AI & Funds #1 – investment discretion

When one instructs a painter to draw a portrait (no matter how detailed the instructions may be) the painter will always have discretion in executing the final product. Likewise, in investment funds, the investment manager is instructed to invest according to various rules and timeframes which are detailed in the investment strategy as approved by the competent financial regulator, but there will still remain investment discretion on how the investment strategy is executed. This discretion is normally based on a number of statistically orientated formulas which analyses quantitative and qualitative data.

Artificial Intelligence (“AI”) is being increasingly used in these formulas to assist investment managers in forming their investment discretion. Although the use of formulas against sets of data has been practised for decades by investment managers, the European Securities and Markets Authority (“ESMA”)’s report on ‘Trends, Risks and Vulnerabilities’ on AI noted in February 2023 that “the scale at which AI can be used, the speed at which AI systems operate, and the complexity of the underlying models may pose challenges to the market participants intending to use them and to their supervisors.” ¹

So far, the use of AI has been generally limited by investment managers as a mechanism to “execute specific tasks that leverage large amounts of data.”² The AI that is being currently applied at the date of this article is largely the result of machine-learning rather than actual intelligence which was generated artificially. Whilst the technology to have an investment fund which is run entirely on AI’s investment discretion is already in existence, the investment discretion is still currently vested within a regulated investment manager where humans are ultimately liable for the consequences of the final decision, irrespective of the level of reliance on the AI.

This raises the question: when does investment discretion shift from humans to AI? Such determination is necessary from a legal standpoint to determine: (1) who has liability which arises from bad investment discretion; (2) when to request authorisation from the competent authority to update the investment strategy; and (3) how should conflict of interests be mitigated.

To determine when the investment discretion has shifted from a human to an AI, one must first define AI. The European Union (“EU”)’s draft AI Regulation is still a proposed legislation which has been sent by the EU Commission to the EU legislative bodies, but to date, it has not been approved yet.   Article 3 of the draft EU AI Regulation says that AI is a:

“software that is developed with one or more of the techniques and approaches listed in Annex I and can, for a given set of human-defined objectives, generate outputs such as content, predictions, recommendations, or decisions influencing the environments they interact with.”³ 

In Annex 1, the EU Commission (the executive branch of the EU) elaborates that the techniques mentioned in Article 3 of the proposed EU AI Regulation are:

• “Machine learning approaches, including supervised, unsupervised and reinforcement learning, using a wide variety of methods including deep learning;
• Logic- and knowledge-based approaches, including knowledge representation, inductive (logic) programming, knowledge bases, inference and deductive engines, (symbolic) reasoning and expert systems;
• Statistical approaches, Bayesian estimation, search and optimisation methods.” ⁴

Subsequently, as a reply to the EU Commission’s draft EU AI Regulation, the European Council (the upper legislative body of the EU) narrowed down the definition in Article 3(1) to only “systems developed through machine learning approaches and logic/knowledge-based approaches”⁵. According to the European Council itself, the purpose for the proposed textual reduction was “to ensure that the definition of an AI system provides sufficiently clear criteria for distinguishing AI from more classical software systems.”⁶

From the Maltese perspective, although the Maltese State has issued a comprehensive policy on artificial intelligence, no AI legislation has been passed by the Maltese Parliament either. Therefore, to date, there is neither a definition of AI from EU law nor from Maltese law.

Without a legal definition of AI, the question on when the investment discretion shifts from humans to AI cannot be answered to determine liability from investment discretion and other regulatory considerations relating to the licensing of the financial products.  As of February 2023, investment discretion is still vested in humans who remain liable if there is a misinterpretation of the result of AI-powered formulas. This is because only human-backed investment managers are licensed by the financial regulators.

Nevertheless, if the proposed EU AI Regulation becomes law and financial regulators introduce policies which govern AI, there would be nothing from stopping investment discretion to be gradually shifted from humans to AI because Article 3 of the proposed EU AI Regulation includes both the stricto sensu meaning of artificially-generated intelligence as well as the more basic form of machine learning which is already being widely consulted by investment managers under AI’s umbrella term.

Such a shift would create more transparency because the steps taken by the AI to reach an investment decision would be recorded in a coding language which can be extracted and reproduced in quarterly reports. Whilst this can be beneficial for investors and investment managers alike, certain investors might still prefer humans to retain discretion over their investments or pension funds. In practice, the probable outcome would be a hybrid model where certain aspects of the investment funds remain managed by humans’ investment discretion and others are allocated to the AI’s investment discretion.