WHEN ALGORITHMS BECOME DIRECTORS: SHOULD AUTONOMOUS AI SYSTEMS OWE FIDUCIARY DUTIES UNDER COMPANY LAW?

INTRODUCTION

Company law has traditionally been based on the premise that a company is run by human beings with judgment, who are accountable and can make decisions in the best interests of the company. Directors are entrusted with great powers in the conduct of corporate affairs and are in positions of trust. To ensure that such powers will be used responsibly, company law imposes fiduciary duties that make it the responsibility of the directors to act in good faith, use reasonable care, avoid conflicts of interest, and consider the well-being of the company and its stakeholders.[1]

But as AI is currently developing and maturing at a fast pace, these assumptions are starting to be undermined. AI systems are no longer confined to doing only routine administrative tasks. They are now becoming even more prevalent in the analysis of financial markets, forecasting consumer behaviour, risk management, resource allocation, and even strategic decision-making. Many companies already have algorithms involved in many processes that used to be left to top management.[2] With the increasing shift towards autonomous and sophisticated AI, a significant legal issue arises: should such an AI system, that effectively acts as a director, be held accountable to fiduciary duties under the Companies Act?

This is no “academic” question. As more and more companies are moving towards self-governing technologies, essential tasks could in the future be outsourced to AI systems. Thus, laws and policies have to take into consideration the effects of algorithmic governance, while regulatory action must not lag behind technological advancements.

THE TRADITIONAL FOUNDATION OF FIDUCIARY DUTIES

Fiduciary duties form one of the corner stones of Corporate Governance. Directors have a duty to conduct themselves in good faith, to use their independent judgment in making decisions, and to avoid self-dealing and decisions that are not in the best interests of the company. These duties are in place because the directors have discretionary power which may have an impact on the shareholders, employees, creditors and the general public.

A fiduciary responsibility is a legal relationship that relies on trust. Directors are expected to apply their judgment, with experience, ethical reasoning and considering the interests of the company in the long run. If directors fail to fulfill these duties, courts could find them personally liable, or remove them from office, or require them to pay compensation for losses due to their misconduct.[3]

Thus, the force of the fiduciary duties relies upon a human decision-maker who can be held accountable in their decision making. AI systems are defying this paradigm because they can have an impact without being conscious, having intentions or even being morally responsible.

ALGORITHMIC GOVERNANCE: WHAT IS IT AND HOW HAS IT EMERGED?

AI is now a key element of business administration. AI enables financial institutions to detect investment opportunities and analyze risks in the market. Businesses use machine-learning to improve their operations and forecast client desires. Automated systems are used by multinational corporations in supply chain management, to assess employee performance, and to prevent fraud.

Despite the fact that human oversight is still “officially” there, in practice decisions are not always made in that way. In some cases, executives might take AI’s recommendations without fully grasping how a specific conclusion was attained. This is especially the case for complex machine learning models that are opaque: the input data to the model can be transformed to a range of outputs that are often hard to explain even to the creators of the model.[4]

In the future, with the advancement of AI technology, companies might grant more extensive autonomy to autonomous systems for decision-making. An AI system could potentially make investment decisions, negotiate contracts, minimize financial risks, or allocate corporate resources, without significant human oversight.[5] These developments make the distinction between a decision-support tool, and a decision-maker, more indistinct.

Algorithmic governance has brought to the fore accountability and responsibility issues. Under traditional legal principles, a system of AI that is capable of effectively managing important parts of corporate management may be found to be unfairly unacceptable.

THE ACCOUNTABILITY GAP

One of the biggest legal questions surrounding autonomous AI systems is the development of an accountability void. The responsibility in conventional corporate law can be connected to particular individuals when it comes to harmful decisions. With negligence or dishonesty, legal systems may impose consequences for the directors.

Things get complicated when decisions are made by independent algorithms. While it is hard to attribute responsibility when an AI system leads to financial loss, discriminate against stakeholders, or resorts to illegal means, these are all scenarios that can occur.[6] There are a number of potential targets for liability.

Responsibility could be leveled at the directors who gave approval for the use of the AI system. However, directors might not be fully aware of the technical process involved in algorithmic decisions.[7]

Second, software developers might be liable. However, developers typically have no control over how corporations are implementing or using their technology.[8]

Third, the corporation could be liable. This may put money into the hands of those who have been affected by it but does not mean that the problem will necessarily be the responsibility of the individual who is responsible for setting up a system.

The lack of clear framework, creates uncertainty, and may lead to the use of automated decisions by corporations without proper safeguards. Thus, the issue of filling the accountability deficit has become an urgent problem in the present company law.

SHOULD AI OWE FIDUCIARY DUTIES?

It has been proposed by some legal scholars that highly autonomous AIs should be given some modest form of legal personality.[9] Per this strategy, AI entities could in theory have rights and obligations comparable to corporations. Imagine AI was legally considered as a person, then it would be possible that fiduciary duties would be attributed to algorithmic decision makers themselves.

While this may sound good in theory, there are real-life and philosophical problems with this approach.[10] The concepts underlying fiduciary duties are loyalty, trust, good faith and honesty. The underlying assumption of these arguments is that there are moral agents, moral intentions. No matter how sophisticated, artificial intelligence does not possess feelings, moral sense, and independent interests.[11] It is not capable of meaningfully being loyal or acting with good faith as would a human director.

