Siri – Mortal sins

Reply to this post

Siri must fly to the cloud.

• Siri was the first digital assistant to hit the mainstream but Apple’s focus on privacy and the unusual way in which it has implemented Siri, has led it to fall from being the trailblazer to the also ran.
• This issue has really come to a head with the HomePod which by all accounts, is a great speaker but has underperformed expectations quite probably due to the poor quality with which Siri has been implemented upon it.
• Siri has three fundamental problems:
  • First, differential privacy: which is the system that Apple uses to ensure that user data in its cloud is unintelligible to any one but the user himself.
  • This works by inserting random code into the data as it is uploaded to the cloud such that it makes no sense when it is viewed in isolation.
  • However, when put together with thousands of other pieces of data the random pieces of code cancel each other out.
  • This gives Apple the ability to understand broader trends of its usage but not what individual users are doing.
  • Apple has used this to promote its’ security and privacy and as a reason to choose iOS over Android.
  • However, RFM research (see here) has found that this limits the quality of the algorithms that Apple can train as the data set has been compromised.
  • Consequently, although Apple collects only slightly less voice data than Google does with its assistant, I believe that it is unable to make use of most of it.
  • This has been a substantial hinderance to its ability to improve and become more intuitive and useful.
  • Second, fragmentation: Apple has committed the mortal sin of the technology industry by fragmenting the Siri platform.
  • By implementing the agent on the device rather than in the cloud, Apple has ensured that the user experience of Siri varies from one device to another.
  • This is an understandable decision from a company so rigidly focused on devices and differentiating with security and privacy, but this decision is now beginning to show some problems.
  • For example, Siri on Apple TV has a different set of skill domains than on the iPhone and the HomePod is incapable of doing some of the most basic things that are available on the iPhone.
  • This also limits the ability to use Siri to control devices from one another.
  • Furthermore, the third-party skills are all implemented on the device meaning that if the user enables Siri on the iPhone to call an Uber, all of his other devices will remain unable to do it.
  • It would appear that there is a central repository where the master code line of Siri is developed which is then checked out and customised by product teams to make it more applicable for the device they are creating.
  • This results in fragmentation and inconsistency that wastes R&D resources and hurts the user experience.
  • This is exactly the problem that contributed to Samsung’s and Motorola’s inability to mount a serious challenge to Nokia in feature phones 15 years ago.
  • In Apple’s case I think that this a reasonably straight forward fix by moving Siri completely to the cloud which would be a great feature to announce at WWDC this year.
  • Third, cloak and dagger: Apple is a closed and secretive organisation which does not work well when it comes to the AI community.
  • Siri is driven by AI and the AI community works on openness and collaboration which it at odds with the way that Apple does things.
  • For example, DeepMind published its method for creating AlphaGo which, in my opinion, was then immediately copied by Tencent to create its own AI Go player.
  • Apple has opened up a little bit and has begun sharing and publishing some of its methodologies for Siri (see here) which will need to increase with time if Apple wishes to benefit from the learnings of others in the community.
  • This does not mean giving away key proprietary and differentiating algorithms, but being more open about solving the many problems that everyone faces when trying to develop intelligent machines.
• These are the reasons why I believe that Siri is falling further and further behind.
• 15 months ago, I rated Siri as broadly on par with Amazon Alexa but since then Alexa has made significant advances that have eluded Siri.
• This problem now appears to be coming home to roost as Siri’s poor showing on the HomePod may now be affecting sales of the product.
• This is not yet a big problem for Apple as a whole but Siri’s trajectory is pointing towards it becoming a liability rather than an asset to the iOS ecosystem.
• Unless fixed, Siri could end up being a reason not to chose iOS rather than a draw to Apple’s products.
• Siri needs to fly to the cloud in the first instance and then worry about its overall low IQ compared to Google, Baidu and Amazon.