Deep learning has been around for decades but it just got interesting. Investors could be at a watershed moment for productivity. There could be far reaching implications for economies and society as a whole. Imagine swathes of low paid labour and professional labour being replaced by machines. How will governments deal with this? A living wage might solve the problem. The results could be quite deflationary in monetary terms.

Known by various names such as machine learning and artificial intelligence (AI). Google (GOOGL) and IBM (IBM) have been deep learning for some time. Arthur Samuel programmed a computer to play checkers better than he could in 1959. How can he teach the machine to play better than he knew how to do? What he did was teach it how to learn. By running through millions of iterations of game tactics it was able to play better than it’s inventor.

Fast forward to 1996 Deep Blue beat Garry Kasparov (the Russian chess genius) at a game of chess under competition conditions. Chess has a huge number of possible moves to enable the player to win. Capturing the imagination of the public at the time because a computer beat a human, and, Kasparov wasn’t happy about it.

It’s no coincidence that the Tony Stark’s computer in Iron Man is called Wilson bearing resembling in name IBM’s latest super computer Watson. IBM’s Deep Blue evolved into Watson who’s feat extraordinaire was to beat beat two of the greatest champions of Jeopardy (the US quiz show) at the quiz. Watson had to play 100 past winners in order to prepare for the match against the quiz playing greats (the learning bit). Watson tackled the double Jeopardy question with ruthless efficiency as this is where the big money is.

IBM calls Deep Learning cognitive technology that is enabling new partnerships between people and computers. The concept has come a long way in the last few years, from the clever computer algorithms that Google use to help you with your search results, or how Amazon is good at suggesting other products you may want to buy, or how Netflix is good at suggesting programming you may enjoy after you’ve used the service.

As Editor of Let’sCompareBets we have written about how we chose investments, specifically for a self invested personal pension. Once such approach is by spotting big themes, one of which is the advancement of technology. Specifically, the concept of Moore’s Law, which simply put, is that the processing power of integrated circuits doubles about every 18 months. This is down to new materials and techniques used in manufacturing these circuits. Robots are getting more clever at an exponential rate: exciting stuff.

Google DeepMind’s AlphaGo (their supercomputer) has beaten Lee Se-dol one of the worlds best players of the challenging Chinese game of GO. Putting this feat into context, there are billions of possible moves in a game of chess. The number of possible moves in the game of GO, due to the large board size, is 577 digits longer than that of chess. So what you may ask? It has taken AlphaGo a few years to surpass the processing power of the human brain that needed a lifetime to lift it’s owner to the status of Grandmaster.

Developments in AI are enough to keep futurologists topped up with inspiration and material to make exciting predictions about the day machines take over the world. What we’re interested in is how AI and Deep Learning are being applied to the world today.

Both Deep Blue, Watson, and AlphaGo are all types of narrow AI. They take on one task and do it better than Humans. The application of narrow AI is no making inroads into society now. Robotics has long taken away low skilled, particularly manual jobs away from humans, in the car manufacturing process for instance. Narrow AI is now disrupting the medical industry and will help solve the problem of preventable deaths or injury caused by errors in diagnosis and poor treatments decisions, as well as the cost of training a doctor. The estimated costs per doctor for training, post graduate work is £1M and even more taking into account continuing professional development. As robots have helped increase the productivity of the car manufacturing process, AI, will now augment the productivity of medical diagnosis. For instance, a diagnostics company based in San Francisco called Enlitic ran tests to compare the ability of one of their diagnostic programs to correctly diagnose early stage lung cancer against a panel of leading physicians. The program out performed the human competition, most notably with the doctors missing around 7% of cases, potentially putting lives at risk.

Watson graduates as a doctor

Dr Watson can scan through massive databases of of textbooks and latest medical journals to recommend the most effective therapies based on the patient’s symptoms. Freeing up the doctors time allowing them to focus conducting therapies. Most recent updates include a version that comes up with its own ideas for treatments based on the latest treatments and drugs, where few doctors would know about them.

Dr Watson is being used in 13 leading cancer centres in the USA and 5000 hospitals. The company leading the charge is IBM. IBM has hovered up many AI related patents and it’s data analytics division make up almost a quarter of revenue but is growing at 15% per year. Potential markets for AI will only grow over time. AI will soon be used to help the London Underground schedule maintenance and prioritising engineering tasks. Many managers working in Hong Kong’s Metro system have been made redundant due to the role out of AI for helping with the process of maintenance. IBM is attractively valued at the moment.

