A lot of chatbot projects in regulated sectors, especially financial services, have reached the realisation that Natural Language Processing (NLP) has severe limitations compared to the unabated hype.
NLP is powered by machine learning.
Let’s call it a more realistic name. Machine-controlled learning. There are several suitable conversational applications for this capability. In the regulated world, there are matters that must remain under human-controlled learning.
There are three distinctive areas not suitable for Chatbot NLP as it is simply inappropriate to empower machine-controlled learning:
- Moral and ethics, which includes the need for politically correct dialogue, as this is so complex that you cannot handover control to the machine to make these judgements
- Regulatory and statutory, as this is so complex you cannot empower the machine to rewrite the law
- Standard Operating Procedures, containing policies and practices interwoven with regulatory and statutory rules, as this is so complex you cannot empower the machine to rewrite say policies
It is a shame so much money and time has been lost developing chatbot NLP when finally there is a realisation that they cannot turn on the machine learning if they go live when it involves the above areas.
The technology for conversation-as-a-service in the regulated sectors is far more advanced than most firms and chatbot ‘experts’ realise. However, it is time the business understood the 101 of chatbots to make informed and reliable decisions for using chatbots
to drive digital transformation. Once a blended approach of machine-controlled learning and human-controlled learning is understood then material change will happen very quickly as conversations-as-a-service becomes mainstream.