Sony AI – 2 from 3

2 of Sony’s AI proposals make sense.

• While Sony appears to be late to the game in AI, it has been working on this for quite some time which has led to 2 out of three of its proposed focus areas making a lot of sense.
• Sony is bringing its AI efforts together into a single organisation which will have operations in Tokyo, Texas and Europe.
• These efforts have been going on for some time as evidenced by a fairly long track record of robot releases which is why RFM has included Sony in its group of companies that are assessed for their AI expertise.
• To date, Sony has fared pretty badly on this assessment and still languishes pretty close to the back of the pack with likes of Facebook, Alibaba and Tencent.
• The robots that it has released to date have been pretty dim and its attempt at a digital assistant went nowhere.
• This may be one reason why Sony has recruited Peter Stone from the University of Texas to run its US AI operation as one of his specialities is digital agents.
• In its initial phase, Sony intends to focus its AI efforts on:
  • First, games: This makes complete sense as any improvement that Sony can make to the AI opponents in its games will make them considerably better.
  • Furthermore, games are the ideal environment for today’s AI as games have the most finite and most stable dataset of almost any task.
  • Data generated by games is also very well characterised meaning that there is no requirement to label the data before it can be used.
  • Sony has a big business in games meaning that improvements here can quickly be leveraged and monetised.
  • Second, imaging and sensing: which is another area that makes complete sense.
  • Understanding the world and accurately characterising it is a critical area because faulty data will produce algorithms that produce gobbledygook.
  • This is why machine vision is a key area for many companies focused on producing AI.
  • As these algorithms become mature, I expect them to be increasingly implemented directly in the camera module giving much better performance as well as efficiency.
  • One of Sony’s most profitable businesses is its image sensor business and offering good machine vision functionality included in its silicon is a good strategy to maintain both market share and margins.
  • Third, gastronomy: which to me looks like a hobby project.
  • Most efforts in this direction centre around a robot that copies what the chef does in order to recreate a dish.
  • The problem with this is the same as with all AI: an unstable dataset.
  • Food ingredients all vary slightly meaning that repeating the exact amounts and actions will result in different outcomes.
  • One of the key skills of a chef is the ability to adjust the technique to deal with the volatility of ingredients resulting in a great dish every time.
  • This is exactly the kind of uncertainty that AI simply can’t deal with at the moment.
  • This leads me to think that this project which, will enable Sony to produce cool concept videos (see here), will end up going nowhere.
• I think that a 66% hit rate for its AI efforts is pretty good in terms of being able to earn a return on these investments leading me to think that Sony’s strategy with this AI division is in good shape.
• However, what is more uncertain is how good the actual AI that it has is.
• The AI-based products that it has produced to date have been poor compared to its rivals and so Sony has a lot of ground to make up.
• This new division will give Sony better direction and purpose for its AI and so I think the end result is a positive one.
• This is one to keep an eye on.