Abhishek Gupta
Feb 22, 2018 · 8 min read

This article has been co-written with sydney, AI District Fellow at District 3

District 3 AI District’s mission is to explore the impact of AI in society, provide recommendations on the ethical development of AI and find ways to leverage novel technologies to enhance standards of living. The aim of this series of articles is to assess how AI will impact the financial sector, we begin the series with an introduction to the importance of this and lay out the foundation for upcoming articles.

Consider the following pieces of information:

32% of the financial sector jobs in the UK are at high risk of automation [1]

Bridgewater, the world’s biggest hedge fund is looking to cut jobs in favour of automation [2]

~3500 fintech startups that have an average $4.5 million valuation looking to disrupt the financial sector [3]

There are strong signals that despite the strength of the current financial sector and the central role that it plays in many developed economies, there will be a lot of disruption that will happen in the coming months and years, a lot of which will be powered by AI-enabled automation. Given the massive promise that this technology has to offer, finance has also tapped into deep learning techniques to gain an edge in a highly-regulated, fiercely competitive landscape.

Artificial intelligence companies saw a massive $4.1 billion in investment in Q4'17 [4] || Photo by Roberto Júnior on Unsplash

We use the term AI-enabled automation with particular care to highlight nuances between different types of automation and to also signal a call for attention on the sea-change amongst the noise of fear-mongering of labor impacts because of Artificial Intelligence, especially when the term AI is used in an all-encompassing way.

If we want to understand deeply how specific sub-sectors of finance are going to be affected by the use of AI, we need to understand the capabilities and limitations of these systems. Specifically, we can look at the underlying machine learning techniques, the types of datasets available and research advancements on platforms like arXiv [5] to gain a more thorough understanding of this rapidly evolving domain.

We don’t go into detail here on differentiating the terms AI, machine learning and deep learning but a great summary is available here. (Slides 6 to 14)

It is important to delineate automation as we’ve seen in the financial sector in the past vs. the newfound abilities that will cause a fundamental shift in the way the financial sector works. One of the key ideas that we need to examine is the degree of autonomy that some of these new systems have as this will be pivotal in determining the impact that these technologies will have on the labor market.

Autonomous vs. semi-autonomous systems

We can roughly break this down into two categories: semi-autonomous and autonomous. While a complete discussion of the different degrees of autonomy is out of scope of this article, autonomous systems are those that after being assigned defined high-level objectives, figure out the optimal path to completion with no human oversight. We can also be more granular and distinguish these systems from ones that would also have the ability to define certain sibling high-level goals that can create global optima. An illustrative example for autonomous systems is a deep space probe that is tasked with exploration activities and doesn’t receive commands from human operators because of it being prohibitively slow.

Primary difference in the spectrum of autonomy arises from the degree of independence the system has in determining the ideal path to completing a set objective || Photo by NeONBRAND on Unsplash

Semi-autonomous systems on the other hand rely on human intervention from time to time, either to re-orient them towards the right path or to co-develop the optimal path towards meeting the high-level objectives. A simple example to illustrate a semi-autonomous system would be a cleaning robot that can move around a space on its own, dodging obstacles but is also amenable to commands to begin and halt its activities from the owner.

We can also discuss here systems that require more frequent vs. less frequent human intervention but at this time such a distinction is purely theoretical in nature and will not enhance our understanding of labor impacts. [6]

Automation so far has typically gotten rid of physical labor which largely impacted low-skilled jobs leading to either displacement or a shift in the labor force towards operating such machines or creating new fields that coexisted with the work being done by machines. [7]

Impacts from AI-enabled automation will be significantly more pronounced than those coming from the automation wave during the Industrial Revolution || Photo by Brina Blum on Unsplash

The second wave of automation got rid of repetitive, menial tasks that require a mid-level of skills and education, the impact of which was a bit stronger as people needed to retrain and move to different fields or take on a more supporting or supervisory role on the tasks essentially being performed by a machine. [7]

Both the above are things that we are familiar with and many eminent economists [8] and public intellectuals have debated ad nauseam the pros and cons of labor-replacing technology’s role in economic growth and whether or not it is sustainable.

But comparing what we have imminent with AI-enabled automation, especially in making comparisons with the Industrial Revolution, downplay and misattribute the impacts quite dramatically.

AI-Enabled Automation

AI-enabled automation is a different beast altogether — we are now looking at the automation of high-skilled jobs that might not necessarily be repetitive. A canonical example of this is radiologists that are being challenged by automated systems that achieve a similar level of accuracy and in some cases finding patterns that even a highly trained radiologist missed.

Bradley Erickson, Director of the Radiology Informatics Lab at Mayo Clinic mentions that when we look at the frequency of cases, the more common ones are very susceptible to automation but there is a long tail of uncommon ones that would still require the expertise of a trained radiologist [9]. Thus, there is a lot of nuance here in terms of a reallocation of workload with the radiologist now perhaps only dealing in extremely complex cases that require cross-disciplinary collaboration to arrive at a final diagnosis vs. the simpler, routine cases passing through the machine.

