ChatBots through the Ages
Chris Green. GotBot Founder & CTO breaks down how far chatbots have come and what’s next for the AI space.
Chatbots have improved leaps and bounds in recent years based on exponential advances in AI and machine learning technology. An interesting question to examine is- where did we start, where are we now and what does the future hold?
Where chatbots began
The first chatbot was actually created in 1966 at MIT - named ELIZA. Fast-forward a few years, bots have certainly come a long way. It’s easy to think of the evolution of chatbots in a generational timeline -
First-generation chatbots -
These have been around for a while and served very basic functionality. Based on simple, hardcoded rules, these chatbots could only perform “question and answer” style execution which made them rigid and inflexible.
Then came the advent of NLP (Natural Language Processing). NLP in basic terms gives computers the ability to process text and spoken words as humans would speak to each other.
Second Generation -
NLP combined brought about the second generation of chatbots which enabled supervised learning through labelling conversational data which could then be trained through machine learning.
This means a users intent could be predicted based on the trained machine learning model. NLP improved accuracy based on the fact that sentences could be deconstructed into something more understandable to computers.
Actual words or entities could now be extracted to accomplish more useful tasks like performing specific API requests which for instance can authenticate a user, order a pizza or get the weather.
This was a massive leap from first generation functionality however there were still issues like managing context. Managing the context of a conversation is the ability to predict the next action based on what was previously said.
With second-generation, this process was a manual task based on applying rules at the intent prediction level.
Third Generation -
This is where are now.
We have solved the cumbersome rule-based context issues that second-generation bots suffered from.
The solution is NLU or natural language understanding which means on top of NLP, machine learning is also applied to conversational data - making it possible to predict the next action so that context is managed in a more flexible and automatic manner making chatbots more human and less robot.
The Category is -
In order to configure a bot effectively, dialogue should be grouped into three main categories and handled differently.
Question and answer
These interactions consist of the user using a single question and the bot responding with a single answer. FAQ’s are a good example of this and NLP is sufficient in accomplishing this task as context does not exist.
The conversation flow of the interaction with the user is based on the context of what was previously said. For example, knowing what products to reveal to a customer based on their gender which was previously discussed.
Rule or event-based
If a user clicks a button, there should only ever be one outcome -
NLU and third-generation technologies solve all these categories very efficiently and there are currently many configuration tools freely available to accomplish this easily.
What’s next for Chatbots?
The answer - we never see exactly where technology is going but the general consensus within the AI/bot space is autonomy.
Bots learn from their mistakes to teach themselves to optimise their performance. Bots also learn from how people interact with them and from other bots and the internet.
While futuristic movies might (terrifyingly) portray bots as self-learning, automatic and quickly adapting to their environments, we’re still far away from that ever becoming a reality.
However, in the business side of things, bots can now adapt to users queries, upsell products or services over messaging and automate a host of responses and actions. So, in between what we have now and the futuristic bots - we’re looking forward to seeing the hybrid iterations that transition from third to fourth generation bots.