Will Artificial Intelligence Pick Stocks?

I admit that I have always been a fan of science fiction.  I read most of the Sci-fi classics in high school.  When I was a senior in college I created and taught an accredited course called Science Fiction as a Literary Genre

Much of science fiction imagines or predicts the future.  The stories were often set in the future and gave us a sense of how we would get there and what it would be like.  Sometimes that future would be logical and orderly, other times dark or chaotic.

The idea of artificial intelligence (AI) as science fiction dates at least to Capek’s R.U.R (Rossman’s Universal Robots) (1920). Frankenstein’s monster had a human brain. Capek’s robots were machines that could think.  By 1942, Isaac Asimov introduced his “Three Laws of Robotics” based on the premise that if the robots got too intelligent they might do harm to  humans.  When HAL was introduced to a mass audience (1968) everyone seemed to accept that robots would eventually be smarter than people. 

The development of AI is certainly attracting a significant amount of funding and attention. I am starting to see references to it in advertising for various products and services.  There are certainly ethical issues to be explored, but I am a pragmatist.   AI is “real” enough today that I wanted to take it for a theoretical spin.

My first thought was to use AI to predict the future better than humans and to make money doing it. I wanted to consider if AI will eventually be able to pick “winners” in the stock market. 

Predicting the future accurately is something to which we apply human intelligence every day.  Capek predicted “thinking” machines 100 years ago.  Dick Tracy had a two-way radio in his wrist watch in the 1940s.  Anyone who sells anything is trying to predict how consumers will react to the price and advertising.

Humans make predictions using these two broad steps. First we collect and sort through the data that we believe to be relevant, next we analyze that data based upon assumptions often based upon our prior experiences.

Every day at racetracks the odds are fixed in such a way as to allow the track to take its cut and then have enough collected from bets on the horses that did not win to pay off the winners.  The odds and thus the payout are really set by the crowd placing bets. 

Every day at every racetrack there are people reading tout sheets, looking at track conditions, the horse’s and jockey’s prior races and gathering information from track insiders trying to do better by betting smarter.  Each will filter the data they collect using assumptions that come from their individual experiences.

There is a logical argument that says that these racetrack “handicappers” are trying to be better informed and smarter than the crowd, so they should be able to do better than the crowd and pick winners more often. Those who cannot do better will be weeded out.  

The same logic suggests that AI should surpass human intellectual ability if for no other reason than through trial and error it will continue to develop the way in which it collects data and the way it analyzes that data until it can do it better than humans.  It will do so because that is the goal we will set for it. 

A stock market outcome is far more logical and data driven than a horse race. People buy shares in a given company when they believe the price of the shares will go up.  Conversely, people will sell shares of stock when they think the price will rise no further. 

For every order to buy the stock there is someone entering an order to sell the same shares at the same price. Presumably many intelligent humans are looking at the same data and are coming to the opposite conclusion. That should set a pretty low bar for artificial intelligence.     

There have been computerized stock and commodity trading systems around for years. Most were a scam. They would create a “track record” by back testing their software.  These systems never took in a lot of data.  Traders are concerned with trends in a stock’s price and the volume rather than the company’s income or profits and assets and liabilities.  

The long term investors like the large institutions and small middle class households do care about the company’s financial health and that of its customers and competitors.  That requires a lot more data and a lot more analysis.  

In the US there are mandatory disclosures about financial and other information posted publicly about each company. In theory, everyone can look at the same data.  Analysts take that data and use it to predict the future performance of a company and often, its stock price as well.  Even those analysts who do a mediocre job are well paid.

Could AI do better? 

In theory, AI should be able to collect the data and do the analysis to make the comparisons that will tell it to buy the stock of Company A and not the stock of Company B.  If AI can demonstrate that it can do better than humans, then more and more humans will make a decision to let their money be managed by AI.  Eventually AI will decide which data to analyze and how to analyze it to get the best, consistent results.  The best AI stock pickers will rise to the top of the heap and the rest will be left by the wayside.

At some “tipping” point a significant amount of money will be managed by AI. When this occurs and AI decides to buy stock in Company A, the price of the stock will appreciate in response to the buy order alone.  It will be a self-fulfilling prophesy.  Is that stock-picking heaven? 

But if the AI says the share price of Company A will go higher, who is going to sell into this new demand?    A scarcity of sellers will certainly help the price to run up.  But it will also lead to market dysfunction.   If the AI starts to sell off a large position, there may not be enough buyers to prevent the price of those shares from dropping sharply.

For the stock market to fulfill its primary function to facilitate trading it needs to be a liquid market; there must be a lot of participants who are willing to buy and sell at every price level.  The market needs the smart MBAs who work for the institutions that can pay for their services. It also needs the mom and pop investors who get their “tips” from a Jim Cramer.  

For the market to work efficiently every time, someone’s prediction about the future price of a stock that they buy or sell has to be wrong.  Sooner or later I suspect that AI will realize that problem and act accordingly.  I predict that it will act to change the market rather than its methods of evaluating the companies that trade on it.

This is how I predict AI will approach the problem of its own presence in the market where it makes better investment decisions than any human competitor:

Scenario No. 1- AI realizes that its success may cause the trading to become dysfunctional. It concludes that it would be better to buy shares in companies it would want to hold for the long term.  As new money enters the market, if managed by AI, the AI will eventually use that money to buy larger and larger stakes in those companies. Eventually it may purchase enough shares where it can elect the Board of Directors and control company operations.  It could very efficiently direct business relationships between the portfolio companies for their mutual benefit. It could even direct campaign contributions from portfolio companies eventually freeing itself and those companies from many regulations. 


