How Deep Learning Will Create Vast Wealth
The data value chain will revolutionize our economies and our lives. Understanding it deeply will help in making great investments, such as $BB and $AMD.
Part 1: Understanding deep learning and predictions.
Part 2: The data value chain.
Note: I´ve moved to SubStack. I´ll still be posting things on Medium occasionally , but now you can find most of my new work here.
Part 1: Understanding deep learning and predictions.
I have previously discussed how advances in information and energy may very well define our history as a civilization, in this post about Spotify. I have been reflecting further on this matter and now I see with clarity that in fact our recently gained ability to produce, store and process data in vast amounts actually represents a leap for us as a species. Our ability to work with data today, may be comparable to the ability we gained a long time ago to work with fire, or maybe electricity.
As I look for new investments and I continue elaborating my mental map of the world, I see that industries across the board are being increasingly digitized. This is fairly obvious and requires no explanation. However, the implications this brings are quite un-obvious to the person who has not spent a lot of time working with data and with novel technologies such as deep learning (neural networks). A first approximation requires briefly reviewing how our reality works.
Last year, I read the book “Intuitive Intelligence” by Malcolm Gladwell. I was shocked to learn that Dr. John Gottman was able to predict divorces with a very high degree of accuracy — most likely above 90%, if I am not mistaken. He would interview couples on tape and ask them to discuss something problematic in their relationship. Later, he would review the tape and assign each second of the film a predominant emotion, out of a number of possible emotions (0 to 10, for instance). He would then check in on the couple every once in a while to check whether they had divorced. Eventually, he was able to train an algorithm that could predict divorce with a far higher accuracy than a board of certified clinical psycologists.
In the above scenario, data was simply gathered and curated on a specific matter. Then, an algorithm was trained with the data and voila — the algorithm could begin to output predictions on it with a high degree of accuracy. Do you see where this is going? There are numerous examples of how our reality is perhaps not as unpredictable as we believe it to be. It´s just that we humans are not very good at making predictions. It turns out that neural networks are very good at making predictions, for the specific reality they have been trained to understand/predict.
Another book that enlightened me on this topic is “Prediction Machines” by Ajay Agrawal et al. The book presents predictions as a commodity. If you think about it, a lot of our economy is developed to make up for the fact the we are terrible at making predictions. The most direct example are waiting rooms. Our economy is full of them. Also, our health systems are built to react to sickness rather than predict it and stop it from materializing. Our inability to predict translates into time and resources wasted. On the other hand, our new found ability to harness data and the power of deep learning has a great potential to make us better off, by commoditizing predictions . We can save time, resources and lives by making better predictions at a reasonable cost. It can create wealth and drive GDP accordingly.
Predictions are anything that a trained neural network (or comparable algorithm) outputs. For example, suppose you want to train a neural network to output whether an image is actually a cat or not. When the network outputs that the image you feed it is that of a cat, it is predicting that the image is that of a cat. If you train a neural network to drive a car and it decides to turn left given a certain stimuli, it is predicting that it should turn left.
What does a world of commoditized predictions look like? Objects that are present today in your day to day live will become smart and will yield data. This data will be transported safely, with the adequate privacy measures, to servers on the cloud. In turn, the data will be fed into AI algorithms, which will in turn output useful predictions, that will save us money and time. Cars will be constantly producing data about road conditions, the state of their different components (tyres, brakes, engine etc). So will planes, wind turbines and any other machine that is used to make our lives better.
Part 2: The data value chain
How does this come to fruition? Through the data value chain.
Things become smart once they begin running an operating system, like BlackBerry´s QNX RT-OS, specialized for mission critical devices. For an object to be able to transmit the data to the cloud, endpoint management is required, to secure the device. Since organizations will connect many devices, unified endpoint management (UEM) will be required — Blackberry is also very good at this. Through UEM, devices will transmit data to the cloud. The cloud, run by companies like Google and Amazon, will operate hardware units (CPUs and GPUs), from companies like AMD and Nvidia, which will in turn run algorithms created using frameworks such as Pytorch and Tensorflow on the data originally sent by the devices. In turn, these algorithms will output predictions which will then lead to applications.
This data value chain will feed into many industries, as use cases are gradually unlocked. Blackberry IVY is an example of AWS and Blackberry working together to create the data value chain for the auto industry and in turn, create an ecosystem in which 3rd party developers can create apps that add value to both drivers and auto OEMs.