The owner/user of standard bike just uses it for utility, if it breaks they can get around it. Life will go on. Just do what you need to do.
An awesome bike owner/user uses this for fun and to challenge what they can do. A virtual agent mimics the best that can be done (“it’s almost as fast as a car”) but does it with much less componentry than a car.
A chatbot is like a standard bike. They are fairly simple. You click buttons and that takes you elsewhere or maybe you say a sentence and it looks for a particular word. That leads to essentially a tour of FAQs — “Do you have an issue with XXX? Please respond/click Yes or No.”
A virtual agent is different. You write a sentence like you would to a human. It uses machine learning to try to pull out what the underlying requirement of the sentence is. This can be done in many ways, through analysing the sentence structure or comparing similarity to other requests.
Why does this matter? Well, a simple example could be if I say “Yes” to a chatbot it may understand. If I say “yeahh” it may not if it is only looking for words like yes, yeah, yup. A virtual agent would likely understand if it has been trained on the same words.
Now let’s extrapolate that idea to a wider sentence: “I have a problem with my phone”. Good luck writing every permutation of that sentence within a chatbot. Perhaps you may look out for words like “problem” and “phone” but then the sentence “Awesome! You solved my phone’s problem!”. If you survived that, do it again with all possible spelling mistakes. Now put every brand and type of phone in there. You can see where the cracks start growing. We need machine learning here.
Don’t get me wrong — for simple conversational journeys chatbots make a lot of sense. When the bot is required for a huge range of problems though we need a better technology answer.