we have adapted to this remote culture and find it delightful. That said, we have created coping strategies for communicating with humans for maximum output. We videoconference a lot. Humour (and trolling) are features.
We have many many chatrooms; each has developed their own mood, idiom, and communication style, as have the humans who inhabit them. Some humans communicate best with words; others in the third person; while some find YouTube videos and animated GIFs more expressive.
But all the rooms are digital proxies for physical and human interaction. In among all of this technology, we also employ robots to make some of the drier, more transactional conditions more humane.
Some of these robots exist to amuse us…. for example we built our own version of horse_ebooks, @horse_safari, which takes actual phrases from our corpus of books – the more incongrous the better – and posts them at moments of awkward silence in a chatroom.
Others exist to help us….This chatroom is Heron, basically the “war room” for Safari Flow, our new consumer product.
The Heron room is inhabited by many humans, the majority of whom do not speak fluent code. Our customer service, QA, design, project management and product engineering teams all use this room a lot – principally human communication, but also with the hum of machine work around them. As a result, and quite naturally, Heron is also inhabited by many robots providing status updates – about code deploys, test methods, GitHub, and other things.
The humans and the robots co-exist peacefully, although when the robots alert them to a problem, the humans take the input very seriously. No-one thinks it is weird to take instruction from a robot.
So we have complicated relationships with robots; and we are not alone in this. It turns out there has been a lot of space between how robots were first thought about, and now.
70 years ago, Isaac Asimov wrote the original three rules for robotics; we don’t necessarily believe in them any more. We enjoy the boundaries provided by these rules, although perhaps breaking those boundaries is more enjoyable? Should a robot still not harm a human? Depending on how you define harm; a lazy or snarky robot is kind of amusing, but mildly harmful.
Asimov said a robot had to obey orders and do useful things – but we are more flexible about occasionally insubordinate robots who don’t do what they are told.
Additionally, that a robot must protect its own existence is kind of an extreme idea, but everyone loves fighting robots, right?
When we think about our ideal robots then, they display human characteristics, but are also snarky, insubordinate, and a little belligerent.
As well as these funny robots, we have also created robots as toys that have become surprisingly useful.
Jirabot is a robot that lives in our chatrooms, and exists to explain a ticket in our tracking system whenever it is mentioned. The humans are lazy and use the shortcodes, which provides no context for (say) HERO-4396. So Jirabot does the legwork for us, providing the description, status, assignee and URL; expanding the shortcode.
Jirabot is therefore helpful, and as such our interactions with it are tinged with gratitude, which means that we communicate in an increasingly human manner, even though we know Jirabot is non-sentient.
Like any relationship this starts off respectful, polite, but then quickly decays into abuse, swearing and sarcasm. It was interesting watching this anthropomorphism happen.
Chirpy is our first robot to have a physical presence, in our SF office.
It was built by our colleague Bill as a toy to troll Peter, every time a customer buys a subscription to Safari Flow – for which our codename was Heron.
Made out of an old box, a Raspberry Pi, Lego, a chopped up Amazon crate and some glue, Bill made chirpy sing the mating song of a heron and flap its wings and blink its lights every time we got paid.
So we have a world of robots, and in January of this year we began thinking about a new product, unaware that robots would become a natural part of this.
Our goal was to build a new, approachable product that would help people gradually become more awesome at their jobs through incremental learning. It would use the existing resources we had at Safari, and we wanted to make it really helpful and easy to turn to. We wanted to make something they might turn to in the same way they turn to Twitter or Facebook, for short bursts of distraction.
Like those products, it should be highly personal, data-driven, and point at interesting things for that user.
Thinking about it in those ways, we realised we could use years of data to point people not at whole books or courses,but the parts of them that were most likely to be interesting (and short) – chapters and clips.
We could say with some certainty that if you’re into “Agile” then you should start with this chapter, then move to this clip, then go on to this chapter, and perhaps try something out in “Teams” after that.
We were focused on how to delight people, make them more awesome at their jobs, but also mindful of the huge apathy that exists in life. Which is why Twitter gives us that dopamine hit. So our goal was just to do anything we could to get people into the chapters or video clips – because then they are happy.
We want to tell them that in seven minutes they could read this piece about big data and come out the other end just a little bit smarter. Our whole goal was to drive them into something useful that they would find surprisingly helpful at making them more awesome at their jobs.
As we thought about it, our empathy to robots coloured how we responded to this problem. We had the books and the data, and it never occurred to us to do anything other than build a robot on top of a recommendation engine, and to believe that a robot would be good enough to interact with the people and tell them what to do. So we did that for six months.
It turns out that now, at the end of building and launching a big, beautiful website that people love, that we started to realise that there were smaller things there that we could play with.
For example we have just connected just the recommendation part to a mouthpiece on Twitter. What happened though was that we forgot to build the small thing at the beginning, we did it at the end. We’ve started to reflect on this and think about what we would do differently next time.
First of all we want to be even more comfortable allowing ourselves to play. We do this internally, and learn a lot.
When Bill built Chirpy, he had to learn a lot about Raspberry Pi, which he did by reading up about it and watching videos in Safari Flow, which meant he gave us very helpful actual feedback about the product that made it demonstrably better. That all came from play, building something to troll his boss. And instead we got a lot of real understanding.
Another thing we want to remember is that we can do tiny things. We don’t need to build a whole website to see if we can engage people with the idea of a personalised recommendation engine. It could be two people sitting in a room pretending to be a robot by replying to Tweets by hand. You may not notice the difference – and what we’re looking for is the data that shows engagement – a follow, a retweet, a reply, a favourite. Tiny can be really tiny – really small, honest things can be worth releasing to the world.
We also need to get out of the building. The only thing that matters in product development is whether you can delight customers – which you only find out by getting the product in front of them. An email with them, a DM, a tweet – this is all speaking with customers, which should be done as early as possible.
We want to end with a challenge. Go out and play; be comfortable releasing things when they are tiny. And whatever you do – leave the building even if that conversation only happens in the digital space.