By Mark Esposito, Aurélie Jean, Terence Tse and Guillaume Sibout
In AI, as in myriad other fields, past and present, technological developments go blazing ahead in advance of the regulators’ ability to define and impose the appropriate legislation and regulatory oversight. The key to ensuring the responsible application of AI technology is algorithmic governance.
Algorithmic governance: from law to ethics
In a groundbreaking decision that reverberated through June 2023, the European Parliament triggered a transformative era by adopting an all-encompassing legal framework meticulously designed to oversee the intricate tapestry of artificial intelligence (AI). This pivotal milestone marked the inauguration of a regulatory regime aimed at finely calibrating the development and application of AI systems based on their potential risk levels to citizens’ fundamental rights. This framework stands as a testament to the Parliament’s commitment to channelling the boundless potential of AI in a responsible manner, safeguarding the bedrock of individual freedoms and societal integrity.
As this vibrant AI landscape evolves, the spotlight inevitably falls on the indispensable role of algorithmic governance. This paramount concept underscores the pivotal participation of both public and private stakeholders. This eclectic group, encompassing end users and architects of AI solutions, occupies the epicentre of this epochal transition. Their mandate transcends the mere construction of cutting-edge algorithms; it entails the establishment of resilient algorithmic governance frameworks. These frameworks embody the bedrock of ethical AI integration, venturing beyond mere regulatory compliance to champion the principle of non-maleficence – a steadfast commitment to safeguarding the interests of even the smallest subset of users. Significantly, the advent of algorithmic governance also seeks to stave off the spectre of algorithmic controversies, which have captured headlines in recent times, casting a shadow over the credibility of AI applications.
Consider the illustrative case of OpenAI, an exemplar in the AI realm. With a proactive strategy, OpenAI could have deftly navigated the introduction of its conversational agent, ChatGPT, and the underlying GPT algorithm. By adopting a prescient approach, OpenAI could have preemptively addressed potential challenges that arise from widespread adoption. This includes the nuanced risks emanating from misinterpretations and approximations during user interactions. This succinctly encapsulates the crux of algorithmic governance, a multidimensional framework that fuses ethical tenets, operational blueprints, and strategic orchestrations.
In its essence, the cultivation of algorithmic governance pivots on the bedrock of ethical norms and best practices that underpin the development and deployment of these algorithms. The quest for equilibrium demands a methodology that harmoniously balances rigour and pragmatism, culminating in effective and quantifiable implementation. Algorithmic governance converges into a dynamic synergy of forward-looking vision, conscientious innovation, and the harmonisation of technological advancement with ethical sensitivities.
This epochal transformation necessitates an approach that transcends the confines of singularly technological considerations, embracing a multidisciplinary perspective. The expanse of algorithmic governance envelops a diverse spectrum of stakeholders, ranging from end users and developers to policymakers and the general public. The collaborative fabric woven among these varied groups underscores the intricate interplay of AI deployment, where the realms of algorithmic science intersect with legal, societal, and moral dimensions.
The complexity inherent in the deployment of AI underscores the indispensability of a nuanced approach to algorithmic governance. This involves a dual-pronged strategy that seamlessly merges top-down orchestration with bottom-up engagement. The former embodies strategic leadership and the enforcement of best practices, while the latter empowers every individual within an organisation to actively contribute through hands-on participation and user feedback. In this orchestration, transparency emerges as a cornerstone, fortifying the architecture of algorithmic governance. It emphasises not only the imperative of sharing source code but also divulging data sets and the criteria governing algorithmic training.
As the landscape of AI evolves, the imperatives of algorithmic governance models emerge as the lynchpin for responsible progress. However, a scarcity of well-defined models poses a challenge, even as entities earnestly embrace the ethos of algorithmic governance. There is no universal panacea; instead, organisations must meticulously tailor bespoke governance models that resonate with their industry, ethos, and goals. This diversity heralds the inherent dynamism within algorithmic governance, fostering a culture of adaptability and innovation.
In summation, the shifting paradigm of AI compels us to recalibrate our strategies and embrace algorithmic governance as our guiding compass. The European Parliament’s pioneering legislative stride underscores the gravitas of ethical considerations within AI deployment. Embarking on this journey necessitates a harmonious blend of technological leaps and ethical mindfulness, ensuring that the profound potential of AI is harnessed judiciously and inclusively. This synergy heralds a future where technology becomes a beacon, illuminating the path of humanity’s advancement.
