Some fascinating insights here, particularly the focus on smartphone Virtual Assistants (VA’s). Few people doubt that AI is the future of Virtual Assistants. Where opinion varies is what we do now, and how we get there.
Demis Hassabis talks about a “dichotomy between pre-programmed and learnt” intelligence, but on the ground, today, there really is no dichotomy – pre-programmed or designed intelligence is the only commercially deployable solution for VA’s. Learned intelligence is fundamentally not ready for commercial deployment for broad or narrow-scope VA’s. RBS’s recent announcement of their work shows whats possible using pure AI – 20 intents and a 30% resolution rate is likely to be some way from being commercially-viable.
With designed intelligence it’s possible to handle up to 1,000 intents with machine-learning derived natural language understanding, achieve an 80-85% resolution rate, an excellent user experience, and for a realistic investment. The remaining 15-20% of instances can be handled effectively through careful design and thoughtful integration with alternative channels. This is the reality today for companies deploying VA’s and natural language interfaces for customer service, and it delivers a very strong ROI and a great user experience.
From this designed experience, the path to AI is clear. There will be no single ‘big bang’, but rather as AI is adopted it will be used to address the ‘long-tail’ of user intents, with the high-runners continuing to be designed to optimise the user experience.
Where VA’s are used for customer service, the enterprise has a massive advantage, not only from a scope that is naturally constrained by their business, but from an army of customer service advisers or agents in call centres who can receive suggestions from AI and help train the intelligence through selecting the most appropriate response or adapting responses for customer consumption, and this can be done in real-time. This trains the AI and helps formalise the wealth of knowledge that resides in the agents heads – the very best knowledgebase an enterprise possesses.
This approach provides massive benefits today to forward-looking businesses, with a clear and realistic path towards AI in customer service over the next 5 years.