Therefore, it might be conceptually problematic to impose fiduciary duties directly on AI systems. A far better means of achieving this is to retain the human responsibility and instead to update the fiduciary standards to match the new algorithmic governance.

CONCLUSION

Artificial Intelligence in corporate decision-making is one of the most crucial issues in the present times in the field of company law. The role of directorship, accountability, and fiduciary responsibility is being challenged more than ever as AI systems take on a more active part in strategic and operational decision-making.

But instead of seeing AI as a legal director, regulators should concentrate on making humans responsible for the usage and supervision of algorithmic systems. AI using directors should face additional monitoring, transparency and risk management obligations. The application of the existing fiduciary theories to the new age of technological innovation can help ensure that companies don’t sacrifice accountability while still exploiting the advantages of artificial intelligence. Corporate governance may not be the next frontier of robotization but nonetheless is set to increasingly feature decisions that will depend on human responsibility as machines become the key input and output.

REFERENCES

CASES

In re Caremark Int’l Inc. Derivative Litig., 698 A.2d 959 (Del. Ch. 1996).

Marchand v. Barnhill, 212 A.3d 805 (Del. 2019).

Statutes & Legislative Materials

Companies Act, 2013, § 166, No. 18, Acts of Parliament, 2013 (India).

Companies Act 2006, c. 46, § 172 (UK).

European Parliament Resolution of 16 February 2017 on Civil Law Rules on Robotics, Eur. Parl. Doc. 2015/2103(INL).

SECONDARY SOURCES

Bathaee, Yavar, The Artificial Intelligence Black Box and the Failure of Intent and Causation, 31 Harv. J.L. & Tech. 889 (2018).

Bryson, Joanna J., Mihailis E. Diamantis & Thomas D. Grant, Of, for, and by the People: The Legal Lacuna of Synthetic Persons, 25 Artificial Intelligence & L. 273 (2017).

Li, Zhaoyi, Artificial Fiduciaries, 81 Wash. & Lee L. Rev. (forthcoming 2024), https://ssrn.com/abstract=4524129.

Matthias, Andreas, The Responsibility Gap: Ascribing Responsibility for the Actions of Learning Automata, 6 Ethics & Info. Tech. 175 (2004).

Siebecker, Michael R., Making Corporations More Humane Through Artificial Intelligence, 45 J. Corp. L. 95 (2019).

[1]Companies Act, 2013, § 166(2)–(4), No. 18, Acts of Parliament, 2013 (India); see also Companies Act 2006, c. 46, § 172 (UK) (imposing a comparable statutory duty on directors to act in good faith to promote the success of the company for the benefit of its members as a whole).

[2]Michael R. Siebecker, Making Corporations More Humane Through Artificial Intelligence, 45 J. Corp. L. 95, 99–101 (2019) (cataloguing the expanding use of artificial intelligence by firms such as Goldman Sachs and Amazon to identify market risks, streamline operations, and inform investment decisions).

[3]Companies Act, 2013, § 166(5)–(7), No. 18, Acts of Parliament, 2013 (India) (providing for disgorgement of undue gain and monetary penalties for breach of directors’ statutory duties).

[4]Yavar Bathaee, The Artificial Intelligence Black Box and the Failure of Intent and Causation, 31 Harv. J.L. & Tech. 889, 899–905 (2018) (explaining that the internal operations of complex machine-learning models are frequently opaque even to the engineers who design them).

[5]Zhaoyi Li, Artificial Fiduciaries, 81 Wash. & Lee L. Rev. (forthcoming 2024) (manuscript at 2–4), https://ssrn.com/abstract=4524129 (describing the prospect of AI systems exercising independent, board-level decision-making authority over investments and resource allocation).

[6]Andreas Matthias, The Responsibility Gap: Ascribing Responsibility for the Actions of Learning Automata, 6 Ethics & Info. Tech. 175, 177 (2004) (arguing that autonomous, learning systems create a gap that traditional concepts of responsibility ascription cannot bridge).

[7]In re Caremark Int’l Inc. Derivative Litig., 698 A.2d 959, 970 (Del. Ch. 1996); Marchand v. Barnhill, 212 A.3d 805, 824 (Del. 2019) (holding that a board’s good-faith obligation requires implementing reasonable information and reporting systems over risks that are “mission critical” to the company’s business, which may extend to a corporation’s reliance on algorithmic systems).

[8]Bathaee, supra note 4, at 922–25 (noting that developers of machine-learning systems generally cannot foresee or control how those systems will be deployed once integrated into a client’s operations).

[9]Eur. Parl. Resolution of Feb. 16, 2017 on Civil Law Rules on Robotics, Eur. Parl. Doc. 2015/2103(INL), ¶ 59(f) (recommending that the European Commission consider creating a status of “electronic person” for the most sophisticated autonomous robots).

[10]Joanna J. Bryson, Mihailis E. Diamantis & Thomas D. Grant, Of, for, and by the People: The Legal Lacuna of Synthetic Persons, 25 Artificial Intelligence & L. 273, 285–88 (2017) (concluding that the difficulties of holding an “electronic person” accountable outweigh the speculative interests that AI legal personhood might protect).

[11]Bryson, Diamantis & Grant, supra note 10, at 280 (distinguishing the legal fiction of corporate personhood, which is ultimately reducible to human principals, from personhood claimed for a purely synthetic entity).