Artificial Intelligence: An Introduction

It has been a while since I wanted to write about Artificial Intelligence. The recent surge of interest in this field has garnered attention from everywhere, including academia, industry, government and many individuals who are interested in such topic because of the potential magnitude of impact it could have on our world and upon our race. Interestingly, Artificial Intelligence did not raise much of my attention in the beginning when I first heard a few times about it during the end of my senior year at Yale, more than half a year ago. When a friend of mine talked to me about his unwavering academic interest in Artificial Intelligence, I could barely share his passion for such field. Although further conversation with him indeed triggered me to think more about this topic, it was not until more than half a year later that Artificial Intelligence entered my sight and grabbed my attention.

The change of my attitude towards Artificial Intelligence is worth exploring a bit, as it shows exactly why AI may sometimes be misunderstood. The reservation and doubt I had on AI when listening to my friend came from disbelief on AI’s potential. To summarize it, it was simply that “AI could never be as advanced as human brains and it can never reach a level where it can replace human function.” In retrospection, my disbelief is not wrong per se, but it certainly does not do justice to AI. The purpose of AI should not be to replace human function (then it’d better just terminate us) but rather to aid us to do our job much better and quicker. Although the penetration of AI in many industries is still quite premature, the amount of investment in AI speaks to the optimism the world now has about the future of AI. Therefore, I think it is a good time for me to write something on this topic, especially after having attended an event held by Hong Kong Artificial Intelligence Society two days ago in IFC.

The event took place on the 22F IFC hosted by Macquarie. In this and the next paragraphs, I wanted to reiterate some of the key highlights and takeaways from the event. To talk about Al generally, it appears to be the case that lots of companies are asking how the adoption of AI could help boost their businesses; however, their understanding of AI is not very comprehensive. Two models of AI were brought up during the conference. The first is the so-called “Retrieval Based Models.” This model is essentially a decision tree model, where a signal input will go through the branches of the tree to yield an output. The second is the so-called “Generative Models,” where it models how the brain works. The second model is a newer concept and there is less a sense of control and structure embedded within the model to the extent that humans may even lose control of it if we start to fail to understand what the machines are doing. Compared to the first model which employs a huge amount of “if statements,” the second model relies on “intelligent” algorithms that can learn and extrapolate from the past. The idea is simple that the machines can be trained to write programs on their own. Although the word “Artificial Intelligence” was coined long ago in 1955 by John McCarthy, a Stanford computer scientist, the idea did not popularize until recent years. This is largely due to Three Factors. The first one is the coming of Big Data. The second is cost-efficient computer power. The third is adoption as there are more talents in this field.

After a brief overview of AI, the focus of the conference switched to institutional trading. To summarize, AI-driven trading algorithm is designed for two purposes --- to 1) minimize the execution cost and 2) to preserve alpha return (including arbitrage on market imbalance). The first aligns more with the sell side, while the second aligns more with the buy side. The current view on AI use in institutional trading is less sophisticated than what the public thinks, as banks and funds are actually pretty slow in adopting AI technology. The implementation of AI in institutional trading is to develop a mechanic and systematic way of selecting from over 1,000 currently available trading algorithms. Through reinforcement, the machine could be trained to “intellectually” select a combination of optimal algorithms given different situations. During the Q&A question, one question deserves some mentioning. The theme of financial crisis was again brought up to the table. The question was how to avoid another financial crisis in the age of AI? Given the limited occurrences of financial crisis in the history, how can machines be trained to sense the coming of another one if not be trained to act only out of the majority of the financial data which were not generated from the periods of crisis? None of the speakers had definitive answers for this question, and the general answer was to analyze the context of the financial crisis and let the machines to study those contexts so that whenever any remote similarity occurred, the machines would raise alerts.

So much for that AI event, the first one I have attended in Hong Kong on such topic. In the past couple years, AI has obtained enough media attention and it has become a catchy word representing not just the actual technology which is far from being clearly defined or well understood, but the future itself because of which it has triggered many questions on human’s existential crisis. Many industries have started to invest in AI and tried to understand how the implementation of AI could help them to reduce costs and generate more revenues. I will try to write more about AI through the lens of different industries and sectors at another time. Financial institution is one of the sectors that are mostly eager to understand and adopt AI in its daily operation. However, despite many moves by the banks and funds to open up the conversation on AI, little has actually been accomplished to accommodate AI to daily operation, not to mention how to use AI to make significant impact on profitability while ensuring all the regulatory risks are responsibly covered. Maybe it is inevitably true that it is a game of trial and error. No one yet knows how AI is going to change the industry landscape, but someone has to try first.

AI is like teenager sex --- everybody is talking about it and bragging about how much he knows about it, but no one has really had much experience with it, not to say master it. It is a sharper sword that can be used to kill fast, for good and, unfortunately, for bad as well. It raises the stake of the game, the game every one of us is inescapably part of. A recent movie “Geostorm” is certainly a good demonstration on how advanced technology could backfire to the extent that it could potentially wipe out our race. That is why it calls upon us to discuss it and think about it in a very responsible and thorough way. Now is the time we must start to do so! Although much is still unknown and unexplored, one thing is for sure: in order to play the game safely, we need to understand, control and master AI technology before completely unleashing its tremendous power.