In an earlier blogpost (and Bloovi article), my colleague Sjoera explained why chatbots have made great progress over the last couple of years. Chatbots appear to be on the brink of a breakthrough: if you can have a conversation with a chatbot that is just as informative as human transaction, and if this bot never needs a toilet break, never needs to sleep and never asks for a pay raise…, then what would stop us from implementing this technology at a large scale?
The most obvious answer to the above question is: context. In any conversation, context is so crucial that it is extremely difficult, even for the most sophisticated supercomputer, to react appropriately to every question, remark or complaint. Just think of phrases with multiple meanings or interpretations. In her article, Sjoera gave the example: “There’s beer in the fridge”. In one context, this can be a reassuring message (“no need to worry about the party tonight, the beer’s chilling in the fridge”), but in another context, it can be an invitation (“Go ahead and serve yourself if you fancy a beer“).
At present, it is next to impossible to render every sentence unambiguous. This is not a real problem, since misinterpretations also occur in human communication. But for a chatbot, it is much more difficult to correct a misinterpretation. Moreover, you can bet your life on it that there will be people who will test the first generations of chatbots to the extreme and who will do their very best to mislead them. If these testers succeed all too often, this would significantly delay the breakthrough of chatbots as a substitute for human communication.
Nicely demarcated environments
If chatbots need conversations to improve themselves but these conversations are met with resistance from the consumer, does this mean that the chatbot is doomed to fail, even before the technology could prove itself? Absolutely not! Speech recognition has also been a laughingstock for years, and yet it is now used by the same skeptics who could not stop themselves from laughing at the time of Lernout and Hauspie. Between the disillusionment caused by Lernout and Hauspie and today’s breakthrough, there is a long period in which speech recognition was successfully implemented in nicely demarcated environments. Some examples are voice-picking – talking to a warehouse system to order goods to be taken from shelves, – and dictation software, which was a popular tool among doctors. The quality of the speech recognition technology benefited from being used in these specific contexts, until it had improved sufficiently and was ready to be released to the general public. This time, however, with much greater success.
I predict the same future for chatbots. At least, without seeing them being thrown under the bus first, I hope. If the initial use of chatbots stays limited to clearly demarcated environments, such as B2B environments, this technology can be spared a lot of image damage, while gaining the necessary experience in a ‘live’ context, making them more and more powerful, year after year.
Warehouses and emergency situations
So, what are the ideal contexts for chatbots right now? Generally speaking: any context that is sufficiently delineated so that the number of potential misunderstandings between machine and human is reduced to a minimum. An obvious example is warehouse management. Due to the limited number of functions and tasks in this environment, this is an excellent practice field for chatbots.
But we can also imagine more complex environments in which chatbots can have an added value. I’m thinking of the dramatic events in Brussels last year. Due to the exceptional circumstances, Astrid, the communication system used by the emergency services, was overloaded, which caused it to shut down and become completely unavailable for use. Suddenly, the emergency services had no choice but to use Whatsapp for urgent communication. Of course, Whatsapp did not offer the same structure and the same features as the Astrid system. This made me wonder: what if a chatbot was listening in on these conversation and this chatbot could inform every party of what the other party was doing? And what if this chatbot learnt from these conversations what is important, and could proactively share relevant information? This would have been a solid alternative for the coordination function that was missing from Whatsapp.
From warehouse management to emergency services – it may seem like a huge step. But both work in a clearly delineated context, in which a chatbot can offer a true added value. While we wait for the development of a next generation of chatbots that can handle information in any context, I already dare to predict a prosperous future for all these communicating digital assistants.