What is algorithmic governance (and what is it not)?
Algorithm governance stands as the web of regulations and practices that intricately envelop the creation and utilisation of algorithms interwoven within AI technologies, all of which are meticulously tailored to serve specific use cases. It transcends the boundaries of being a mere ethical charter or a collection of abstract principles. Instead, it takes form as a comprehensive and systematic amalgamation of measures. These measures are not just intended to ensure the effective and efficient operation of the algorithm in question, but also to erect formidable safeguards against potential pitfalls. These pitfalls may range from insidious technological biases to any inadvertent divergence from legal obligations. In many ways, this construct parallels a legal framework, albeit one that necessitates a congruent judicial mechanism for perpetual enforcement and ongoing assessment of compliance.
Algorithmic governance, in essence, is a multidisciplinary and cross-cutting construct that calls upon a diverse range of disciplines, including sociology, politics, and anthropology. This intricate tapestry emerges as a conduit that seamlessly binds together a multitude of stakeholders engaged in a given project. It goes beyond mere technocratic considerations to integrate various levels of technological acumen and general understanding of algorithmic science. Notably, it also incorporates the rights, obligations, and responsibilities of these stakeholders in the context of the algorithm under scrutiny.
Algorithmic governance, in practice, necessitates the facilitation of two parallel and harmonious approaches: a historical, top-down orchestration and a grassroots-driven, bottom-up engagement. The former confers authority on steering committees or project leaders to inject and enforce best practices, thereby permeating the project’s trajectory with a strategic compass. Simultaneously, the bottom-up approach empowers every member of an organisation’s workforce or the end-user community. This engagement is manifold in its scope, ranging from tangible contributions within the project’s development to indirect but substantial involvement through the collection of invaluable user feedback. In unison, these diverse contributions combine to nurture the evolution of the project from its inception to full-fledged deployment.
Furthermore, effective algorithmic governance doesn’t exist in isolation. It necessitates a framework that upholds transparency in a manner that is universally understandable to stakeholders involved in projects encompassing the design, procurement, or utilisation of algorithms. Importantly, it’s crucial to underscore that absolute transparency of an algorithm’s source code should not serve as a mechanism for evasion of algorithmic governance. Rather, transparency should encompass a comprehensive unveiling, extending beyond mere source code disclosure. It should encompass all pertinent data sets utilised during the algorithm’s creation and the criteria driving its training process. This notion brings to light a delicate conundrum: while transparency is vital for accountability and oversight, excessive disclosure could inadvertently curtail innovation by exposing proprietary intellectual property to the public.
In summary, the lynchpin of algorithmic governance is a profound commitment to transparency, accountability, and the harmonious interaction of stakeholders. This complex symphony unfolds in a multifaceted framework that not only ensures the ethical, robust, and responsible integration of AI but also instils a culture of continuous refinement and advancement. In a rapidly evolving landscape where technology advances at an unprecedented pace, algorithmic governance stands as the beacon that guides AI’s evolution to align with human values, societal aspirations, and ethical standards.
Algorithmic governance models
Algorithm governance rises above the realm of mere ethical principles, manifesting as an intricate and comprehensive framework that intricately governs the landscape of algorithm creation and application within AI technologies meticulously designed for specific use cases. It assumes the role of a dynamic construct, interwoven with a plethora of protocols and regulations that not only ensure the accuracy and efficiency of algorithmic operations but also proactively anticipate and mitigate potential pitfalls. These pitfalls encompass a spectrum of concerns, from subtle technological biases to the intricate labyrinth of legal compliance. Just as a law stands incomplete without a robust judicial system to enforce it and continually assess its applicability, algorithmic governance demands an all-encompassing approach to safeguard its implementation and sustained effectiveness.
The very fabric of algorithmic governance transcends the boundaries of any single academic domain. It is instead an intricate and multidisciplinary tapestry that elegantly weaves together disciplines as diverse as sociology, politics, and anthropology. This cross-disciplinary approach empowers algorithmic governance to act as a connective thread, uniting a diverse array of stakeholders intricately involved in a project’s lifecycle. From the end users who engage with the technology to the creators who shape it, algorithmic governance seamlessly integrates their varying levels of technological literacy and broader comprehension of algorithmic principles. In doing so, it creates a harmonious fusion of the technical intricacies and the moral imperatives that underpin the very essence of algorithmic science.
