I’ve been noticing a little trend about chatbots lately, so I wondered what’s going on with them? Just a few years ago, chatbots were a bit of a rage. With the hype around IBM Watson winning on Jeopardy, the potential seen in Siri and the introduction of Amazon’s Alexa, it seemed that chatbots were primed to take off. In 2016, we heard all sorts of stories about companies developing chatbots for both generic and specific uses. Here at Perficient, my colleague Eugene Sefanov blogged about Bank of America’s chatbot, Erica. Facebook introduced chatbots into Messenger in April 2016, and by the end of the year, Venture Beat reported that 34,000 chatbots had been built into Messenger. Since then, if you’ve been paying attention, we’ve heard about customer service bots from Staples, KLM, 1-800-Flowers, Starbucks, Lyft, Sephora, etc. Here is a video about Staples’ Easy Button bot from 2016:
So what is the trend I noticed?
It seems that the hype surrounding chatbots has decreased considerably since mid 2017. Ok, that’s not so long ago, but it seems like an eon in the age of artificial intelligence. In doing Google searches, I find lots of articles about chatbot success and potential up until around August 2017. Of course there are more articles since then, but it just seems the number of articles has dropped quite a bit. For such a hot topic, I wanted to understand what happened with chatbots.
It appears to me that chatbots have suffered from the “one of the most overhyped technologies” syndrome that hits so many other early, wonder-inspiring technology. We’ve all seen overhyped ideas that sound amazing, but have a relatively short hype-life. That’s not to say that these technologies don’t eventually live up to their original hype, but often it takes a lot longer and more effort than the hype seems to imply.
I think chatbots are now exiting from this recent period of overhype. Now instead of hype, some are even calling chatbots dead! In January 2018 Wired wrote: FACEBOOK’S VIRTUAL ASSISTANT M IS DEAD. SO ARE CHATBOTS.. You’ll find many other stories about bots that are now defunct. I won’t mention any company names here, because its not fair to call them out for getting out ahead of the curve and experimenting with potentially game changing technology.
Why are chatbots struggling?
Early on chatbots were built the same as instant messaging applications. Most required you to type into a computer or phone to interact with the bot. While this is in the spirit of ‘online chatting’, it’s not what most consumers think of when you mention chat. To me and others, chat is speaking and not typing.
As an experiment, I recently tried 1-800-Flowers’ Gwyn bot. This service bot is in beta still, but the idea is to help you with ordering gifts, a seeming pretty simple concept. However, when I opened Gwyn’s page in Safari on my iPad, I saw no way to interact with Gwyn. In fact there was no input field to type into. I thought maybe this used voice control (yea), but I couldn’t find a way to speak to it (boo). I checked it on my phone, and there I found the typical input field, but no voice control. Most people would have expected a voice experience on the iPad or Phone. Maybe the chatbot interface is one problem.
However, we’ve seen tremendous market success with Amazon’s Alexa service. Apple Siri has been pretty popular. Google Home has also been received nicely in the consumer market. I think the voice interface of each of the platforms has been a key contributor to their success. So a voice interface is probably better for chatbots too.
Understanding intent is hard
Another problem is that it is challenging to build conversational systems to understand a user’s intent. Amazon Alexa, Apple Siri and other ‘personal assistants’ are solving this problem on a daily basis, but sometimes still struggle to understand intent. In a series of articles about creating a chatbot, Gidi Shperber at Towards Data Science.com put up this image:
Chatbot failed intent
This is an extreme example of failure to understand intent, but you can see how serious a mistake could become. (In fact, I tried this out recently with Siri, and she actually did offer to call emergency services. So kudos to Apple for solving that intent issue.) However, training our machine learning tools to understand intent for specific tasks can be easier. Understanding that “get me a taxi” means that I want a Lyft ride from my current location is within the realm of possible right now. In general, getting to 80% accuracy for specific tasks is more doable and probably sufficient for most.
Chatbots are not dead
In contrast to Wired, I don’t think chatbots are dead. I think chatbots have a lot of potential still. All the experimenting that has been done so far indicate that the chatbot interface is one key to its success – more voice and less keyboard. In addition, making your chatbot specific to tasks is more likely to be successful, but still requires a significant effort to make it work intuitively and accurately.
So don’t believe the hype of chatbots’ birth and death. There are plenty of examples of good, well-working chatbots. We just need to work through these hype periods and learn from the early experimenters.
