Bias and Harm Mitigation is the fourth principle requiring those responsible for AI-enabled systems to proactively address the risks of unintended biases or harms, whether during initial rollout, as systems evolve and learn, or when redeployed in new contexts.
This principle is essential for ensuring fairness, maintaining public trust, and safeguarding the ethical and lawful use of AI in Defence. It emphasises designing and operating AI systems to minimise unintended biases and prevent harm to individuals, groups, or communities. Mitigating these risks is both complex and vital, demanding that developers and decision-makers anticipate challenges throughout the system’s lifecycle while balancing operational effectiveness with fairness and inclusivity (see Human Centricity).
Definition of Bias and Harm Mitigation:
Those responsible for AI-enabled systems must proactively mitigate the risk of unexpected or unintended biases or harms resulting from these systems, whether through their original rollout, or as they learn, change or are redeployed.
AI-enabled systems offer significant benefits for Defence. However, the use of AI-enabled systems may also cause harms (beyond those already accepted under existing ethical and legal frameworks) to those using them or affected by their deployment. These may range from harms caused by a lack of suitable privacy for personal data, to unintended military harms due to system unpredictability. Such harms may change over time as systems learn and evolve, or as they are deployed beyond their original setting. Of particular concern is the risk of discriminatory outcomes resulting from algorithmic bias or skewed data sets. Defence must ensure that its AI-enabled systems do not result in unfair bias or discrimination, in line with the MOD’s ongoing strategies for diversity and inclusion.
A principle of bias and harm mitigation requires the assessment and, wherever possible, the mitigation of these biases or harms. This includes addressing bias in algorithmic decision-making, carefully curating and managing datasets, setting safeguards and performance thresholds throughout the system lifecycle, managing environmental effects, and applying strict development criteria for new systems, or existing systems being applied to a new context.
1. What is a bias?
2. What is a harm?
1. A bias is a systematic inclination, preference, or prejudice for or against something, someone, or a group, often in a way that is unfair or unbalanced. It can manifest in decision-making, data processing, or judgement, influencing outcomes by introducing distortions or inaccuracies. Bias can be intentional or unintentional and may arise from personal beliefs, societal norms, flawed processes, or incomplete information. In the context of artificial intelligence, bias refers to errors or disparities in data, algorithms, or system outputs that lead to unfair treatment or outcomes for certain individuals or groups. Addressing bias is essential to promote fairness, objectivity, and reliability.
(See card: What do we mean by "bias"? How can I address bias in algorithmic decision-making?) 2. Harm refers to any damage, injury, or negative impact caused to an individual, group, system, or environment. It can take various forms, including physical, emotional, psychological, financial, reputational, or societal. Harm may arise intentionally or unintentionally and can result from direct actions, neglect, or systemic issues. In the context of artificial intelligence and technology, harm often refers to adverse outcomes caused by errors, biases, misuse, or unintended consequences of systems or decisions. Addressing and mitigating harm is crucial to ensure ethical practices, safety, and trust in systems and interactions.
(See card: What do we mean by "harms" and what are they?)
Bias and harm can only be mitigated if they are recognised and then understood. That means there is a clear link with Understanding and the testing and assessment that is central to Reliability.
8
8
8
8