Recent years have seen a proliferation of chatbots and so called ‘personal assistants.’ These supposedly serve both the convenience of the user, as well as the goals of a company.
However, these bots often realize poor adoption rates, high user frustration, and limited benefit to the corporation. In fact, they often hurt an enterprise.
Current reality is that natural language applications have severe limitations, failing on tasks that even a child could easily handle. They have no memory of what was said earlier, cannot learn simple but arbitrary facts interactively (unless they were specifically programmed for it), don’t reason about their tasks, and have no common sense.
Imagine having a human personal assistant who could not understand or remember simple facts relevant to yourself and future interactions; like for example: “Always use my email@example.com email”, or “I don’t really like sushi”, or “My boss will be in NYC for the next 2 weeks”. You would reasonably expect a PA to remember this and to take it into account in future interactions, restaurant and meeting requests.
This is the domain of intelligent, hyper-personalization.
The better chatbots today already provide basic personalization in the following ways:
- Statistical, demographic preference prediction such as ‘products you probably like’
- Database parameterization such as name, location, appointments, prior purchases, etc.
- Hard-coded personalization options/questions: ‘Remember this card/address?’
What hyper-personalization entails is a fundamental shift in thinking: The AI bot must be seen as the customer’s assistant, not the company’s. Each user has their own PA, able to interactively learn their individual preferences and situations – a personal bot that actually remembers and utilizes both implicit and explicit information from prior conversations.
Any company that can offer hyper-personalization will be rewarded with much greater customer satisfaction, loyalty, and increased engagement while reducing service costs.
Why is Hyper-Personalization not yet common?
A part of it is corporate philosophy – companies unable or unwilling to provide proper personalization. A simple example: In my years of running a call automation company I’ve come across far too many enterprises not implementing CallerID and call transfer integration – leaving the caller to have to provide and repeat unnecessary information. Even though everyone knows that this is the number one customer complaint when dealing with call centers.
The bigger issue, however, is technological.
Current mainstream AI systems are inherently unable to learn interactively and to deal with such complexity — they lack intelligence. Today’s chatbots are based on what DARPA calls ‘First and Second Wave’ technology, utilizing a combination of ‘machine learning’ and simple conversational flowcharts or dialog managers.
Hyper-personalization requires ‘Third Wave’ technology. It demands AI that has deep, contextual understanding, that can learn by itself, reason and disambiguate, and hold meaningful ongoing conversations. This ‘cognitive architecture’ approach works more like our own mind operates, interactively and in real time.
While there are some smaller companies and teams working on this new type of AI, existing large companies, plus the associated ecosystem of technology vendors, are all focused on ‘big data’ statistical solutions – that’s their strength and thus their focus. These systems simply cannot provide the level of fluid, interactive, contextual intelligence required for meaningful conversation, and no amount of tweaking will imbue them with ‘Third Wave’ capabilities.
As Geoff Hinton, the ‘Godfather of Deep Learning’, opined: “My view is to throw it all away and start over”.
We look forward to a (near) future of highly-intelligent, hyper-personalized personal assistants that truly adapt to the needs and goals of the individual user.