Apple – Edge issues

Xnor.ai will not fix the fragmentation issue.

• Apple’s acquisition of Xnor.ai increases my concerns with regards to Apple’s approach to AI and may make it fall further behind.
• Apple is reportedly paying $200m to acquire Xnor.ai whose speciality is the ability to efficiently run large machine learning models on edge devices.
• There are obvious advantages in being able to do this in the short-term, but I think without the agent being primarily resident in the cloud, any personal digital assistant is going to have real problems in delivering a decent and consistent experience on different devices.
• I think that there are two things that have driven this acquisition:
  • First, Google: In May 2019, Google announced that it was now able to run some of the core models of Google Assistant on the phone.
  • The main one is speech recognition which makes the assistant much quicker and responsive as the analysis and inference of the request are processed on the device before being sent to the cloud to get an answer.
  • This puts Google yet another step ahead of Siri increasing the already very large gap to its biggest competitor on the smartphone.
  • Xnor.ai will help Apple also run its machine learning models more efficiently on the device.
  • Second, Privacy and security: Given that users pay Apple for access to its ecosystem when they buy the device, this is an area where Apple has pushed hard for differentiation.
  • This drive has pushed it to implement Siri locally on all of the devices within its ecosystem with no user information being relayed to the cloud.
  • I have long viewed this as a huge mistake because it means that Siri has been fragmented with a different version of the assistant on different devices.
  • This has resulted in an inconsistent experience where different versions of Siri on devices owned by the same user are not aware of each other and have different capabilities (see here).
  • However, Apple seems to be sticking with this strategy and the acquisition of Xnor.ai will allow Siri to do more on each device but will be, in effect, doubling down on this choice.
• The acquisition of Xnor.ai should make Siri and other AI algorithms run more quickly and efficiently on Apple devices.
• It should also allow Apple to move more processing (such as speech) onto the device to give a more responsive and rapid experience.
• However, I think it will also exacerbate the fragmentation of Siri meaning that the cross-device experience for users will worsen.
• The optimum digital assistant experience is still one where the assistant resides in the cloud such that a more consistent experience is delivered across the whole ecosystem with no fragmentation.
• Everyone except Apple has chosen this route and I continue to think that this remains the single biggest block to Apple closing the gap om its peers.
• There are ways of getting the both of both worlds, but this will not work well for digital assistants.
• An example of this is Anagog which builds a profile on the device of the user and then polls the cloud and downloads information that is only relevant to that profile.
• This works very well for delivering targeted advertising while maintaining privacy, but I don’t think it will work very well for digital assistants.
• The net result is that I think that Apple is going further down a road that hobbles its ability to deliver high-quality AI meaning that it will not be closing the gap on Google.
• This is not going to hurt Apple’s appeal in the short-term as phones are still selling predominantly on cameras and battery life.
• However, as AI becomes more and more important in delivering the ecosystem experience, Apple’s advantage in the overall user experience will begin to erode.
• Hence, I continue to think that taking profits on Apple’s long-rally is not a bad idea.