Ensuring AI use is ‘ethical’ has a host of benefits for any private organisation, including protecting brand reputation, boosting revenue, and preparing for future regulation . However, without any official regulation governing AI systems in the UK as of yet, the path to defining and implementing ethical AI appears obscured by many unknowns. Such uncertainty may disincentivise companies from engaging with AI, but it is increasingly recognised as a necessity to stay relevant, optimise, and compete – or risk falling behind without it.
For the organisations who have already implemented AI technologies, or have imminent plans to do so, research within the field has accelerated to match its growth and provide sought-after insight. Advisory organisations such as Gartner and The Alan Turing Institute have published their recommendations for principles around which to base a governance framework, which have been supported and echoed by businesses driving ethical AI use such as AstraZeneca . The policies by which many organisations will decide to govern their AI use are centred around 5 guiding principles: Fair, Accountable, Private and Secure, Explainable and Transparent, and Human-Centric and Socially Beneficial.
These principles inextricably include notions of ambiguity and attempt to define matters of personal opinion and experience. To help unpack them, Kubrick Data Management consultants Oliver East and Max Herinck share their insight into the practical steps you can take to ensure these guiding principles deliver results.
Diversity of thought is key to action this sentiment. Utilise democracy, whether at board-level, with focus groups, or in committees, to give guidance for best practices which considers all stakeholders and social groups an AI system can impact in order to mitigate against biases within both the system and the data with which it functions.
AI tools are not moral decision-makers, but algorithms which are created and approved by humans with accountability. Determining a system of checks and owners removes any sense of sci-fi radicality and places the onus on those who create, monitor, and can intervene. Consider the impact of the possible outcomes of the AI system to determine the stages at which human intervention and approval is required throughout its process.
Private and Secure:
As with any development in the digital revolution which utilises data, privacy is a risk to be weighed up and mitigated against. The organisations who are willing to invest in AI should simultaneously build their data management and governance capabilities to ensure their data use is of the highest quality and GDPR compliant in order to protect their AI systems from the source.
Explainable and Transparent:
Organisations can protect their reputation and build trust by demonstrating the purpose of their AI system and its limitations and risks. To avoid masking interactions that consumers have with AI systems, businesses may be advised to avoid assigning human qualities, such as names, to systems like chatbots. By working with relevant stakeholders, explanations of an AI system and its output can be tailored to its audience. For example, the results and process of an AI diagnostics tool should have different explanations for patients and their doctors.
Human-Centric and Socially Beneficial:
The outputs of AI are for the advancement of our society, whether economically or socially; we must consider this first and foremost. The impact assessments which determine human accountability and intervention in the early stages of AI implementation can also be used to map the social benefits of an AI system and highlight the biases which can undermine its fairness.
Oliver and Max reflect on the importance of ethical AI for every industry and the difference between AI regulation and principles of ethics:
After spending a year working on a company-wide ethical AI initiative, I have seen first-hand the importance of embedding responsible AI practices. Although the field is in its relative infancy, it will only grow in significance as awareness of the myriad of ways it affects our lives increases. Engaging stakeholders across the business, starting the conversation around ethical AI, and providing best practice and guidelines can ensure ethical considerations are embedded in the development and deployment of AI systemsOliver East, Data Management consultant
Unlike prescriptive regulation, principles are inherently high level and flexible. Companies need to recognise and embrace this fact in order to reap the benefits of principles-based governance while avoiding the pitfall of merely ‘ethics-washing’ projects by subscribing to principles with no basis in reality.Max Herinck, Data Management consultant
To learn more about Kubrick and how we can support your AI, data, or technology transformation, get in touch: email@example.com