Accountability as a deal accelerator: what buyers and sellers need to know about bias and responsible AI

AI algorithms are often opaque 'black boxes'. Assessing accountability for decision-making processes and outcomes is necessary to protect deal value, but it is not straightforward. How can tech buyers and sellers ensure concerns around AI transparency, explainability, bias and discrimination don’t derail deals?

The potential for bias and discrimination

The explosion of generative AI brings a sharper focus on responsible AI. As AI systems become more intricate and pervasive, technology businesses face greater scrutiny regarding transparency, accountability and bias mitigation.

Whilst regulators primarily rely on existing antidiscrimination laws to tackle AI bias, they are starting to demand greater accountability. Authorities in the EU and US are targeting high-risk uses of AI and requiring developers to publish impact assessments and reports, such as the Algorithmic Accountability Act (2019) in the US.

The ability to interpret and explain AI decisions is crucial to fostering users' trust and preparing for future legislation. In a recent YouGov survey, nearly two-thirds of UK consumers want businesses and brands to be transparent about their use of AI. However, a lack of transparency and explainability in algorithms brings growing concern about the potential for bias and discrimination, particularly in hiring, lending and criminal justice.

Nascent regulation and limited oversight in the development of AI are also problematic. Compared to other technologies, new AI algorithms and applications are not subject to rigorous testing before launch under current regulatory frameworks, which increases the risk of unintended consequences or harm if algorithms are biased or faulty.

Acquiring a liability

A lack of transparency raises questions about accountability. Who should be responsible if an AI system makes a biased, discriminatory or criminal decision? Should it be the company that created the algorithm, the data scientists who trained it, or the end user who implemented it? Regulators are grappling with these questions and looking at ways to address accountability.

The EU’s position is clear. They are prepared to pursue any company that uses AI to break the law even when no human intervention has occurred. In the US, the Federal Trade Commission has stated that companies must be accountable for their AI systems and take steps to prevent bias and discrimination.

Buyers of AI assets must understand how algorithms make decisions and any potential for bias and discrimination. The EU’s AI Act demands that companies understand the coded algorithmic models and data that make up their AI systems. In the case of a dispute or compliance query, can the seller provide understandable and justifiable reasons for AI actions and outcomes? Buyers could inadvertently acquire a significant liability without demanding AI transparency and explainability.

Further regulation and guidelines for the ethical and transparent development and deployment of AI systems are on the horizon. The EU introduced the AI Act for high-risk applications, and other countries are following suit. The UK is shaping its strategy via bodies like the Centre for Data Ethics and Innovation, and the US is focused on sector-specific guidelines. All three approaches demonstrate a move towards more responsible and accountable AI.”

Jonny Bethell, Partner, Taylor Wessing

Protecting deal value

AI explainability is crucial in M&A deals. A comprehensive understanding of AI systems is vital to avoid integration challenges, hidden biases or decisions that could lead to regulatory fines or reputational damage.

So how can buyers and sellers be confident that AI decision-making processes and outcomes can withstand ethical scrutiny and the gaze of current and future legislation? How can they ensure an AI asset doesn’t become a liability?

Strategic buyers must conduct enhanced due diligence on AI systems. This includes understanding how the system makes decisions, what data it uses as inputs and how it handles sensitive information. Tech sellers must also be transparent about their AI systems and clearly explain how they work and any potential biases or risks leading to discriminatory outcomes.”

Jonny Bethell, Partner, Taylor Wessing

Responsible AI diligence: a checklist for buyers

Horizon scanning: Understand how changes in the law could impact value creation and the scalability of potential M&A AI targets in key markets.
Technical audit: Examine the external outcomes of the AI decisions and the system's inner workings. Can the seller clarify the methodologies and data sets used for training and ongoing algorithmic improvement?
Bias and fairness mechanisms: Assess whether the AI systems exhibit unfair biases or whether the merged entity or tech stack will create new biases. What safeguards are in place to prevent bias in AI decisions? Can the seller prove that their system actively works to maintain fairness across all demographics and variables?
Continuous reinforcement: Consider the seller’s plan for continuous improvement and accountability. Will the system be monitored post-transaction to ensure it meets the agreed standards? What mechanisms are in place for updates that may affect legal compliance and ethical practice?

Prioritising transparency: a checklist for sellers

Governance and ethical codes of conduct: Put in place clear guidelines on how AI technology is developed and used and proper oversight to ensure compliance.
Documentation and reporting: Exhaustive documentation is essential. Buyers need reports, assessments and audits covering all aspects of legal, financial and ethical AI operations.
Training: Demonstrate that training algorithms and underlying data are balanced and fair.
Explainability: Work on making your AI more communicative. Buyers want to understand the logic and decision pathways of the AI before committing to an acquisition.
Commitment to continuous improvement: Show a commitment to continually refining and updating AI processes. Have a clear algorithmic improvement plan, and be ready to articulate it during negotiations and due diligence. Participating in industry standardisation efforts can help demonstrate a commitment to responsible AI practices.