The chatbots are generally individual AI agents that live inside your favourite messaging app (Facebook Messenger, Skype, Slack, etc) and are specific to the service you need. Some will be able to book you a meeting with a colleague, whilst others will order you your favourite pizza at 1am. The idea is that you don’t have to (god forbid) speak to an actual human being, and the agent will learn your preferences as you do more stuff with them. It also means that you don’t have to open another app to, say, check in to your flight – in theory you can do it all from your chat app.
At the consumer level, where the use of the chatbots is free to the user, the results have, to date, been mixed. There is an initial learning curve (for the human) to master the specific phrases each chatbot wants to hear, and even then they tend to fail as soon as the conversation goes off-piste. Another challenge is that there are many chat apps that host the chatbots. Some will work with some chatbots but not others, so you may end up having to switch apps after all. Each of the major players (Facebook, Microsoft, Google, etc) will all being vying for your loyalty, which means that, as with so much else in the social media space, the user will end up confused and frustrated. They will also end up giving away an awful lot of personal data in the process.
But exactly how clever is the artificial intelligence that powers these chatbots? According to this article from Bloomberg, many of the chatbots have people sat behind them checking and answering the questions instead. Some of these offerings, such as Facebook M, are open that there is a helping human, but pretty much all are opaque about how much of the work the humans do. Bloomberg suggest that in some instances the AI element is hardly engaged, with humans taking the majority of the requests and then ordering the pizza online on another computer.
At an enterprise level, in contrast, virtual assistants reach a whole new level of capability. In many cases it can be difficult to discern whether you are chatting to a real person or a computer (the classic Turing test). At the end of the day you get what you pay for; IPSoft’s Amelia, for example, is able to take in vast amounts of information (the operating manual for an oil rig, for example) and then answer and resolve questions in natural language with the users. What this does prove though is that the AI technology is capable of good work, it’s just not quite there yet at the consumer level.
Another option for enterprises wanting to exploit chatbot technology but don’t have an Amelia-sized budget is for them to build their own (or, more precisely, have someone else build a chatbot specifically for them). In theory this is relatively simple: once you’ve worked out what problem you are going to solve with your chatbot and chosen which messaging platform your chatbot will live on (Slack, Skype, Facebook, etc), then it is either a case of going to a chatbot store (such as Botlist) and picking the best match, or setting up a server and finding a someone to build it for you. Right now there are quite a few companies, such as Chatfuel, that claim to be able to have a chatbot up and running within a matter of minutes, and I’m sure that’s the case. But as we’ve seen above, not all chatbots are good chatbots. In reality there are quite a few hurdles to overcome to create a really good chatbot experience. For example there is work to be done on analytics, flow optimization, platforms upgrades, error checking, API integrations, routing and escalation to live human support, understanding NLP, no back buttons, no home button, etc. There are definitely opportunities to be grasped by any forward thinking enterprise that wants to exploit chatbots, but they must think carefully about how they approach it and what they want out of it. A good chatbot can open up completely new customer channels, a bad chatbot can damage your whole reputation.
And finally, I must leave you with this thought: what would happen if you got two chatbots talking directly to each other? Well, luckily, someone’s already tried this little experiment, and the results are both funny and creepy. Let’s hope that this isn’t a glimpse of a future world run completely by robots…