The future of Artificial Intelligence and Virtual Agents seems to be promising, and we are getting to the point where a bot will know what we are saying, but it might be a while before it knows what we mean.

There is a slight difference here, because a bot that knows what we are saying, could have been ‘trained’ on a repository of pre-built answers, and given enough terabytes of pre-fed dialogues, could maybe also pass the Turing Test! But would we really call that intelligent, and would any volume of pre-fed dialogues be enough to match the human minds ability to try & trip it over.

A bot that knows what you mean, would understand & assess and then respond. And also learn or unlearn as the case maybe, and such a bot would effectively be able to side step what happened to Microsoft’s bot Tay, that got ‘trained’ to become a racist in less than 24 hours based on the twitter messages it read & got trained on, parroting them back to the users without actually knowing what they ‘mean’.

Artificial intelligence that understands what you mean, has been the long held ambition. Avid lovers of technology may have gotten used to the many stories and predictions around artificial intelligence year after year. If you can recall, in 1957, Herbert A. Simon predicted that it would take less than ten years for a PC to win the game of Chess. It of course didn’t materialize until in 1996, close to four decades later.

Similarly, Marvin Minsky in 1970 predicted that a machine with the intelligence level of a man would be unveiled in less than eight years. Of course, it was yet another funny scientific mirage.

Did these and other stories depict a field full of many over-ambitious people? Certainly not!

Yeah, the long-awaited technology is here

The many intelligent, pioneering AI prodigies & patrons may have been wrong in their predictions, but certainly they had a great vision. There are strong supporters like Google, with its investment in Deep Mind, and also opening up TensorFlow. Let’s not forget IBM’s Watson and Amazon’s Echo and the list goes on.

The expectations are higher than what the 1997’s ground-breaking discoveries elicited. Unlike the technology of such times characterized by the famous brute-force computing using decision trees and algorithms for decision-making, the recent one is an “ice-breaker.” In a bid to create a machine so human that it learns from examples, the rush to finally make Artificial Intelligence and Virtual Agents real is at its full speed.

However, if we are to take a step back, the ability of the bot to understand the ‘meaning’ comes from the element of context. If for instance you, dear reader & I were in a room where for some reason, someone had also left an elephant, and I exclaimed ‘Why is it here’, you would have the context and therefore be able to respond back appropriately. If however you were on the other end of the phone or messenger, you would not have the context and hence your ability to respond back intelligently gets limited.

Why, even though we are asking, Chat bots might not have & so be the answer

Whilst a lot of effort & almost all the investment is going towards building bots that can derive the context on their own, and so the conversation can go anywhere (open domain systems), we are definitely not close to that as yet.

And so Enterprises face the issue of having chat bots that are being used to service customers, and when they get asked an unscripted question, it’s a wrong answer or no answer. Couple that with the fact that 80% of millennials don’t return back after one bad customer experience, and the money spent on the bot might not return the dividends it promised.

Another factor that influences accuracy, and that might be inherently against the chat model, is the length of the conversation. That is to say shorter conversations yield higher accuracy, and the longer the conversation the more difficult it becomes to generate the accuracy.

However, all is not lost, and the answer is obvious - if the context can be brought in the dialogue, so that the interaction can be focused ones, with pre-set expectations, the accuracy of the interaction goes up multi-fold. So borrowing from our previous example, if you were on the other end of the messenger and I shared the photo of the elephant in the room with the caption ‘why is it here’, you would have the context, the dialogue could be kept short and you would be able to give an intelligent response, maybe something like ‘Because it’s not your office, but the zoo. So the question is why are you there?’

So until we get to the point where the bots are able to discern context on their own, in the near term I suspect once all the hype & mad rush to build chat bots has mellowed down a little & once the accuracy & resulting customer experience is clear, almost as a mid-step, enterprises will eventually move away from Q&A based chat bots and adopt some form of closed domain bot system, still over the chat window. This is so that they are of course able to leverage the benefits of the social messaging & chat platforms that have become the default digital haunts of the millennial & certainly Gen Z, and also to ensure that the service & advice being provided is personalized, accurate & worthy of the customer experience such large brands aspire to enable.