Overall, it might reduce the need for many radiologists and could also potentially lower the barrier of entry for new radiologists as they can work alongside these systems to make judgements which would have previously required a high degree of skill. We observe this in the field of cybersecurity as well where new entrants can leverage AI-enabled tools to aid their work. [10]

A similar trend can be observed in financial trading where a lot of what used to be done by humans; analyzing trends and making decisions on which trades to make, is now done by algorithmic systems more efficiently and at scale. Quantopian presents a great example of how human traders can leverage a platform enabled with algorithms to run their trades.

Trading systems comprised of humans and algorithms working together will become more common as shown by Quantopian || Photo by Mark Finn on Unsplash

Another example is analyzing audit trails in financial transactions, if we translate the parameters that an analyst uses to distinguish valid from fraudulent transactions, then we can apply auditing at scale rather than sampling from the entire volume and looking for anomalies. [11] For example a survey in 2015 found that false declines, legitimate transactions that are wrongly rejected, cost retailers $118 billion; automated and more efficient fraud detection can help to mitigate these losses. [12] An initiative from Mastercard called the Decision Intelligence System doesn’t limit itself to pre-defined rules but learns on the fly to become better at fraud detection and prevention. [13]

Such a process also allows reduction in costs to meet compliance and regulatory requirements which are currently estimated to cost the industry $270 billion a year [14].

The benefit of automating these activities allows employees in the financial services industry to focus more on high-value tasks and helps them become strategic partners in business. [15] So instead of looking at this as a labor-replacing technological shift, we can think of it in terms of a labor-displacing shift where managers at these institutions need to identify how to harness higher levels of productivity from fewer errors, higher quality, speed and accuracy of work.

In associated fields of providing support services in finance and generally outside, with sophisticated voice recognition and language parsing, we can get rid of clunky automated call menus and increase customer satisfaction and engagement. This can also be extended to robo-advisors, chatbots that use predictive analytics and a chatbot interface to provide financial advice to customers, for example, Erica from Bank of America. [16]

This will have far reaching impacts as we fundamentally alter domains that require a high degree of education.

This wave is being dubbed “white-collar automation” and gives us a sense of the impact that this is going to have. As a recent example Goldman Sachs showed just how devastating automation can be to traders. In 2000, its U.S. cash equities trading desk in New York employed 600 traders. Today, that operation has two equity traders, with machines doing the rest. The rise of firms like Sentient Technologies, Numer.ai among others is evidence enough that the role of humans at least in this particular part of finance is long gone. And this is even before the full brunt of AI has taken over Wall Street. [17]

The next article in this series will focus on the historical trends in the intersection of AI and finance. This will be followed by an article on the AI and finance ecosystem.


Thanks for reading and feel free to share your thoughts in the comments section!

For more information on the work that Abhishek Gupta does in the ethical development of AI, please visit his website

If you’d like to follow Sydney Swaine Simon’s research, please visit his page


References:

[1] Richard Berriman, John Hawksworth, March 2017, Will robots steal our jobs? The potential impact of automation on the UK and other major economies

[2] Simone Foxman, Katherine Burton, and Saijel Kishan, Sept. 15, 2016, Bridgewater to Cut Jobs as Dalio Says Some Areas ‘Bloated’

[3] Analysis combined from public data available on fintech startups on CB Insights and AngelList

[4] Jennifer Samuel, Jan. 15, 2018, 2017 global venture capital investment hits decade high of US$155 billion following a strong Q4: KPMG Venture Pulse

[5] http://www.arxiv-sanity.com/search?q=finance

[6] Federal Agencies Ad Hoc Autonomy Levels for Unmanned Systems Working Group Participants, September 2004, Autonomy Levels for Unmanned Systems (ALFUS) Framework

[7] PwC Editorial Board, Feb. 6, 2018, PwC economists assess how and when workers will be affected by coming waves of automation

[8] Rebecca J. Rosen, Sept. 3, 2015, In Praise of Short-Term Thinking

[9] Dr. Bertalan Meskó, The Future of Radiology and Artificial Intelligence

[10] Ben Dickson, Aug. 31, 2017, Why the “AI is destroying jobs” argument is a bit overblown

[11] Robin Grosset, Nov. 15, 2017, The Impact of Artificial Intelligence and Machine Learning on Financial Services and the Wider Economy

[12] Al Pascual, Oct. 20, 2015, Future Proofing Card Authorization

[13] Matthew Cochrane, Jan. 4, 2017, Mastercard is using AI to improve the accuracy of its fraud protection

[14] Martin Arnold, Jan. 25, 2017, Banks’ AI plans threaten thousands of jobs

[15] Workday, Nov. 3, 2017, How AI and Automation Will Shape Finance in the Future

[16] Harriet Taylor, Oct. 24, 2016, Bank of America launches AI chatbot Erica — here’s what it does

[17] Kevin Maney, Feb. 26, 2017, Goldman Sacked: How Artificial Intelligence will transform Wall Street

District 3

Based at Concordia University, we work closely with the Montreal innovation ecosystem to ensure every innovator and entrepreneur can create an impact.

Thanks to Sydney Swaine-Simon and Rana Abbasi

Abhishek Gupta

Written by

SE #machinelearning @Microsoft | Founder @mtlaiethics | #AI #Ethics researcher @mcgillu | @mcgillu grad #deeplearning

District 3

Based at Concordia University, we work closely with the Montreal innovation ecosystem to ensure every innovator and entrepreneur can create an impact.

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade