#866 November 29, 2017

The use of artificial intelligence (AI) in automated decision-making is a special interest of mine so I thought I would provide some comments on Van's recent excellent article.

The NY Times published a story in May about a self-directed artificial intelligence program playing the world's reigning champion in Go, perhaps the most complicated board game in existence. The software was special because it did not learn how to play Go from a set of human experts. Instead, it learned to master Go by playing games against itself constrained only by the rules of the game. This is an example of setting aside human constraints in order to explore the solution space in an automated way.

The previous generation of Go software was developed by assembling human generated decision rules by Go game experts and that software had defeated a number of globally ranked human Go players. The newer version of the software taught itself how to win by playing games against itself. Then in a heads up competition against the previous generation of Go software, the new software won 100 games — out of 100 games. Starting from scratch, the new software took only a few years to develop a truly formidable skill that crushed the previous best human+machine effort. It also beat the reigning human Go world champion three times in a three game series in May. That performance clearly illustrates the kind of potential that machine learning software holds.

There are a couple ways that AI can learn to be effective in trading the markets. The speed and power of modern computing make daily updating for neural networks possible as well as re-optimizing them based on the most current pricing patterns. This iterated approach adapts and uses the power of AI to find deep patterns in large data sets that would be hidden from a human centric view. The neural network simply explores the potential solution space with brute force computing and makes micro improvements in a way that is responsive to the deep order that can be found in data. This approach is especially effective for complex systems with many variables, like the market.

Another AI technique is to use a model of biologically inspired genetic algorithms. This technique also uses brute force computing to make random mutations among the many combinations of trading rules and then computes the results of these rule sets competing against each other dynamically. The mutation and randomization factor ensures that the solution space is not constrained by preconceived human notions of what should work. This is a very powerful way to find surprising patterns hidden deeply in data.

A third way of applying artificial intelligence is in the area of machine learning, in which the software can develop and then test learning algorithms that again are not constrained by human intuition or prompting.

All of these approaches set aside the preconditions of human sense making and use the power of computing to rapidly and efficiently explore the solution space. These approaches do not remove the human factors from the equation completely, however, because the data sets that they are engaging with still reflect the impact of human psychology in the form of the other agents and actors that are competing in that space. So this means that the deep patterns that they are learning to exploit are still conditioned by human psychology so in that sense, human variables are still found inside the problem space.

This notion of self-programming and adaptive learning software that can bootstrap itself to high performance without human intervention or control is precisely the concern that Elon Musk and Stephen Hawking have expressed about the dangers of AI for the human race. We are rapidly approaching a moment in the future at which a yielding of control to automated intelligence becomes a significant issue.

The role of human decision-making in market trading is still viable when it comes to selecting and blending risk parameters and strategy guidance in the application of these kinds of tools. It is still possible to use the power of human intuition and sense making to find adaptive ways of responding to changing markets. It's clear, however, that traditional rules and heuristics are vulnerable to exploitation by AI which can decode and exploit static rule sets.

In that September 13, 2017 article here in the Tharp’s Thoughts newsletter, I gave you two items that I keep a close eye on as my “canaries in a coal mine”. My early warning system that market momentum could be changing consists of market breadth and global market strength.

In another article three weeks ago, I wrote that market breadth was diverging from price. That was the first cautionary signal from that long-term indicator that we had seen in well over a year. The cumulative advance-decline line was drifting lower vs. a market which was making new all-time highs. Here was a yellow flag. I told you then that this could resolve itself in one of two ways:

Special Note: I received so many wonderful thoughts and comments about last week’s article “Why Gratitude Matters for Traders and Investors.” If you didn’t get a chance to read those ideas, I believe it would be well worth your time.

See full workshop schedule, discounts, pricing, logistics, etc.

The Super Trader Summit is in December.

ALL YOU HAVE TO DO IS CLICK HERE!

For a list of nearby hotels for our Cary, North Carolina locations, click here.

Click here for more information about RDU.

Email us: [email protected]