AI ethics and data governance

Give a brief overview of the initiative, its objectives, and why you decided to undertake it. 

EOS at Federated Hermes (EOS) launched a combined thought leadership and public advocacy project in 2019 that aimed to help companies better understand the risks associated with artificial intelligence (AI) and data governance. The firm has created frameworks and tools that investors can use to address issues around freedom of speech, supply chains, data privacy, surveillance, user manipulation, bias and discrimination. 

The cloud computing and analytics services offered by Big Tech companies such as 

Alphabet (Google’s holding company), Facebook, Amazon, Microsoft and IBM can be incredibly useful for companies that want to use Big Data to profile their client base, develop predictive analytics and customise their services. However, a lack of awareness of the potential inherent biases in the data or analytical modelling, can lead to accusations of discrimination and unfair outcomes. Examples include the fine levied on Liberty Mutual¹ in the UK over its complaints procedure, and accusations of gender discrimination levelled at Goldman Sachs as a result of a credit card algorithm, which it denied.² 

EOS’s objective was to make companies aware of the risks related to privacy, bias and discrimination when using AI, and then to develop a risk‐aware culture at different levels within a firm. EOS has built its engagement strategy around two white papers: 

1. Investors’ Expectations on Responsible Artificial Intelligence and Data Governance was published in April 2019, in conjunction with law firm Bryan Cave Leighton Paisner (BCLP), and addresses concerns from a legal and compliance, and a technical perspective. 

2. Artificial Intelligence Applications in Financial Services was published in December 2019, in conjunction with Oliver Wyman, Marsh and BCLP, and addresses issues that affect asset managers, banks and insurers. 

Describe how your project is aligned to Active Ownership 2.0, including: 

a. The significance of the systemic, real‐world outcomes it seeks. 

EOS wants to help companies become more transparent in how they use Big Data and machine learning. It expects companies to commit to overseeing the respect of all human rights and to publish AI principles, applicable use cases and white papers that highlight their challenges and limitations. For example, Google’s white papers on AI and on Disinformation are positive first steps. Although the material social issues may differ from sector to sector, the right to privacy, life, and equality and non‐discrimination will always apply, as these are fundamental human rights. The example of Alphabet raises a number of issues: 

• Right to Privacy:  Through its technologies and relationships, Alphabet currently collects sensitive personal data, which may include health, biometric, and real‐time location tracking. It deploys listening features on Google Home products, which was not disclosed when this was launched. Also, under Project Nightingale, which was revealed by a whistleblower and widely reported in the media in November 20193, Google allegedly collected the personal medical data of up to 50 million Americans from one of the largest healthcare providers in the US, without informing patients. This reportedly included individual names and medical histories, which could be accessed by Google staff. The allegation raises urgent questions about privacy, due process, and the civil rights of billions of individuals. 
• Right to Political Participation: While Google has sought to limit microtargeting of political advertising, which risks the manipulation of specific groups or individuals, and sought to limit the spread of disinformation, political advertisers can still target using age, gender and area code. 
• Right to Freedom of Opinion, Expression, and Information: The right to freedom of information is jeopardised by micro‐targeting and profiling, with no transparency on the algorithm that puts people into targeted buckets. 
• Right to Health: Multiple sources have documented health and safety impacts on YouTube content moderators, especially mental health issuesarising from their work. 
• Right to Equality & Non‐Discrimination, Right to Life & Security: YouTube was used to share scenes of the New Zealand mosque massacre in 2019. At present, the proliferation of hate speech and acts of violence still appear to occur on the platform.  

b. The ambition, ingenuity or effort in the responsible investment tools/activities that were deployed. 

EOS has engaged with 60 companies across the tech, banking and pharmaceutical sectors in the US, Europe and Asia. This included sending letters to companies outlining its concerns and requesting further information on their approach to AI and data governance risks. The firm also conducted an initial benchmarking of the management performance at banks on AI/data governance. 

In 2019 EOS escalated its engagement at Alphabet, supporting a shareholder proposal calling for the establishment of a societal risk oversight committee, and speaking at its annual shareholder meeting, where the firm called for board directors to answer to shareholders. 

