by Vagelis H. 06/09/2020
We are bombarded by news articles on how artificial intelligence (AI) is advancing and machines are becoming smarter.
People are excited about how chatbots learn how to respond to humans. And they envision the day where a chatbot can be our personal assistant.
But what is mostly omitted from these futuristic articles is that these great results are only possible in very limited domains — for example, ordering pizza or making an appointment — or in meaningless chit-chat scenarios.
People who use Siri or Google or Alexa already know how limited the understanding of these platforms is, once you go away from the 3–4 domains of focus (play music, weather forecast, driving directions).
Suppose you want to build a chatbot to answer questions of people with blood test appointments, such as, what to eat, where to go, how long it will take, and so on.
How can you create a chatbot for that?
To use AI techniques — specifically deep learning — one must have access to huge amounts of training data. In this example, this would be thousands or millions of conversation logs between patients and nurses, talking about vaccination preparation. Unfortunately, such training data is rarely available.
So, is there another way to create a chatbot? Yes, an expert may manually design the chatbot, in the form of a flow diagram. For example, the diagram may say that if the user asks “Is parking free?”, the chatbot should respond “Yes”.
Does this mean that we cannot use any AI in this case? No, we can still use “limited” AI. In the above example, if the user types “Can I park my car for free?” the chatbot may employ AI to figure that this question has the same meaning as “Is parking free?” and hence reply “Yes”. Or, if the AI is smarter, it may also figure that for the question “Do I have to pay for parking?” the answer should be “No”.
So, to summarize, AI can be used to improve the performance of chatbots, but in most scenarios it can only serve as an enhancer to the chatbot logic specified by an expert.
If you want to get more technical, check out my recent tutorial’s slides at https://smartbot360.com/news/.