1st Chatbot Summit in Israel 2016. (CC SA 4.0)
It’s so hard to speak to a real person on the phone these days. Almost any time you need to call your bank, doctor or any other service, you’ll probably be greeted by an automated service seemingly designed to prevent you from speaking to someone who actually works for the company. And that could soon get worse thanks to the rise of chatbots.
Chatbots are artificial intelligence programs, often deployed in apps or messaging services. They are designed to answer people’s questions in a conversational style rather than just pointing them towards information like a search engine. Companies such as Uber, Lufthansa and Pizza Express are already using them to field customer enquiries and take bookings, and many more are on the way.
They have the potential to improve some aspects of customer service, and are certainly easier to use than automated phone systems that struggle to understand even your basic personal details. But they’re also another hurdle separating customers from a human who can actually answer difficult questions and, crucially, show the compassion and good will that strong customer service is often based on. There’s a chance that chatbots could cause both consumers and companies to find this out the hard way.
Automating customer service, or at least part of it, is an inviting thought for many companies. Not only can it reduce the exposure of human employees to many of the unpleasantries of the job, it can also help screen out many common or trivial problems early on before the expensive attention of a human is needed. This could help companies reduce costs while calming customers who just need simple solutions to standard problems.
But replacing human employees with artificial ones isn’t that straightforward. To begin with, language, with all its variations and errors and despite really impressive progress in automatic recognition and translation, is still a tricky issue. Automated representatives are still significantly less competent and language-aware, and for some problems would be difficult or impossible to communicate with.
Good but not good enough
Talent is the ability to perform well. Mastery is the ability to fix an unusually difficult situation. There is an art to handling the exception, and good customer service is often about the unusual or unanticipated cases involving potentially angry customers. While chatbots can convincingly source answers to basic questions, AI isn’t yet smart enough to deal with the rare and exceptional examples.
Companies might not initially see this as a problem, as it introduces a way to separate customers whose service requires going the extra mile. Only those whose problems confuse the bot need to be passed on to a human employee. But going through this frustrating process of talking with a bewildered computer is likely to make the customer more angry with the service. In the long term, this could encourage them to take their business elsewhere, especially if it is difficult to get a human representative to come to the rescue when the bot fails to help.
I experienced a version of this issue myself when trying to book a cab following a train breakdown. When I called the one local company I had a number for, I was put through to an automated service that was completely unable to understand the pickup location, in all variations of naming and pronunciation that I could think of.
Through some serendipity, I was connected to a human representative, but before I had time to utter my predicament, he told me “I shall put you through to the booking system” – and the infernal loop resumed. The sad story ended with a very long walk, a lucky pickup by a human-driven black cab in an otherwise utterly deserted area, and a vow to henceforth shun the first company whenever possible.
Automated systems might be able to handle regular cases. But they can’t yet adapt themselves to exceptional circumstances or even recognise that the flexibility of human intervention is needed. And the problem, from the consumer’s point of view, goes further than that. Some situations require not just human understanding and problem-solving, but a level of compassion and empathy.
A chatbot can be programmed to adopt a certain style of interaction, but that will still sound oddly out-of-place in unexpected or difficult contexts. There is currently no practical road map in AI research for how to implement something that convincingly resembles human compassion.
Sometimes angry customers need kind words and the chance to express themselves to someone willing to listen, as well as or even sometimes instead of actually having their problem resolved. And often good customer service relies on gestures of good will made at the discretion of individual employees following their own feelings of empathy rather than a set of fixed rules.
This would be very hard for AI to replicate because it depends so heavily on the context of the situation. In my opinion, context understanding is still one of the major elusive and unsolved problems of AI, and is likely to remain so for quite a few years to come.
Despite this, the cost-saving promise and other benefits of automation appear so appealing that chatbots and other AI customer services are still set to drastically expand over coming years. The likely outcome, at least in the mid-term, seems to be an even more technocratic treatment of complaints with less flexibility. Or worse, once the algorithms become increasingly refined, an opaque decision-making process with very little room for the mellowing intervention of a human supervisor.
If we want to avoid this, we need to realise that the way to help is not paved with good intentions, but found in appreciating how limited AI currently is at understanding contexts, exceptions and the human condition.
About the Author
Daniel Polani, Professor of Artificial Intelligence, University of Hertfordshire