Following the 2019 AGM, EOS stepped up the pressure on Alphabet. 

Despite product‐level improvements, EOS still believes the company falls short on governance oversight on human rights‐related matters concerning the use of AI, including its overarching strategy on government‐related contracts and operations in non-democratic markets. 

In November 2019, the firm sent a private letter to Alphabet signed by over 80 institutional investors representing nearly $10 trillion in assets under management and advice, raising concerns about the company’s lack of responsiveness on ESG-related issues and requesting a dialogue with Alphabet on human rights‐related issues. 

EOS was a co‐lead filer with three other institutional investors who put together a shareholder proposal for Alphabet’s 2020 AGM. The proposal called for the establishment of a Human Rights Risk Oversight Committee to help anticipate and oversee management of the adverse human rights, and societal risks and impacts, associated with Alphabet’s technologies. With approximately 53% of Alphabet’s voting shares controlled by the company’s executive officers and board members, 16% support for the resolution translates to roughly 45% of the independent votes.  

EOS has also carried out public policy and market best practice engagement, including: 

• June 2019: presented at the Institute of Business Ethics on AI ethics 
• July 2019: participated in a meeting with the OECD and Thun Group on AI and human rights in Geneva 
• Sept 2019: HSBC invitation to speak on AI ethics. HSBC published its principles in February 2020. 
• Feb 2020: PRI London Forum 
• April 2020: contributed to ShareAction publication on responsible AI 
• May 2020: UK All‐Party Parliamentary evidence session on corporate decision‐making and investment using AI 

Subsequent to this project being reviewed as part of the PRI Awards 2020:  

• August 2020: Moody’s seminar on AI and fintech 
• September 2020: All-Parliamentary Group Corporate Responsibility Meeting on Responsible approach to AI. This discussed how data assets and predictive analytics are changing the way society functions. It explored ideas ranging from regulation to taxation, and raised awareness of supply chain issues given the concentration of cloud providers, plus the impact of AI on the most vulnerable segments of population.  

c. The challenges associated with this initiative (e.g. free rider issues hindering first movers) how these were overcome, and what was learned. 

The first challenge was to develop a foundational understanding of the long‐term sustainability issues relating to AI and data governance, and then translate this into an actionable engagement agenda. 

EOS at Federated Hermes tackled this challenge by publishing the two reports mentioned above. It placed particular emphasis on the connection between AI ethics and the impact on human rights ‐ to help foster wider acceptance and understanding by stakeholders. 

Challenges also remain in ranking the leadership of companies with trusted AI applications beyond the technology sector, as brand value is closely connected to product offering, customer services and other factors. 

EOS developed a multi‐criteria appraisal system for scoring technology companies’ approach to AI and ethics, which can be used as an input to ESG integration, as well as monitoring the progress of engagement on this topic at individual companies. 

Subsequent to this project being reviewed as part of the PRI Awards 2020: 

In September 2020, EOS published the trusted AI assessment framework. This is a ‘three lines of defence’ model that it recommends to companies for trusted AI implementation. Each category of the assessment is mapped to the principles and analytical framework (legal and financial factor analysis, and salient social impact analysis) that EOS highlighted in the April 2019 paper. 

Outline the results, including evaluation of its success against the objectives; were there any adjustments to the forward agenda; were there any insights learned from this project that can be applied more broadly? 

Subsequent to engagement by EOS: 

• Alphabet has made improvements in AI governance at the operational and product levels. (Although EOS continues to press for improvements at the oversight level.) 
• Facebook has established a safety advisory board. 
• Chinese insurance firm, Ping An, became one of the first major financial institutions globally to publish a set of AI ethical principles. 

In addition to these successes, a European conglomerate sought EOS’s guidance on its AI ethics approach, and the board chairs of two other European financial conglomerates responded to EOS’s letters, detailing the progress made within their companies on AI ethics and data governance. 

EOS has also expanded its engagement targets to go beyond financial services and cover healthcare and consumer companies, with positive feedback and insights gained from those engagements.