Artificial Intelligence, Machine Learning, and Deep Learning: A Primer
This is the first installment of a three-part series exploring the impact of artificial intelligence (AI) on investment management.
We are witnessing the beginning of the artificial intelligence (AI) era.
The computer program AlphaGo defeated the world’s top player in the complex Chinese board game of Go for the last time in May 2017. The program had run out of human competition. So its developers designed AlphaGo Zero to simply play against itself without the aid of any historical game data. AlphaGo Zero taught itself how to beat all versions of AlphaGo in 40 days.
People have been playing Go for millennia. And yet all the human wisdom accrued during those countless hours of competition across the continents and throughout history turned out to be no rival to an AI program with 40 days to itself.
And AI’s footprint is not limited to board games. Its imprint can be seen in countless industries and professions, from finance to medicine to accounting. JPMorgan’s COIN program performed 360,000 hours of finance-related work in a few seconds. An AI program at the University of Nottingham can now predict strokes and heart attacks more accurately than doctors. Accounting firms EY and PwC are testing drones that use AI in their audit work.
The threat AI poses to white-collar jobs is obvious. But before embarking on an elaborate discussion about whether and how AI will put human investment managers out of business, we first need to define what AI, machine learning, and deep learning are all about.
For the technically initiated, Deep Learning by Ian Goodfellow as well as online courses by Andrew Ng and at the Massachusetts Institute of Technology (MIT) are all great resources. But what about investment professionals with backgrounds in finance, not linear algebra and computer programming? What do we need to know to understand AI’s potential impact on the industry and our careers?
What is artificial intelligence (AI)?
At a basic level, AI is a branch of computer science that, to paraphrase Bill Gates’s mission when starting Microsoft Research, seeks to build computers that can see, hear, and understand humans.
Alan Turing’s work to make thinking machines, summarized in “Computer Machinery and Intelligence,” was a major milestone in AI’s history. In that 1950 paper, Turing asked, “Can machines communicate in natural language in a manner indistinguishable from that of a human being?” This is the essence of the famous Turing Test, which has become a key benchmark for generations of AI researchers in evaluating the power of their programs.
There seems to be general agreement that AI became an independent branch of scientific investigation in 1956, at the Dartmouth Summer Research Project on Artificial Intelligence.
For our purposes, the term AI applies to programs that simulate human cognitive abilities as well as those that process and apply the information captured. Natural language processing (NLP) and speech and image recognition applications are examples of AI. NLP seeks to understand written language texts. Speech recognition — say, turning voices or spoken language into text — is a related field. Image processing is another parallel field and is often referred to as image recognition or computer vision.
As a discipline, AI is fast-evolving and so is its definition. Applications that counted as AI just a few years ago, optical character recognition, or OCR, for example, may no longer.
What are machine learning and deep learning?
The quality of machine translation has improved by leaps and bounds in recent months. Recommendations from commercial services, such as Amazon and Netflix, have become ubiquitous. As with AlphaGo and AlphaGo Zero, these developments have all been driven by advances in machine-learning and deep-learning technologies.
So what are machine learning and deep learning? The term machine learning was coined in 1959 by computer scientist Arthur Samuel in reference to “a field of study that gives computers the ability to learn without being explicitly programmed.” Machine learning applications are AI programs that can write additional programs themselves to interpret input and predict output.
While machine learning may be a new term that many investment managers have only recently encountered, neural network is a related concept that finance professionals, particularly quants, may be more familiar with. Neural network is a form of machine learning inspired by how human brains process information. Yaser S. Abu-Mostafa of California Institute of Technology (Caltech) compares that relationship to the one between a plane and a bird.
Deep learning is among the hottest buzz words today. Many claim deep learning’s emergence has revitalized AI research. Deep learning is basically multi-layer neural networks: programs that process the initial input in multiple stages to generate the final output, at each stage taking the output of the last stage as the input. It is reminiscent of how we tend to break down complex tasks into a series of smaller steps. Deep learning is one of the many approaches to machine learning.
Together with the increase in computing power and the avalanche of data now available, advances in deep learning techniques have helped bring about the AI spring we are experiencing today.
Where are we on the journey of building a “seeing, hearing, and understanding” machine?
AI terminology and methods will continue to evolve. For example, Yann LeCun has recently proposed the term differential programming should replace deep learning. What really matters for investment professionals are the toolkits these innovations have made available to us.
So where does the scorecard stand today? Computers are making strides and may be outpacing us humans.
- In the ImageNet competition of 2017, AI programs beat the best human record by an increased margin, emphasizing that computers can now “see” images better than us.
- Last year, Google and Microsoft speech recognition programs transcribed as accurately as humans, so computers can now “hear” just as well as us.
- Two AI programs have succeeded in reading better than an average adult as of January 2018, so we can say computers can now “understand” us.
- In January 2018, Google debuted the Cloud AutoML platform that puts the power of machine learning in more programmers’ hands. So more machines will be able to see, hear, and understand.
AI technology has made tremendous progress in the last 12 months and many more tools are now at programmers’ disposal. The next big push will be applying AI across industries.
In the words of numerous industry heavyweights, AI is the new electricity.
I have no doubt it will light up many bulbs.
At the 71st CFA Institute Annual Conference in Hong Kong, David Pope, CFA, will discuss natural language processing (NLP) and corporate earnings sentiment analysis.
If you liked this post, don’t forget to subscribe to the Enterprising Investor.
All posts are the opinion of the author. As such, they should not be construed as investment advice, nor do the opinions expressed necessarily reflect the views of CFA Institute or the author’s employer.
Image credit: ©Getty Images/ nevarpp