Furthermore, this orchestrated integration doesn’t just stop at merging the realms of technology and ethics. It encapsulates the entire spectrum of stakeholders’ rights, responsibilities, and obligations in relation to the algorithm at hand. This holistic incorporation ensures that algorithmic governance operates as a unifying force, providing a cohesive roadmap that guides every participant towards responsible and ethical practices. It ensures that the pursuits of innovation are inextricably tied to the principles of accountability, transparency, and user-centricity.
In the grand tapestry of algorithmic governance, this multifaceted approach culminates in a symphony where technological advancements resonate harmoniously with ethical considerations. It facilitates a cohesive environment wherein algorithmic governance evolves as a guiding light, ensuring that AI technologies continue to enhance and enrich human experiences while adhering to the highest ethical standards. Just as society advances, so too must the principles governing its technological creations evolve, adapt, and flourish. This is the essence of algorithmic governance – a dynamic, multidisciplinary foundation that paves the way for a future where technology and ethics coalesce in unprecedented ways, fostering responsible innovation and enduring progress.
Algorithmic governance encapsulates a dual-pronged strategy that navigates both historical precedence and contemporary perspectives. This twofold approach integrates the wisdom of the past with the dynamism of the present, setting the stage for a comprehensive framework that transcends traditional governance paradigms.
On one hand, algorithmic governance preserves the essence of the historical top-down approach. This strategy empowers designated steering committees or visionary project leaders to inject and enforce optimal practices and their pragmatic applications. Through this avenue, a carefully curated collection of best practices is seamlessly infused into the project’s DNA, guiding its trajectory with a strategic compass. It’s akin to the conductor of an orchestra, harmonising the diverse elements to orchestrate a symphony of seamless functionality and ethical responsibility.
Conversely, algorithmic governance also embraces the innovative spirit of the bottom-up perspective. This paradigm shift democratises the governance process, endowing each individual within an organisation’s workforce or the end users themselves with a profound sense of agency. In this dynamic landscape, contributions take on various forms, from tangible, hands-on involvement in the project’s development to indirect, yet substantial, engagement, such as the collection of invaluable user feedback. This fusion of perspectives lends the project an inclusive ethos, ensuring that it resonates with the very individuals who engage with its outcomes.
As algorithmic governance unfurls, a core tenet lies in the imperative of unwavering transparency. Transparency operates as the bedrock upon which ethical accountability is built. It’s not merely about shedding light on the source code, but a comprehensive unveiling of the intricate layers that constitute the algorithmic machinery. However, it’s imperative to navigate this transparency with a balanced perspective. While the complete disclosure of source code is vital, it should not be exploited as an escape route from the vigilance of algorithmic governance. Transparency transcends the surface level and delves into the disclosure of data sets and the nuanced criteria that underlie algorithmic training. It’s a nuanced dance, for excessive disclosure could potentially impede innovation by exposing proprietary intellectual property. Striking this balance is critical to foster innovation while upholding ethical considerations.
Nevertheless, the cornerstone of transparency remains non-negotiable. It must be enforced unconditionally, despite the intricacies and challenges it presents. The drive for transparency is not just for the immediate gains but for the long-term ethical ascendancy, where algorithms operate as tools of empowerment and enablers of progress.
In essence, algorithmic governance is not just a strategy; it’s a symphony that harmonises the historical with the contemporary, the technical with the ethical, and the visionary with the pragmatic. It thrives in the dynamic interplay of perspectives, transforming traditional hierarchical structures into collaborative ecosystems. It is within this intricate interplay and harmonious transparency that the future of algorithmic governance emerges, a future where technology and ethics converge to propel innovation while safeguarding the values that define humanity’s progress.
In this intricate and symbiotic ecosystem where technology and ethics intertwine, algorithmic governance emerges as the lynchpin that steadfastly anchors AI’s transformative journey. It assumes a dual role, functioning not only as a protective shield that safeguards against potential pitfalls but also as an invigorating catalyst that propels innovation while upholding the sacred tenets of ethics. The delicate dance of striking this equilibrium bears witness to our profound capacity to wield cutting-edge technology with an unwavering sense of responsibility. It underscores our commitment to steering the remarkable potentials of AI towards avenues that elevate society, ensuring that the remarkable advancements it offers are channelled in ways that not only advance progress but also align harmoniously with the guiding principles of ethical excellence.
Which model of algorithmic governance should be adopted?
