The thing about predicting the future is that it’s really, really hard. The drive to know what will come has been a part of human ambition the first drawing on the back of a turtle shell. Throughout history, we have used methods ranging from decapitating fowls to rolling dice on binomial trees, and the results, just as much, have varied. Predicting the future has always been an inscrutable craft – something reserved for those elite few with knowledge and power.
Whatever algorithms the oracles of Delphi used to direct the fate of our ancient forefathers are now lost. (Alas, if only the ancient Greeks believed in the concept of open source and GNU licensing!) Even so, with no insight to their methodology, both the great and common man certainly made use of oracular services, whether through faith, desperation, or reason – making this particular craft a cornerstone of human history and development. Today, this field is evolving yet again. Armed with new tools, ideas, and hardware, there is a revolution in precognition. Perhaps for the first time in human history, we’re making clairvoyance into a science. Predictive modelling via neural networks, trend analytics via deep learning, and AI-driven data intelligence are some of the fastest developing technologies.
Everything has a start. Before we can go modelling the state of an entire civilization millions of years into the future, as Isaac Asimov envisioned in his Foundation series, we first must set more practical goals; baby steps, as they say. My goal (and particular interest) is in predicting foreign exchange rates. These minute upticks and downticks of currency rates are in fact driven by titanic clashes between empires in statecraft, trade, and war. To predict what will happen to the forex rate of a particular currency seconds, perhaps minutes, into the future, by drawing upon data and trends in the world around us is both a worthy and a profitable goal.
The problem is complex. The world is a monumental machine – and the number of moving parts will only grow as time goes on. What happens in Yemen may affect politics of the Americas, and trade spats in Asia will ripple across both hemispheres. Yet these are the exact factors that influence currency rates. To identify all the factors would be a nearly impossible task (until at least, in a few decades, when quantum computing and Lovecraftian data structures become the norm). To add to the complexity, these factors are not static. Every piece of news brings in new factors. Every change in political power introduces new entities and removes old ones. Every conflict, every threat, and every policy change demand adjustments to our model. We are immensely fortunate, then, to have new tools, invented and developed in recent years, which help us address them.
We can build machines that can read through hundreds of news articles, tweets, blog posts, and discussions in seconds via natural language processing. These algorithms can extract key topics, subjects, and even sentiments. From this, we can build a corpus of key entities, ideas, and events. Although natural language processing algorithms will never outperform a human when reading the same piece of text, the sheer speed and efficiency advantage makes machine reading much more able to cover significantly wider ground, giving us unparalleled access to a significant percentage of published media, something that’s simply not possible a decade ago.
We can design machine learning algorithms that map complex relations. A piece of news affecting Mario Draghi, would, then, also have ripple effects to the European central bank, the European Union, and the Euro currency as a whole. With the aid of modern computing hardware, combined with the newest in machine learning, we can create artificial intelligence algorithms to design methods that can extrapolate these relations at speeds and efficiencies unimagined five years ago. The same algorithms can then take existing relations, identify trends, and project the future. Combined, utilizing methods and tools in the rapidly developing fields of artificial intelligence and natural language processing, we can not only model an incredibly complex world, but take a very good guess as to what’s going to happen next.
Surely, we’re far from creating cyber-Nostradamus, and our predictions fall far short of one hundred percent accuracy. Under laboratory conditions, we’ve only gotten it down to generate predictions for 30 seconds to one-minute time windows for FX market data – a far cry from the seers of old advising Caesar on his battles though Cisalpine Gaul. But that’s the best part about science – start small, establish a domain, and expand. There is certainly a drive to get it right – if it works in one case, then we can duplicate it for other cases, for other instruments, and in other markets. Often, applications such as ours become the gateway for something greater. We seek to build a model with a mold that can be generalized and mass produced, giving orderly structure to chaotic complexity. The world is difficult to model at once, yes, but when we zoom down to the smallest building block, oftentimes, like a fractal, patterns emerge.
The development of artificial intelligence, meshed with natural language processing and purpose-specific hardware, is quickly becoming a focal point in the field of computer science. We, as a people, are finding more and more capabilities in our constructs, and trusting them enough to outsource some of our decision making. Perhaps one day, our machines will be able to see the future better than we, and not only make accurate predictions on time scales much grander than a few minutes, in fields far less restrictive than the market data of a particular financial instrument, but they may also be entrusted with the decisions that lead us to a better tomorrow.
This is my prediction of the future – that one day, our constructs will be able to glimpse at tomorrow, and through that lifted curtain, tell us, with certainty, that the best is yet to come.