SEBI's Circular: The black box conundrum and misrepresentation in AI-based Mutual Funds

Last week, SEBI released a circular imposing reporting requirements for the use of artificial intelligence or machine learning in mutual funds. SEBI is, via the Circular, conducting a survey and creating an inventory of the AI/ML landscape, including the use natural language processing, neural networks, statistical heuristic methods, etc. The aim, through the quarterly reports which are now to be filed, is to develop an in-depth understanding of the adoption of such technologies in the financial market, which will guide AI/ML policies in the future. The primary concerns expressed by SEBI, for the future, are with the black box nature of AI/ML systems, and the need thereby to ensure that there is no misrepresentation to investors on the abilities of such technologies. A recent case that perhaps triggered the release of this Circular at this juncture is the report of a Hong Kong businessman who is suing for investment losses to the tune of $20 million, triggered by a robot. [caption id=“attachment_6625241” align=“alignnone” width=“1024”] The logo of the Securities and Exchange Board of India (SEBI). Reuters[/caption]

75 percent of trading globally is via algorithms

Algorithmic trading in itself is nothing new, and globally, 75 percent of trades are executed via algorithms, and so are 30-50 percent of trades in developing economies like India (as per the Global Algorithmic Trading Market 2018-2022 report published by Research and Markets). While this involves the use of algorithms with more or less fixed codes and strategies, the use of AI and Machine Learning in trading has led to algorithms that are continuously evolving, developing newer strategies as they learn more. To outline some uses of AI and ML in trading today, the company Trading Technologies uses an AI platform which identifies complex trading patterns, on a massive scale across multiple markets, in real-time. CLSA, a global Asian investment group, uses machine learning and Natural Language Processing to identify market signals from news and research documents. Taking a slightly different approach, an Indian company Auquan provides a platform to crowdsource data-driven trading strategies from a community of data scientists, developers, and machine learning engineers. It then uses machine learning, big data and predictive analysis to help companies translate the human skills into trading profits.

Retail investors and AI-based trading

In addition to institutional traders, brokers and fund houses which are tapping into the revolutionary potential of AI, retail investors are also becoming aware of algorithm-based trading platforms that are available to them. The dangers of this are, naturally, the possible advertisement of ‘guaranteed returns’ on account of the AI, and the risk thereby to more gullible investors and to the market in general. The assumption that AI could be a better decision maker than a human only adds to this problem. [caption id=“attachment_4429165” align=“alignnone” width=“825”] An Indian broker reacts while trading at a stock brokerage firm in Mumbai. Photo courtesy Reuters[/caption]

Li Kin-Kan’s suit on supercomputer K1

The case of the Hong Kong businessman, Samanthur Li Kin-Kan, demonstrates the concerns that arise. Li Kin-Kan allowed a robot based hedge fund, controlled by a supercomputer K1, to manage $2.5 Billion in March, 2017. This was after being convinced by Tyndaris, a London based investment advisory firm that was offering K1, of the abilities of the supercomputer and on being shown simulations promising returns in double digits. K1 apparently would comb through real-time news and social media to make stock market predictions and would execute trades and adjust its trading strategies as it learnt more. As per reports, the supercomputer quickly started to lose money, including losses of $20 million in a single day, which led to the suit. This case, in fact, is said to be among the first cases where humans are going to court over investment losses caused by a robot. Due to the impracticality of suing the robot itself, Li Kin-Kan is instead suing Tyndaris for allegedly exaggerating the abilities of K1. Tyndaris, in turn, has countersued for unpaid fees and has claimed that it never guaranteed that the robot would make returns.

The black box conundrum

The case has brought the spotlight onto the concerns that arise, both of which are outlined in the SEBI Circular. The first is the black box issue, where the lack of understanding as to how or why an AI takes a decision calls into question who is to be held accountable when things go wrong. The second is with misrepresentation to investors as to the abilities of AI-based technology. The use of robots and AI in decision making has long since raised issues of ethics and accountability. One can recall the issue on who must be held accountable for a death caused by a self-driving car- the car manufacturer, the programmer, or any human who may be present in the car at the time. These issues are only compounded when factors like machine learning come into the picture, where the users or makers of the algorithm are quickly no longer able to understand how it functions. This black box issue has now moved into the investment space as well.

SEBI and international regulations on algorithmic trading

While regulations on algorithmic trading have been issued in India and around the world, these do not address the specific issues that arise from a consumer’s perspective, such as the black box issue. In the EU, for instance, Article 17 of the Directive on Markets in Financial Instruments has requirements like ensuring business continuity, notifying authorities of use of algorithmic trading, keeping records of high-frequency algorithmic trade orders, etc. Disclosure requirements, for instance, include factors like a description of the nature of its trading strategies, the trading parameters or limits, and key compliance and risk controls. [caption id=“attachment_5947921” align=“alignnone” width=“1280”] Representational image. AFP[/caption] The aim of these measures is towards the broader goal of ensuring market stability, such as to prevent and detect an AI initiated flash crash.  The same is reflected in Indian regulations on algorithmic trading as well. This was first allowed by SEBI in 2008, followed by detailed guidelines in 2012 and 2013, which were revamped recently in 2018. These measures were chiefly directed at stock exchanges, and involved requirements like implementing economic disincentives for high daily order-to-trade ratios, encouraging co-location facilities, providing tick-by-tick data free of charge, and requiring the monitoring of algorithmic trading through tagging of algorithms to ensure an audit trail. The boost to investor confidence The concerns expressed in this latest Circular are more directly consumer-centric, with the SEBI expressly seeking that any advertised financial benefit on account of AI/ML should not constitute misrepresentation. Among the details sought in the quarterly reports include specific information on how the use of AI/ML is portrayed in the product offering and the claims that have been made about the AI/ML system. SEBI has, under, the Circular, also sought several details to gain a larger picture on the use of AI, such as of the type of area where the AI/ML is used; involvement in order initiation, routing and execution; dissemination of investment/trading related advise/strategies; use in cybersecurity to detect attacks; the safeguards in place to prevent abnormal behavior; etc. While there has been activity around algorithmic trading regulation by SEBI for a while now, this Circular is welcome for its focus on investor protection. This can also serve as a boost to investor confidence in AI-based trading. It will also be interesting to see the approach taken by SEBI to address the black box issue, given the worldwide nature of the issue. The author is a lawyer specializing in technology, privacy and cyber laws.