Within the rich and intricate tapestry of algorithmic governance, it becomes evident that a single model cannot encapsulate the vast and diverse landscape that encompasses this evolving field. Rather, an array of adaptable models stands poised to be meticulously tailored to the distinct contours of various industries, individual companies, and their unique aspirations. This appreciation for diversity, however, does not diminish the importance of establishing a foundational structure, a guiding template that lays the groundwork for the development of personalised and finely tuned governance systems. This foundational framework serves as a scaffold upon which the intricacies of algorithmic governance can be delicately woven, ensuring both consistency and adaptability.
This overarching structural blueprint is a compass that navigates the multifaceted journey of algorithmic project development. It commences with the ideation phase, a pivotal juncture where the contours of the business challenge are carefully delineated. This initial spark of conception sets the tone for what follows. The journey extends to the use phase, where the vital input from end users is integrated, forming a crucial feedback loop that iteratively refines and enhances the algorithm’s performance and user experience.
Intricately intertwined within this encompassing structure are the business and technical specification stages, where the practicalities of implementation are meticulously outlined. An often-overlooked aspect of this process is the environmental considerations linked to the extraction of raw materials for hardware production, a reminder that technology’s progress should align with ecological sustainability. Moreover, the spotlight falls on the computing power necessary for algorithmic training and execution, a facet closely tied to the energy consumption that underpins these technological marvels.
Data collection phases, representing the veins of algorithmic evolution, hold equal significance. Rigorous sampling methodologies and representativeness tests serve as the bedrock, ensuring the data’s integrity and relevance. The journey then navigates through the intricate realm of computer programming and algorithmic training, a process that involves moulding raw data into refined intelligence. The culmination lies in validation and deployment, a pivotal point where the algorithm’s readiness for real-world application is meticulously assessed.
However, the journey doesn’t cease here. Meticulous post-usage tests come into play, scrutinising the algorithm’s performance in real-world scenarios, often involving a diverse multitude of users. These post-usage tests serve as a final checkpoint, validating the algorithm’s efficacy beyond the controlled environment.
Amidst this intricate orchestration, nurturing the technical comprehension of stakeholders assumes paramount importance. This extends to both the business professionals who drive the project’s vision and the end users who interface with the technology. A collective understanding of the algorithmic nuances fosters a shared responsibility, ensuring that every stakeholder contributes meaningfully to the project’s success.
In the grand mosaic of algorithmic governance, this multi-phase structure stands as a testament to the meticulous dance between innovation and ethics. It’s a dynamic narrative that accommodates diversity, facilitates precision, and respects the dynamism of evolving technologies. As industries transition into the age of algorithmic empowerment, this adaptable yet foundational framework is the cornerstone upon which ethical and technologically pioneering future is built.
A pivotal facet of this journey involves the continual calculation of a risk score, an evolving metric that encapsulates the scientific, technical, ethical, and reputational risks aligned with the algorithm’s lifecycle. This dynamic evaluation process ensures a proactive approach, deftly addressing any potential pitfalls that could emerge.
In tandem, the inception of algorithmic governance warrants the drafting of charters and principles of good practice. These serve as the lodestars, offering guidance and solutions to the myriad questions and nuances that invariably surface throughout a project’s lifecycle. The presence of an internal or external ethics committee further enriches this landscape. Their involvement is instrumental in fostering a harmonious governance process, overseeing its trajectory, and ensuring its seamless operation.
A pivotal determinant of the triumph of algorithmic governance resides in the methodical and disciplined application of explainability calculations. Operating as sophisticated statistical methodologies, these calculations orchestrate an intricate symphony that elegantly unravels the algorithm’s inherent opacity. Through adeptly controlling and skilfully extracting the algorithm’s operational logic, these calculations wield a profound influence, serving as a safeguard against an array of potential issues that may emerge in the algorithmic terrain. This proactive and preemptive approach holds the potential to thwart errors, bugs, and the insidious algorithmic biases that can precipitate technological discrimination.
This exceptional capability finds its embodiment in its capacity to preempt facial recognition algorithms from manifesting biases rooted in skin colour or to forestall scenarios like the gender-based disparities in credit limits that were a result of the Goldman Sachs algorithm embedded within the Apple Card application. These instances serve as poignant reminders of the critical role that explainability calculations play in ensuring that algorithms function with an ethical equilibrium, preserving the sanctity of user interactions and the equitable distribution of opportunities.
