Interesting discussion on the use of voice interfaces and services. There is a certain amount of cheerleading involved in the talk; but that is to be expected with vendors in the room. If found it interesting that one of the panelists; Sam Liang of AISense moved out of where2.0 services and into voice. because location is a great gateway to lots of rich contextual information and voice is desperately in need of context and by extension user intent.
It is interesting to get a perspective on the organisations involved in the discussion on voice interfaces:
- SRI International
All of them seem to be well behind where the telecoms voice services managed to get to like Orange’s Wildfire.
Key takeouts from this:
- 50,000,000 voice devices to ship this year (2018). A total installed user base of 100,000,000 (presumably excluding voice interfaces on smartphones)
- AISense is looking to build in voice biometrics that would prompt you about who a person is. Privacy implications are profound
- The panel struggled to articulate an answer to privacy concerns beyond ‘services need to build trust’ and transparency
- Information security and hacking wasn’t a point of discussion; which surprised me a lot
- Context still seems to be a huge issue, I think that this is a bigger issue than the panelists acknowledge. Google still struggles on user intent, without adding the additional layer of understanding voice. The biggest moves seem to be ‘social engineering’ hacks, rather than improvements in technology
- Amazon and Microsoft don’t have plans for advertising services on voice (at the moment)
- We’re very far away from general purpose voice services
- Work has only started on trying to understand emotion
- Orange’s Wildfire and The Register on its shutdown. Wildfire’s problem seemed to be a failure of marketing more than anything else. We haven’t seen anything else like it. Even Siri is only scraping over the ashes of the work done on Wildfire 15 years ago
- Google’s published research on speech processing. What becomes apparent from looking at the list of research is how basic the current state-of-the-art currently is
- Stuff that I have written that touch on context dependent services