Tim Morgan inspects the concept of Smart Capital in his tenth blog post for our Emerging Fellows program. The views expressed are those of the author and not necessarily those of the APF or its other members.
We are starting to embed automated decision-making into capital itself. We routinely embed automatic control systems into our processing plants and factories ensuring that optimal use is made of those capital investments. Advertising and sales are increasingly given over to algorithmic management. No industry seems to be untouched by automation. We are infusing our intelligence into our capital systems. So how smart can our capital get?
Once upon a time a computer was a job description, not a machine. Human computers did the hard work of accurately calculating everything from astronomical phenomena to tracking weather patterns. That changed with the advent of stored instruction computing machines. Programmed algorithms could be systematically created from a combination of well-defined repeatable steps incorporating not only mathematical operations but conditional (if/then/else) decision-making logic as well.
We have been developing this computational capability for decades. We still are limited by the need to design and transcribe programs most of the time. The logic is still simplistic and rigid compared to human reasoning. But that design limit is quickly giving way to complex machine learning algorithms. Artificial Intelligence has been a field of study since the beginning of the digital computing era. Now the early promises of decades past are rapidly being realized.
A.I. researchers are harnessing our exponentially increasing torrent of data to train machine learning algorithms. This has resulted in A.I. techniques like Generative Adversarial Networks (GAN) which use competing Generator and Discriminator neural networks to solve problems based on older human-curated examples. GANs quickly learn tasks like creating human-like art, designing 3D objects, and accurately identifying tumors in X-rays.
Other advanced A.I. techniques are moving beyond the need for human training or big data sets. Google’s AlphaGo A.I. beat world Go champion Lee Sedol by 100 games to 0 in 2016. AlphaGo’s neural network heuristics were initially trained using a database of 30 million moves from 160,000 masters-level games. Yet in 2017 with no access to that database and just three days of self-play AlphaGo Zero beat AlphaGo by the same 100 games to 0 that AlphaGo beat Sedol. Go masters worldwide have begun eagerly studying AlphaGo Zero’s unusual moves to inject new strategies into their sport.
Computing advances will not stop with digital computers and machine learning. Researchers around the world are rapidly developing Quantum Computers to take computing capabilities to a whole new level. A leaked paper recently revealed that Google has demonstrated the theorized principle of “Quantum Supremacy”, or the ability of quantum computers to quickly solve problems that conventional computers cannot. Their quantum computer solved a problem in about 3 minutes that the world’s most powerful supercomputer could not perform in 10,000 years.
The cognification of capital via computing will not stop. It will accelerate. Capital will incorporate computer’s gains in self-training and abilities to solve ever harder problems. Capital will acquire more and more ability to self-manage with less and less need for human decision making. The ultimate endpoint may be that it no longer needs our direction. If that happens, capital will go from being owned to autonomous. If it does, we will need to pay close attention to what it wants.