Are AI chatbots really intelligent?
The first thought that we get, as we come across a chatbot is that, if they are really “intelligent”. And if they are, then how? Is it some sort of Artificial Intelligence? Prior to entering this field, I thought that the chatbots in retail are just fed with numerous Question and Answer pairs(FAQs). In reality, a ChatterBot seems to be a box of surprises.
What makes a conversational AI chatbot intelligent?
A friend of mine said, “chatbots remind me of kids”. I thought it might be an interesting example to explain how chatbots work. Remember how our parents taught us to say thank you? The picture illustrates 3 scenarios, which are kind of self explanatory.
What is interesting here is the 3rd scenario, where nobody is teaching the kid to say thank you, but she still does.
Here we should notice 3 things:
The child understands from its previous experiences that the 3rd scenario is also about giving and taking.
The child understands it has received something pleasing.
The child knows it should thank the other person every time it receives something from them.
How did the child learn this? Here’s where human intelligence plays its role. Similarly, when a machine tries to do the same, it needs some intelligence too – which is nothing but Artificial Intelligence.
Can a conversational AI bot learn?
A conversational AI chatbot just cannot reply to whatever we ask, unless it knows and understands the language we speak. Knowing the language helps the bot to understand the semantics of textual data. The concept of language models help us represent any textual data from a language in the form of numbers, preserving all the semantics! So, does that mean a chatbot can speak any language if we create a language model for that particular language? Indeed!
So now that we know that our ai chatbot can learn our language, it now needs to –
Understand the intention of the user
Identify any specific entities mentioned in the query
Keep track of the context in which the conversation is happening
Generate a reply to be sent back to the user
Relate to the kid’s example above? The way the kid’s mom taught her, our ai bot too needs someone to teach them. Basically, we need to teach our bot how to perform the above tasks. In technical terms, we call this Machine Learning.
Do you see how all these fancy terms come together to form a chatbot? Sweet!
So, can a AI chatbot answer anything and everything?
Well, depends on how you train them. Usually chatbots are trained to serve a specific use case. It may be for example a customer service chatbot, an ai assistant for retail shopping or could be even industry specific such as chatbots in retail or chatbots in fashion industry. These domain specific bots are more powerful due to their specificity and hence are more popular as well. However, generic chatbots can also be built which can address more general queries which could be as vague as “what should I do today to stay positive”. These chatbots require a huge amount of data and are often the hardest to train.
Chatbots are not simply hardcoded machines but use Artificial Intelligence and deep learning techniques to learn from given data to try and imitate humans while conversing.
Chatbots can be trained to serve any specific AI use case or to carry general human conversations depending on the need and specificity.
Do have a look at “Are AI and Machine learning the same” by Tejas Venugopal to get a clear picture of AI, Machine Learning and Deep Learning.
Also i recommend the article by Joseph Mathew Traditional Online Chatbot or Domain Specific Conversational AI? What can create a delightful customer experience online? to know why domain specific chatbots could be a better choice for your business.