The Responsible-AI Checklist for Sustainability Teams (Ethics, Risks & Governance)

Nowadays, Artificial Intelligence (AI) is everywhere, from recommending the next Netflix show to watch to shaping how we fight climate change. As AI rapidly expands into all areas of business and society, concerns are increasing: Can AI benefit or harm the environment and society? How can we have responsible AI used transparently? Who will be held accountable if data privacy is compromised, and how will it be maintained?

These challenges are both technological and societal, and they can directly impact sustainability goals. Businesses must develop and implement responsible AI with ethical practices in mind. For this, sustainability professionals have a dual responsibility. First, they should leverage AI for ESG to keep up with current trends. Second, they must ensure that AI adoption across the entire organization aligns with sustainability standards and stakeholders’ expectations. 

However, there is a plethora of methods and tools available, alongside legally binding regulations, to help put ethical principles into practice. For example, sustainability teams can integrate algorithmic impact assessments into their reports, support IT teams in implementing understandable AI techniques (i.e., LIME), and ensure that organizational practices adhere to laws like the EU AI Act. The Sustainability Intelligence course provides strategic insights into how responsible AI can accelerate sustainability performance by improving risk assessment, data-driven decision-making, and operational efficiency.

The scope of this article is to be a starting point for the readers to understand AI ethics and the principles that they want to establish. To assist them in this journey of integrating responsible AI with their sustainability objectives, we’ll go over the fundamental pillars.

Three Major Risks of AI

The first step in establishing ethical business practices is proactively identifying both the opportunities and the risks of AI. These risks can directly undermine stakeholders’ trust and erode the company’s values. 

1. One of the most pressing matters that responsible AI dwells on is algorithmic bias and discrimination. The general notion is that machines are inherently objective; however, this couldn’t be further from the truth. AI reflects the data it is trained on. If incomplete or unrepresentative datasets are used, then certain groups could be at a disadvantage. For example, from 2014 to 2017 Amazon used an AI system to screen CVs. The dataset consisted of resumes submitted to the company over the last ten years, which were mostly composed of men. The AI started filtering out women’s CVs, ranking all-female universities or clubs lower than the all-male ones. 

• The absence of transparency and explainability is another critical risk. Stakeholders can’t trust a system they do not understand. To earn their trust, companies should disclose to the end user when an AI model is used in decision-making. Moreover, the methods, metadata, and expertise used to train the AI model should be accessible. AI shouldn’t be treated as a black box that we can over-rely on. It needs to be able to explain the results before the lack of clarity damages the company’s reputation. For instance, if the company can’t explain the calculations in its GHG emissions reports, then stakeholders may view the results as unreliable and even accuse the company of greenwashing.

• Another important issue is data privacy. Sensitive data in datasets that are not adequately protected could leak. Businesses must reassure stakeholders that their information is safe and that strong security protocols are in place to prevent any hacking attempts.  

Building An Ethical AI

To develop a responsible AI model, companies must pay attention not only to compliance but also keep their core values in mind. The AI systems should be designed with fairness, transparency, and accountability as their primary pillars. To achieve this, businesses should first decide on their principles that will guide the AI development.

1. A fundamental principle that companies could embrace is ensuring that AI is human-centric. It should augment human judgment and not replace it. Decisions should be made by humans alone who can take responsibility for their actions. For example, an AI system that accepts or rejects clients’ requests for loans could lead to discrimination. A human should review the decisions and should apply clear, fair standards when reviewing them.

• Moreover, the teams that develop the responsible AI models must be diverse and multidisciplinary. In order to make the model more applicable to a wide range of stakeholders and not to reproduce biases, experts in data science, sustainability, ethics, and sociology could assist in designing the system and identifying any blind spots.

• Finally, the data used to train these models must be fully documented, with a clear understanding of its source, potential biases, and the rights to use it, ensuring the AI’s foundation is solid.

While the above steps can offer a strong internal practice, companies should also look for international frameworks, which provide insight into trustworthy AI.

OECD AI Principles

The OECD AI Principles are an international framework that is widely accepted as a basis for responsible AI. Adopted in May 2019, these principles were the first of their kind to be endorsed by governments around the world. Five value-based guidelines for stakeholders have been established by the OECD. 

The first principle is inclusive growth, sustainable development, and well-being. AI should benefit society as a whole, not just a chosen few. 

The second principle is respect the human-centered values and fairness. This entails respecting rights and refraining from discrimination. Supplier-screening systems, for instance, ought to identify labor risks without unjustly penalizing small enterprises that are unable to provide comprehensive reports.

The third principle is transparency and explainability. As discussed above, when AI is used, businesses should be transparent about it. As well as be open about the data sources and assumptions they use.

The fourth principle is Robustness, security, and safety. AI systems ought to be secure, dependable, and tested. For example, climate risk models need to be verified to ensure that they don’t mislead investors about their vulnerability to droughts or floods.

Lastly, the fifth principle is accountability. Businesses are still accountable for the results of AI. Instead of placing the blame on the model, the company must address and clarify any errors reported.

Everything discussed so far in this article, risks and ethical practices, connects with the OECD framework. This shows that responsible AI has shared foundations: fairness, transparency, security, and accountability.

Conclusion

Responsible AI can accelerate processes, but only if it is developed under fairness, transparency, and accountability. Businesses can ensure that AI helps rather than harms their long-term ESG goals by recognizing the risks, promoting ethical behavior, and following OECD AI principles, which are a valuable benchmark. To talk in “successful terms,” such goals can only be accomplished if each company is committed to practically embedding these ideas into their operations and not giving in to compromises. If you are looking for a useful road map with case studies and resources to investigate these practices, the sustainability intelligence course has been created for professionals who wish not only to learn but also to take the lead.