The Rise of Messaging Bots

3 minute read

Bots aren’t really anything new. It’s hard to track down exactly when they appeared, but they exist at least since 1992 where they were used to automate tasks in the first standardized chat of the web, IRC. Since then, they went through a few hype cycles, every one of them impacting more people, extending from technical to more consumer audiences.

Bots are appealing for a simple reason: they are meant to feel “like people”, they can understand commands and do tasks. The hype cycles happened every time the technology improved enough to create a sensible change in their ability to either understand commands or execute tasks.

While voice activation is captivating and getting better every year, the reason of the current raise of the bots is the pervasiveness of messaging tools that in 2016 range from hundreds of millions to a billion users on each platform, with Facebook, Whatsapp and WeChat at the top.

This means that while IRC bots reached at best a few hundred people at once, this generation doesn’t just have a reach of about 14% of the world population, but is also far more integrated with better awareness, better understanding and better computing power to give complex answers.

However, they are still far away from being truly natural, and this is a key understanding to not buy too much into the hype. Natural language understanding is way better, but is still limited. Their ability to give proper answers has increased rapidly, but bots aren’t still truly aware of the content they reply with.

Today’s technology allows bots to shine, but only under these conditions:

  1. Expert: they need to be experts on a specific field, not generalists.
  2. Able: they need to be well connected to the tasks they are meant to execute.
  3. Horizontal: their field of expertise needs to allow a wide range of shallow questions and mostly direct dialogues.
  4. Unprimed: the questions they get asked are already in the mind of the user.

Expertise is important because a specialized bot can be far smarter in understanding the question and providing the answer as it excludes anything that isn’t related to its field of expertise. For example, a bot that is contacted to do troubleshooting doesn’t have to be aware of everything that can be asked, but only to the topics related to troubleshooting. It doesn’t have to guess it’s doing troubleshooting, it’s already pre-determined.

Ability is similarly important because the bot needs to be able to read information, operate and give answers that are on point. A bot that isn’t connected is a bot that is only partially useful, as the users will have to do the task themselves. This means direct API through other relevant systems, or sensors to understand the environment the person is in.

Horizontality is particularly interesting, and also often overlooked. Given it’s currently hard for bots to grasp the wider context of a discussion, they are most effective when their expertise can be expressed either as answers to direct, individual, questions, or by driving themselves these questions.

Unprimed questions are also far more effective as messaging apps are a form of dialogic interface, and dialogic interfaces don’t show much interface beyond a blank field and a way to type or dictate a question. Normal interfaces are intuitive because they show the user what they can do, but dialogic interfaces are much more opaque, thus require the user to already know the question in advance, and the bot has simply to answer it.

This of course doesn’t mean that all bots must strictly follow these points, just that given today state of technology these are bots that can be built efficiently. Further advancements in both natural language processing and artificial intelligence for example will allow them to be even better, but that requires a higher degree of investment.

While the future for bots is going to be quite incredible, the present is already quite interesting.

This article was originally written for Ernest‘s blog, a startup I’m advising.