Machine Learning: Attack of the Clones

Robots will soon look at a thousand charts and news stories simultaneously with your perspective.

HASBRO TOYS

Hasbro Inc. “Star Wars: The Clone Wars” toy action figures are displayed at the Hasbro New York Toy Fair 2008 in New York, U.S., on Friday, Feb. 15, 2008. The U.S. Toy Industry Association, whose members include Mattel Inc. and Hasbro Inc., said its board unanimously approved a plan for a new testing system following the recall of tens of millions of Chinese-made toys last year. Photographer: Jin Lee/Bloomberg News

JIN LEE/BLOOMBERG NEWS

It’s toward the end of February, and another Valentine’s Day has come and gone. Although romantic and special for some, others (including, admittedly, many in finance) entered this week wondering what would happen if a computer could hold up their end of a conversation over a dinner date. Luckily, technology has just the answer for them.

Industry-leading news web site TechCrunch has announced that “love is becoming more scalable,” thanks to the ever-increasing connectivity provided by social media. We can now text, e-mail, Facebook message, upvote, like and otherwise reach out to hundreds of loved ones, friends or bare acquaintances to let them know we care — which sounds, frankly, like an insane amount of work. Good thing that former Google developer Ashish Bhatia has invented a system that might just allow us to respond to all the heartfelt tweets and Facebook messages without even reading them — and without anyone knowing it isn’t really us.

As described in his patent application for Google, Bhatia’s much-touted idea is that we can use machine learning to streamline — and, yes, automate — our online social lives.

It was Alan Turing’s 1950 paper, “Computing Machinery and Intelligence,” that introduced to the world the namesake Turing Test and asked if we could imagine a computer that could win “the imitation game.” Behind the blaze of innovations since it was composed, the Turing Test remains the founding text for most of our conversations about artificial intelligence, for any purpose. Can we build a machine or a program that behaves as if it were in fact a human brain — or acts so convincingly that we accept it as a version of ourselves? Though Spike Jonze’s recent film, Her, imagines a form of artificial intelligence so rich and full as to persuade human beings to form full romantic relationships with their computer operating systems, we haven’t yet found a good replacement for a handwritten letter expressing complex human emotion, much less for, say, a mother’s love. But when it comes to the financial services sector, near-future applications for machine learning and artificial intelligence may prove invaluable.

So what does this new kind of automated-avatar learning have to do with finance? As with any other profession, the workflow any investor can accomplish is generally fixed by the limits of his or her experience, skill and intelligence. Discount what are for now unforeseeable events, chance and so forth, add a widely varying array of access to information (market indicators, economic data, however granular or flattened, however sophisticated or simple), and you have a basic outline of the capabilities of any individual financial analyst — one analyst, one brain, surrounded by the tools of the trade and possessed of differing abilities to use those tools. But imagine if that person were not just one person, if the hard limit that is a single brain was in fact a much softer limit than we thought.

Bots equipped with machine learning like Bhatia’s are designed to act like people, that is, like the human you. When used for social media marketing or to assist you with sharing massively scalable Internet love, they behave in accordance with your preferences, both expressed explicitly by you and learned over time from patterns implicit in your behavior. As described by the BBC, they “can mimic your usual responses to updates and messages from friends and relations to help cope with the daily data deluge.”

Sponsored

Deriving real-time insight from “the daily data deluge,” scraping data from the web as well as compiling and analyzing more conventional sources of financial information, could easily become an automated process. These days it’s not difficult to imagine that bots might learn to make rapid, granular decisions about which stocks to buy, in keeping with any given investment style, that is, in keeping with your investment style. Distinct from high frequency trading algorithms that follow general and blunt rules (for example, buy when price is less than X and volume is greater than Y), bots can be trained through machine learning to identify the signals you would identify, independent of any overly stringent rulebound conditions. The behavior of an asset might exhibit certain traits that would be impossible to reduce to a logical set of conditions, and nevertheless you find it appealing. Bots can learn, simulate, replicate and amplify the reach of that idiosyncratic perspective. You can only look at one chart or read one news story at a time to draw your personality-driven inferences; bots can look at millions simultaneously, and eventually will be able to do so with your perspective.

Bots would not make investment decisions independently but would pass them on to traders, as a set of near perfectly engineered recommendations derived by functional digital replicas, or avatars, of your own brain.

Though the idea of a 12-brained trader is certainly exciting, it also seems squarely within the realm of possibility. Hedge funds like New York–based Rebellion Research have already begun to develop machine-learning algorithms that recommend stocks the same way e-media giants Amazon and Netflix recommend books and movies. Rebellion’s founder points out that the technology might be useful for all sorts of activity, from systematic buys to high frequency trading. Most online retailers have some version of an intelligent recommendation system, even if it is merely to display similar products based on keywords related to color, size and style; to refine these systems and to massively augment them with machine learning are more than imaginable, more than possible, simply doable. (There has of course been pushback; the Wharton School of the University of Pennsylvania recently published a basic breakdown of the reasons machine learning won’t do for finance what we want it to.)

Machine learning might mean multiplying, near infinitely, the brain power of the best analysts and placing it back at their own disposal. A not-so-extreme version of this process might work as follows: A dozen artificially intelligent versions of any analyst — say, you — take in information and parse it, making sense of it exactly as the human you would. That information is sorted and then passed to you, the human trader, to act upon as you like. The result is, essentially, the creation of digital clones of a good trader, sent out like Bhatia’s bots into the world of fast-paced information. And unlike the bots you might one day employ to fully answer your e-mails, Facebook messages and even romantic texts, financial-analyst-amplifying bots won’t require a lot of explaining to your date when their use is discovered.

Related