Cryptocurrency is a fundamentally new type of investment, one that has investors and economists both excited and nervous about the future. Sometimes referred to as digital currency, cryptocurrencies like Bitcoin utilize blockchain technology (or other decentralized means of exchange and recordkeeping) to log transactions in tamper-proof ledgers.
Though the concept has been around for years, interest in cryptocurrencies has spiked with the dramatic rise in Bitcoin prices (and, a short while later, its 50 percent crash). Now software engineers and financial advisors are wondering: Is it possible to use machine learning and AI to model the potential growth of cryptocurrency and better predict its future fluctuations?
There are some key opportunities and limitations driving the future here.
Opportunities for AI prediction
Let’s start by looking at the opportunities for development and growth.
AI prediction and management in other areas
There are many overarching investment strategies to consider, and so far, human engineers have been able to semi-master trading algorithms that can fulfill those goals. Robo-advisory firms like Wealthfront and Betterment are already managing billions of dollars of assets in stocks, bonds, and index funds. If we can manage simple assets and currencies already, it stands to reason a cryptocurrency trading algorithm isn’t too far beyond our capabilities.
Observation of patterns from other novel commodities
Cryptocurrencies aren’t the only new, volatile investment asset available. Though on the surface, cryptocurrencies are designed to function like any other global currency, their volatility suggests a pattern somewhat closer to those of commodities. Studying these patterns and learning from them could be a shortcut to faster success.
Stabilization and future intrigue
Cryptocurrencies are currently volatile, thanks to a combination of uncertainty and consumer excitement. However, it’s unlikely that this volatility will last forever. Assuming cryptocurrencies remain active, it’s likely that they’ll eventually stabilize and become more predictable, making them easier for trading algorithms to handle.
The fact that so many consumers are interested in cryptocurrencies (especially Bitcoin) is driving more engineers to study this area and offer products that make trading easier or more profitable. At least one platform, VantagePoint, already incorporates cryptocurrency predictions into its offerings, correlating Bitcoin prices to 30 other markets to project price fluctuations.
There are also some limitations we should bear in mind.
Lack of data
Bitcoin is only a few years old, and most of its contemporaries have only been born in the past several months. Unlike the stock market, which has nearly 100 years of data to study, we’re in the dark with cryptocurrencies; this is a fundamentally new type of investment, so there isn’t enough raw data to make suitable long-term predictions.
Currently, even our best trading algorithms are still the product of a human mind and are therefore subject to all the biases and limitations of the average human. If we don’t understand how the market works, how can we possibly design a trading algorithm or predictive AI technology that can outperform us?
Processing and tech limits
The more sophisticated a machine learning algorithm is, the more hardware and processing power we need to run it. As an example, one of DeepMind’s latest projects, AlphaGo, needed 1,202 CPUs and 176 GPUs, more than 25 times as many as the single-computer version of the algorithm. For a high-profile company with ample funding, this is an easy obstacle to surmount. But for an amateur entrepreneur or a prospective machine learning engineer, this can be prohibitive.
Also, consider that to develop a truly successful cryptocurrency trading algorithm, a company would need to win the public trust. The company would be responsible for not only proving that the algorithm works better than the average human being, but that cryptocurrency is a worthwhile long-term investment. So far, despite the consumer excitement over the currency, lingering questions about the long-term efficacy and acceptance of the currency prevent it from being a mainstream source of exchange, especially in the international scene.
Is it possible for AI to model the growth and recession trends of cryptocurrencies better than a human? As of now, the answer is no, but it’s certainly possible, given enough time and enough effort put into the project. Until then, the volatility and unpredictability of cryptocurrency will likely remain a staple feature of the investment type.
Larry Alton is a contributing writer at VentureBeat covering artificial intelligence.