I have talked briefly before about the threat that is posed by decoupling of the job market from financial indicators. The fact that automation has been making it easier and easier to produce more with less human labor has driven the long term trend favoring capital over labor, and is a key element in the wealth inequality we see today.

In recent years a new spin on automation has also emerged in the financial industry, and I think it may represent an existential threat to our way of life. Automated trading on financial market already accounts for half of the trading volume that happens today. At its best, automated trading could offer a means to deliver capital efficiently where it is needed, but at its worst algorithmic traders could threaten our wealth, our economy, and our freedom. 

Let it be resolved that algorithmic trading represents an immediate existential threat to the the modern world. 

The most prominent form of automated or algorithmic trading today is known as high-frequency trading (HFT). For a good introduction, this documentary provides some insight into the way that the shadowy world of HFT operates. Essentially, this kind of trading relies on buying and selling assets just moments ahead of slight changes in price up or down. By acting on informational disparity at exceptionally small time intervals,   these these HFT firms can make a tiny bit of profit on each trade. In the world of HFT, those with the fastest information get the profit, and the profitability of these firms is driving a scramble for faster data transmission. 

It is thought that HFT was responsible for the flash crash of 2:45, so called because the market lost almost 10% of it’s value in mere minutes, only to recover the value in minutes more. This kind of rapid and chaotic occurrence may be a property of markets with so many feed-backs ready to act rapidly on slight changes in the market. This speaks to the danger posed by a market dominated by algorithmic trading and which is disconnected from human rationality.

HFT represents the obvious extreme of a system where the cost of transactions has approached 0. Means to slightly slow the market flow or increase the transactional costs have been effective at slowing the growth in this kind of trading (having actually fallen since a high in 2009, at as much as 73% of trading volume). But if we go beyond simple HFT and look deeper at algorithmic trading, the real power and danger of algorithmic trading become apparent.

Algorithms are mining vast stores of data on everything from weather conditions and crop yields to political changes and historical stock prices. These algorithms aim to find correlations which predict the price for anything which can be bought or sold on an electronic market. By being the first to identify correlations which have been previously unrealized, a lot of money can potentially be made.

For example, an algorithm might notice that a drop in the stock price of a ketchup maker always follows in the minutes after an increase in the price of tomatoes. Thus the algorithm can then act to profit by acting very quickly on small changes in the price of tomatoes. These algorithms have no need to understand what ketchup is or why people like it in order to profit of the correlation they establish.

Naturally, there is a huge profit incentive to make these kinds of trading machines more accurate by empowering them with more intelligence. Algorithms which can monitor twitter and other news sources and look for news about various investments, are very useful. These bots are certainly not nearly as good as humans at understanding natural language, but what they lack in understanding they make up for in speed. These algorithms can potentially act on a news item in the order of milliseconds after news is released, capitalizing on the time that it will take humans to read and understand a tweet.

So far what I have described may not seem too scary. Algorithms looking for correlations in market performance sounds just like what human traders do. Figuring out where capital can be best allocated for most profit is what the market is supposed to do, and these algorithms are only helping us do it faster. Maybe this means is that some traders are out of a job, and what should you care about that?

But this is where it gets a bit scary. The institution of algorithmic trading as the main force driving modern markets means that the decisions of algorithms are increasingly the basis of market price. As these algorithms start to go up against eachother, they are also being used to discover ways to manipulate the market itself. Scarier still, it is unclear how connected the gaming between algorithmic traders is to the fundamentals of economic function (like how many people can afford bread tomorrow).

This example from a Sean Gourley’s 2012 TEDx talk really drove the danger of these algorithms home for me. This algorithm is rapidly selling and buying natural gas futures in an effort to find an algorithmic market breaking point. Once the price hits a certain threshold, other trading algorithms then act and the price quickly drops almost 10%. Perhaps by accident, perhaps by intention, this algorithm has found a means to manipulate the price of natural gas in the real world. This kind of gamesmanship between algorithms could realize huge profits for whoever controls the most advanced algorithms.

The advent of algorithmic trading extends the game that has always existed in markets, but now the speed is faster, the stakes are higher and we can’t be sure who is in control. 

The manipulation abilities of trading algorithms may already (and if not, soon will) extend beyond this kind of inter-algorithmic effects. Given that trading algorithms can act on human informational sources, such as Twitter, as news is released, it is not outlandish to imagine that these algorithms could also be producing information in an effort to manipulate the market. Given that algorithms are becoming better at turning basic information into natural language, it seems possible that an algorithm could be designed to Tweet out false information about a company to try to depress the stock price.

If we take the ketchup manufacturer again and we imagine they are in a precarious position due to a new bill to remove subsidies for tomato growing. Imagine a bunch of tweet/comment/news bots aimed at pushing the public dialogue to make it seem that the subsidies are going to be removed. If massively parallelized, this kind of attack on public sentiment could have a significant effect on the ketchup manufacturer and provide an opportunity for major profits. I think it’s likely this kind of algorithmic sentiment manipulation is already happening on some level.

Even this kind of sentiment manipulation is only a drop in the bucket compared to what may become possible in the near future. The astounding profits which can be made in this kind of algorithmic trading is driving huge investment in artificial intelligence. In the near future, algorithmic traders will be capable of much more complex manipulations to try to move market prices.

Rather than taking creating the illusion of a sentiment shift about tomato subsidies, an algorithm could instead attempt to influence those specific individuals who are going to be making decisions about tomato subsidies. Perhaps by identifying those congressmen who are on the fence about subsidies, a targeted campaign to manipulate the opinions of those in said congressman’s district could have a real effect on the outcome for ketchup manufacturers. This may seem a bit ridiculous, but even a tiny effect on the perceptions and opinions of one individual can make a big difference if spread across a wide enough group.

Like all of the other elements I discuss here, political manipulation aimed at maintaining market position is absolutely not something new. These kinds of practices are a well established part of our world, whether we like it or not. What is new, is that just as computers have always done, algorithms make it possible to scale these kinds of manipulations to make them so much wider much faster, that we ultimately can’t be sure how much of an impact they will have.

So where is all of this headed? Movies like the Terminator made us imagine that killer military robots with super strength and bad-ass weapons could take over our world. Maybe what we should be afraid of isn’t the army of military drones, but an army of Gordon Gekko-bots capable of manipulating every aspect of our legal and political systems in an aim to maximize market profits.

The fundamental problem remains the same as it always was, money doesn’t care about the betterment of human life, if we fail to firmly attach our own betterment to the betterment of the algorithmic markets and the automated economy then too many of us may end up left behind.