In several recent notes I’ve written I have mentioned in passing this term personalized machine learning. It is impossible to talk about artificial intelligence in a broad sense without talking about, and understanding, machine learning.
At a high-level, we are still many years away from a true artificial intelligence, meaning a computerized simulation of human intelligence which is capable of engaging with a human just like another human would. I say “years away” because of several factors. Firstly, computer scientists are still developing and modifying the underlying math/algorithms which make the foundation of machine learning. We have decent natural language processing as well today but we still can’t have a generic and normal conversation with a computer. Devices we speak with today are still capable of taking our words, turning them to text, then running them as a query on the backend trained network to deliver a response. What computers can not do, is understand us. More deeply, they don’t know us, understand our context, or pick up on nuances in our tone or voice. These are all things necessary to develop and also some of the most difficult problems computer scientists are trying to solve.
The second challenge facing machine learning is raw processing power. We simply don’t have enough processing power in our devices, or in the cloud, to take massive quantities of data in real-time and process, train, or learn from that data nearly instantaneously. We are literally at least five to eight years away from that kind of processing power and in reality more like ten.
However, while the biggest companies in the world will spend most, if not all, or their efforts in machine learning around general data/information, the area I’m watching the most is personalized machine learning. We will no doubt achieve a more generic or general AI well before we have a personalized one. A general AI will be one that is capable of understanding things in a general sense, or with general type data like news, sports, and other types of communal data where massive public data sets are freely available. To a degree, some of these AIs will feel personal as companies like Amazon or Google attempt to add elements of personalization to their version of these services none of them will truly be personal or understand me in intimate ways like a close friend or family member would. This type of true personalized machine learning is the hardest one to accomplish, the one with the highest payoff in terms of consumer value, and also one only a few companies are in any kind of position to execute.
I argued in this article called personal vs. communal AI that Apple and Google are the two companies closest to the consumer in terms of seeing intimate details of their lives. Google may or may not be attacking this the same way Apple is because of their business model, but Google has been clear their mission is to organize the world’s data (not necessarily your data) where Apple may be better positioned to focus on the concept of personalized machine learning that Google is given their market strategy. I saw this for a few reasons.
Out of both companies only Apple has personalized their assistant. If we are going to let a machine become our true assistant it will need to be anthropomorphized so we develop somewhat of a relationship with it. Google has not done this which is telling of their strategy. Amazon, Microsoft, and Apple are closer to where I think the market will respond to assistants by giving them names. But, only Apple out of all these companies is closest to the consumer. Which is why I think Apple is more likely to focus on personalized machine learning than any of the main platform companies of today.
You can actually see Apple’s first attempts at this in some small, yet significant examples not so hidden in iOS. For example, when someone texts you “where are you” you see a share location auto-response option show up in the text field. iMessage understood the context of the text message and a useful response would be a way to instantly share your location with whoever just asked. Or if you start typing “my cell is” in any text field you will see your cell phone number show up in an auto-response making it very easy and quick to share your number. Another one of my favorites is when using Yelp to find a restaurant or coffee shop, when I go to iMessage and tell someone to “meet me at” the last place I looked at on Yelp shows up as an auto-response. There are countless numbers of examples from iOS that I’ve observed that are trying to get at a more personalized experience which is why I feel Apple will try to focus their machine learning efforts on a personalized experience than one that excels on a more generic/general data like Google will.
This theme of personalized machine learning will be extremely difficult due to the more small and limited data sets of data unique to me which will exist. Machine learning needs large data sets of common data and that is completely different than what is needed for personalized machine learning which is massive quantities of random and unique data. That being said, this concept is one I’m watching very closely in 2018 to see how all these companies provided machine learning as a service attempt to personalize the experience to the individual.
