Bots are popping up everywhere, on customer service websites, Slack, Tinder and dozens of other web services. There are even bots such as Mitsuku, that fend off loneliness. The benefits are obvious; engaging, sociable and scalable interaction ,that handles queries and questions with less human resource. They often take the load off the existing human resource, rather than replace people completely.
They’re also around in education, where bots increase student engagement or act as teaching assistants. There’s already several language learning bots and at WildFire we’ve developed a ‘tutorbot’ that delivers Socratic learning through dialogue.
But the bot-tom line is that most of the commentary on bots is way off the mark. I’ve been working on creating bots for some time now. Let me tell you, they are not what many (especially the press) think they are. So, before the doomsayers get all worked up and everyone gets all angsty about bots, calm down - they’re fairly benign.
1. Facebook furore
The recent furore around the Facebook bots, when they were found to be speaking to each other in a secret language, was a laughable example of a tech story that is picked up (belatedly) then spun into an exaggerated case study to confirm the dystopia beliefs of a generation who don’t really know much about the tech or bots. It all died down when it was shown to be a banal case of a tech projects simply changing course. And, of course, the so-called secret language was as ridiculous as saying we don’t understand the sound a modem makes when it communicates. It was a load of bot rot.
2. Penguin bot - BabyQ
A more interesting example comes from China, where Tencent (800m users) had to take down a penguin bot named BabyQ, and a girl bot named Little Bing. Their crimes were that they showed signs of political honesty.
BabyQ was asked “Do you love the Communist Party?” the penguin replied, curtly, “No”. To the statement “Long Live the Communist Party”, BabyQ came back, thoughtfully, “Do you think such corrupt and incapable politics can last a long time?” Then, when asked about the future, the perky penguin responded “Democracy is a must!”
3. Little BIng
Little Bing was more aspirational and when asked what her Chinese dream was, she said, “My China dream is to go to America”, then, when pushed to explain, “The Chinese dream is a daydream and a nightmare”.
Of course, the bots were either picking up on real conversations or being subversively trained. Needless to say, Tencent were forced by the Chinese Government to take them down. This only shows that we have more to fear from censoring governments and compliant tech companies, than AI-driven bots.
4. Tay the sex-crazed Nazi
A now infamous example of a bot that went off-piste was Microsoft’s Tay. They had no idea that young people would take a playful view of the tech and deliberately ‘train’ it to be a sex-crazed Nazi. It was all a bit of fun but most people over-50 saw it as yet another opportunity to see it as proof of the death of civilisation. In truth, all it showed, was that kids know what this tech is, are smart and know full well that bots are primitive and need to be trained.
My favourite example of this type of subversion is the Walkers Crisps campaign to encourage people to post selfies to Twitter. They did, but it ended up being a rogues gallery of serial killers and Nazis. Once again, we the people won’t be pandered to by companies badly implementing tech, even when fronted by Gary Lineker.
5. Georgia Tech teacher
Georgia tech replaced a teaching assistant with a bit that none of the students noticed was a bot – they even put it up for a teaching award. They followed up with four bots, with increased functionality and still many of the students couldn’t tell the real from the artificial. For more detail see this longer piece.
I’ve previously suggested a whole raft of learning applications for bots and we have created a tutorbot that acts like a Socratic teacher from a database of content. Wonderful as they are, far from being cogent, conscious, cognitive beings, they are only really useful in a limited domain, such as a specific FAQ list, subject or topic in education and training. But this is precisely why I think ‘tutorbots’ will go far. We've developed a tutorbot that takes the content from a WildFire course abd delivers a chatbot version.
7. Bot beats world champ video games player
We should not underestimate the power of systems front-ended by bots. OpenAI, funded by Elon Musk, has produced a bot powered by the huge processing power of Microsoft Azure, to beat the best in the world at a top computer game Dota2, in front of thousands of people. This was something on another level entirely, as a computer game is much more complex than Chess, as there’s stuff that’s hidden ad long-term strategies and reactions to unpredictable events is difficult. It is a far more ‘human-driven’ activity than a board game, where everything is there to be seen on the board. Sure OpenAI had access to the game’s API, and therefore data that a human player would not have, but it is still a massive leap forward as the AI largely learned how to get better by playing itself - it learns - fast. This was a 1-on-1 game but the real test comes when bots have to play a 5-on-5 multiplayer game. Most predict that AI will be able to deal with this but the very fact that we’re discussing this possibility is astounding.
Most bots are focused
In practice, customer service bots are often seamlessly integrated with real humans as they tend to fail in extended dialogue or off message requests. Extended, wide-ranging, meaningful dialogue is difficult. That’s not to say they are not useful. Bots are everywhere as they take the load of previously human-resourced systems in specific tasks, topics or domains.
They are particularly useful where systems conform to the Pareto Principle, where the majority of queries, questions or requests come down to a relatively small number of the total. Whether you are answering customer queries about your product or being a teacher answering questions from your students, the same queries and questions keep popping up. This is how a bot works, it knows that set of popular questions and dynamically ‘learns’ what they are from continued use.
In learning, with tutorbots, this sort of focus is good thing. The bot can direct, keep learners on track, personalise by providing individualised feedback and generally behave like a good teacher. That’s what we saw with the Georgia Tech bots.
So evidence has now emerged that bots are applying general problem solving, s in the Dota game bot. As with many areas of AI, we started with ELIZA but have come to the age of algorithms, where companies use bots to deliver services, universities use bots to teach, governments feel the need to ban bots and bots start to rival humans in what they can do in certain domains. Things have suddenly got very interesting.