Last updated on April 18, 2018.
Watson seems to be IBM’s brand for anything AI related. You may have heard how Watson competed in the TV quiz show Jeopardy and won. Well, I doubt they used much of that technology in their chatbot tool, but it was a great promotional gimmick.
They’ve actually built a pretty standard, and pretty solid tool, not too dissimilar from TensorFlow. You can define intents and entities, giving examples for each. Once you’ve deployed your bot and got some examples from real users, you can train it further by correcting any incorrectly identified intents or entities.
Trying out the bot is gloriously simple. A button opens a right-hand drawer in which you can chat with your bot. A flashing bar at the top shows if Watson is currently training your model, which typically doesn’t take too long. This aspect of the tool is a win over Tensorflow, where testing your bot is a little more convoluted.
One design decision they have made that is a little questionable is that entities and intents are not explicitly related. Unlike dialogflow, where an intent has an associated set of required entities, this type of requirement is specified in the dialog specification itself. I can define a “node” which looks for a particular intent. I can then specify that certain entities are required to continue. I can also specify questions to ask the user if these are missing.
The end result is that you can build something similar in behaviour to what you would build with tensorflow, just that the workflow is a little different.
I am a little sceptical of this decision. It feels wrong to me that you can define an intent without specifying how it relates to entities. For example, imagine an intent around booking a train journey. If I’m taking the train, there will be a required departure point, a destination and certain time restrictions. These are inherent to the intent itself. But Watson’s design implies that it might be possible to want to take a train without knowing those things. In practice, their approach will probably work, but it just feels wrong…
The free plan allows 10,000 API calls per month, five chatbots, 100 intents and 25 entities. The standard plan has unlimited API calls at a cost of $0.0025 per call. You can have up to 20 chatbots, 2,000 intents and 1,000 entities. They also offer a “premium” plan of unspecified cost if this does not meet your needs.
Choosing the right chatbot tool can be difficult. There are a lot of tools, and their quality varies a lot. The purpose of ChatbotTech is to make choosing a chatbot tool as painless as possible. We've reviewed and rated the top chatbot tools so you can quickly find the one that is right for you.
Sign up below and we'll send you a free PDF containing all the reviews, and ChatbotTech's guide to choosing a chatbot. We'll also send you occasional news and updates (no more than once a week).