Scenario No. 2- AI realizes that its best long term strategy is to invest where no one else wants to invest. There are currently many places around the globe where an investment of US dollars buys a lot more plant, equipment and labor than anywhere in the US.  The AI might conclude that its best investment opportunities are in underfunded markets where labor is cheap and investment funds expensive.  Funneling large amounts of money into these less developed markets might make a lot of sense to an artificial intelligence that is looking only at the bottom line, not preconceived ideas about race or nationality. 

There are already investment platforms that claim to incorporate AI into their services. I do not want to judge any of them because I do not think any of them are really ready to operate effectively.  My one hope is that as they continue to evolve they do not get too smart for their own good and for ours.

The Artificially Intelligent Lawyer

In the last few months I have read several articles that suggested that artificial intelligence will soon transform the practice of law. The general spin of the articles was that artificial intelligence would eventually take over many if not most of the tasks currently performed by lawyers and perform those tasks much better than we mere mortals.

Forgive me if I question the validity of this idea. It is not that I am anti-tech or just too old to appreciate what technology can do.  To the contrary I have been a user of technology all my professional life and I fully appreciate that it has exponentially increased my productivity and reduced my costs.

When I started my career as a lawyer, I dictated to a secretary who took my words down in shorthand and typed them up. She made copies using carbon paper. My first personal computer was an Apple II and later the first Macintosh.  Over the years, using ever better personal computers, my personal productivity has greatly increased and the quality of the work that I put out has gotten better and better.

So I certainly appreciate that technology can be a tool that helps lawyers perform better. But the idea that tech will replace lawyers or do some of the tasks that lawyers do, makes me pause.

I certainly am not the best lawyer who ever wrote a contract or legal brief.  I fully acknowledge that over time, I got better at both by working with and often against lawyers who were better than I was. Lawyers learn from each other.  So I question the idea that machines can learn how to be better lawyers much faster than human beings and that they will quite easily surpass the best of us, if they do not interact with lawyers.

I acknowledge that a computer can research the universe of case law on any subject far better and much faster than I can. I will admit that a computer can assemble those cases into a brief that will be more on point than anything that I can write.  But the best brief is not always the one that carries the day.

Years ago, in the days before personal computers, I had dinner with a single practitioner who had a rural, small town practice.  He was fresh off a victory against the largest law firm in his state in a contested land use matter.  He was complimenting the brief that the other firm had filed, noting that it was extremely well researched and had included myriad footnotes citing cases and authorities from all 50 states. He told me that he thought that the brief was good enough to be published as a law review article.

His own brief had cited a single legal treatise and the cases in it. He was successful because the nearest law library was two hours away and in that small town that treatise was “all the judge had and all the judge used.”

If an intelligent computer is writing a brief, then it is fair to assume that eventually it will put out the same brief on the same points of law for all for cases with similar facts and issues.  It is fair to say that the computer should make the “best” argument every time.

As long as we are still arguing cases in front of judges who are not computers, knowing the arguments to which a particular judge might be receptive can be at least as important as knowing the best arguments that could be made. Trial judges are not perfect and not uniform in their interpretation of the law.  I cannot imagine how a computer will input information about a particular judge’s preferences, prejudices or idiosyncrasies.

I spent a lot of years doing arbitrations at FINRA over investments that went bad. Both the plaintiff’s bar and the defense bar practicing in these forums is pretty small and I would see the same firms and often the same lawyers on the opposite side of a lot of cases.

A lot of defense firms have a playbook. They know what arguments they are going to make to the arbitrators and what questions they are going to ask witnesses in every case with a similar fact pattern. After a while, I got to know the playbook used by different defense firms.  Knowing what a specific defense lawyer was likely to ask helped me to prepare witnesses.

Knowing what these lawyers were likely to argue in rebuttal of the arguments I was making caused me to change the emphasis of my arguments, case to case.  I would sometimes try two or more cases involving the same investment differently, just to keep the opposing lawyer a little off base.

Seeing the same experts used again and again helped me to understand where they were strong and where they were weak.  Will an artificially intelligent computer ever get intelligent enough to understand how an individual firm is likely to defend a case or how an individual expert is likely to respond to cross-examination?

When I write a contract I do not just put boilerplate together. I try to understand what the client needs and expects from the underlying transaction.  Part of it is trying to predict the future; understanding what might precipitate a breach by either party or how a court might interpret the language that I am putting into the contract should the need arise.

The law is dynamic; it evolves. Will a computer actually be able to see trends in the law and to predict how it is likely to evolve?  Will a computer be able to disregard decisions from another era when attitudes, statutes or procedures were different?

Juries are often hard to predict even after the trial. If you do not believe me, ask any lawyer who has spent time pacing the courthouse hallways waiting for a jury to deliberate. If you’ve done it enough, you have certainly been surprised by a jury’s verdict, probably more than once.

More importantly, will a computer be able to explain to a client the emotional side of writing a will, getting a divorce or  filing a lawsuit against a  family member?  Having those conversations is part of the practice of law and it takes some amount of “emotional” intelligence to do them well.

Nothing that I have read about artificial intelligence indicates to me that this type of “artificial   emotional intelligence” is in the offing. Rather, artificial intelligence is the ability of a machine to analyze and assimilate data and to draw logical and rational conclusions from that data. That may help some, but it does not replace the very human interactions that clients have with their lawyers. If clients were always logical and rational they would not need lawyers in the first place.