What industries might be next? If doctors are willing to take a hit, law could be the next industry to get attacked by the machines. Peter Diamandis has gone on record saying that the world might be a better place with a few less lawyers. In a litigious society like the USA the machines may have a fight on their hands with the established order on that front.

X marks the spot that may help sculpt the future of mankind

Google X, now know as X, is the secretive research facility funded by Alphabet the parent company of Google. Some quarter have suggested that Google’s research and development budget is being focused on producing a superbrain. The concept of a google superbrain definitely captured my imagination. X is facilitating the progress of AI from narrow applications to broader applications. The ultimate goal of which will be to enable AI to solve the ‘Turing Test’ which is to convincingly pass itself of as human. That is a massive task. It is one thing to surpass the processing power of a limited part of the human brain (involved in playing the game GO), but, something totally different to match the processing power of the human brain as a whole.

Google makes a lot of money. The company holds significant amounts of cash, and, is pumping a large portion of that into research and development. A certain proportion of which is going into a pipe dream of moonshot projects. Larry Page and Sergey Brin’s punting money is going into it’s moonshots and they expect only a small percentage of these pie in the sky projects to shape the future direction of the parent company. Tony Stark eat your heart out.

What are the top R&D projects being incubated by Alphabet’s superbrain?

    • Google Glass is one punt the company hasn’t landed… yet.
    • AlphaGo was developed by DeepMind which was originally a British company that was bought by Alphabet in 2010. Developing algorithms that can learn for themselves DeepMind will become intrinsic to X’s mission to development some mind blowing technology. Currently DeepMind is working in partnership with Nurses and Doctors in the National Health Service (NHS) to scale their technologies to improve patient care and outcomes
    • Project wing is a fledgling drone delivery project
    • The Self Driving car project. After more than 1 million test miles the project is about to be spun off into a separate company. AI enables a self driving car to perform the many functions enabling it to work and be safe, for example, learning what the difference is between a tree and a family standing at the side of the road. Self driving cars require a lot of sensors to feed data back to the AI at it’s core. One company we have suggested previously is TT electronics (TTG) that supplies sensors to companies like Mercedes who will be testing their self driving lorry convoys on UK roads.
    • Baseline. Is essentially a medical study of baseline health of people. Using the big data crunching ability of the superbrain. This has massive potential for conducting medical trials and test that involve very large samples of people
    • Project Calico and grown into a stand alone company. Now we’re getting really ‘out there’. This one’s into life extension. The elixir of eternal youth. In practice they are using the superbrain to help develop better therapeutic drugs.
    • Boston Dynamics is an offshoot of X labs. The creator of Alpha dog and Atlas has now been up up for sale by Alphabet.

So what will really fire all this AI stuff out into the stratosphere technological advancement?

Developments in hardware which run AI algorithms could allow AI to achieve parity with the processing power of a human brain. How realistic is that? There are 100 billion neurons in the brain. Each neuron can make on average 7,000 connections with other neurons. That is 100 billion to the power of 7,000. A number that is not comprehensible: which makes up huge processing power. Transistors on a computer chip are the closest thing there is to a neuronal connection. Currently 20 billion is the most transistors that can be put on an integrated circuit. There is a huge gulf between these numbers; the superbrain has a long way to go, even before it can match the intelligence of an eight year old child as fair as broad AI goes. Narrow AI is what will be really affecting our lives in the future.

The base material for computer chips is silicone which is a semi conductor. Lots of transistors on a chip produce heat which will eventually stop the progress of Moore’s Law. The next break through to allow the continuation of Moore’s Law and the growth in the like’s of Alphabet’s superbrain is a change to better materials especially those that have super conductive properties like Graphene. A superconductor can operate at room temperatures without getting too hot. At which point, this topic moves into the realms of futurologist type predictions of machines that learn to make better machines. The concept of Skynet in the film Terminator, 1984, would have seemed very improbable to most people at the time. Moonshot indeed.

See the potential of AI in action

Boston dynamics’ Atlas; https://www.youtube.com/watch?v=rVlhMGQgDkY
AI used in drones; https://youtu.be/w2itwFJCgFQ and https://www.youtube.com/watch?v=YQIMGV5vtd4
Algorithms experimenting with self assembly; https://www.youtube.com/watch?v=xK54Bu9HFRw; this one makes me think of Prey, the book by Michael Crichton.