By embracing these multifaceted dimensions, algorithmic governance emerges as a comprehensive and all-encompassing framework that embraces strategic structuring, meticulous risk assessment, unwavering ethical oversight, and a resolute commitment to technical transparency. This holistic and multifaceted approach not only erects an impregnable defence against potential pitfalls that may befall algorithmic endeavours but also illuminates the path toward an AI-driven future where the dynamic interplay of innovation and ethics converges harmoniously. This symbiotic coexistence is not just a fleeting aspiration but a tangible reality that beckons us to mould the future of technology with both ingenuity and ethical consciousness.
Where to start?
At the crux of fostering effective algorithmic governance lies a meticulous journey that begins with the company’s discernment of a precise business need ideally suited for algorithmic technology. This initial step entails identifying a use case that seamlessly integrates the potential of algorithms to address business challenges. To navigate this uncharted terrain, a synchronised effort between technical and business teams becomes indispensable. Collaboratively, they decipher the contours of each phase within the algorithmic governance framework, meticulously shaping its trajectory.
This collaborative endeavour involves sculpting the very essence of algorithmic governance for each phase, punctuating them with meticulously crafted questions, actions, and tests. These granular aspects serve as guiding beacons, illuminating the path to meticulous execution and oversight. At the heart of this lies the calculation of the algorithm’s risk score, a multidimensional assessment that evaluates scientific, technical, ethical, and reputational risks. Moreover, devising governance success metrics underlines the commitment to measuring the effectiveness and impact of the governance structure.
Once these parameters are established, the algorithmic governance blueprint is not confined to a singular use case. Rather, it undergoes validation through practical application across diverse scenarios, serving as a litmus test for its scalability and adaptability. This iterative assessment not only fine-tunes the governance model but also validates its efficacy in varied contexts.
The journey culminates in the seamless integration of the evolved algorithmic governance framework into the organisational fabric. Employing an iterative and agile approach, the company refines and adjusts the governance structure based on real-world deployment experiences. This dynamic evolution fosters a system of algorithms that transcends mere technical excellence, embracing inclusivity and environmental responsibility. The algorithmic solutions that emerge from this meticulous process stand as a testament to the harmonious coalescence of cutting-edge innovation with ethical and societal considerations.
Beyond the organisational sphere, this transparent approach to algorithmic governance holds the potential to catalyse broader changes. By openly communicating and sharing their governance frameworks, companies contribute to the larger discourse surrounding algorithmic ethics. This extends to legislative debates, where the insights derived from practical governance implementation play a pivotal role in shaping the contours of future laws governing algorithms. Thus, the journey that begins with a focused business need matures into a ripple effect that resonates across industries, legislation, and society at large, fostering a landscape where technology and ethics are seamlessly intertwined.
About the Authors
Dr Mark Esposito is Professor of Business & Economics at Hult International Business School, where he directs the Futures Impact Lab. He is Faculty at Harvard University’s Division of Continuing Education and Affiliate Faculty at the Microeconomics of Competitiveness programme at the Harvard Business School. He was appointed as a Professor of Public Policy at Mohammed Bin Rashid School of Government in Dubai. He serves as Senior Advisor to the Strategy& group at PwC. He is co-founder of Nexus FrontierTech, and the Circular Economy Alliance. He is co-author of The AI Republic, Understanding How the Future Unfolds, and The Emerging Economies under the Dome of the Fourth Industrial Revolution. His latest book, The Great Remobilization, is with MIT University Press.
Aurélie Jean, PhD, computational scientist, entrepreneur and author. Aurélie Jean has close to 20 years of experience in computational science applied to a broad range of disciplines. After 11 years of academic research, Aurélie is now running two companies, including a deeptech AI startup in early detection of breast cancer. She is the author of several bestseller non-fiction titles on algorithmic science, as well as a columnist on science and technology. Aurélie is teaching algorithmic science in executive education. She is also an investor and a board member of several companies in the United States and in France.
Terence Tse is professor of finance at Hult International Business School. He is also a co-founder and Executive Director of Nexus FrontierTech, an AI scale-up. Terence has appeared on television, in radio shows and in periodicals. He has given seminars, workshops, and speeches for and to some 50 organisations. Terence has written three books, with the next to be published by MIT Press in 2024. He is on the board of various entities, including Nexus FrontierTech, Thyreality, Tolar HashNET, and Circular Economy Alliance. Previously, he was in investment banking and consulting. Terence has a PhD from the University of Cambridge, UK.
