March 28, 2023

AI Ethics - Getting Out In Front

How can you be sure the AI technologies you are using are ethical? Karl Schulenburg outlines three key guidelines to help you protect yourself and your stakeholders.

Ethical issues will always emerge that you hadn’t predicted or prepared for that can throw your AI project off track, or kill it altogether. It’s best to get out in front of them as much as you can. It's also a good example of how best practice is to bake your ethical guidelines in from the outset. 

CAVEAT: Not everyone is responsible for all the AI solutions they are using and in the case you are using a 3rd party data model then while you don’t have direct access to that data model, you owe it to your stakeholders to interrogate the model in some way. For instance you may be able to review the kind of data sources the app uses or the algorithms it employs, and the historical accuracy of its decision-making. Or if this kind of access isn’t possible, you may need to run experiments to query the outputs of the app to check its returning equal and unbiased results for everyone using it.

So, what guidelines should you use to protect yourself and your stakeholders from unethical AI? I asked Chat GPT and the answer I got back is a great place to start I think:

1:  Bias

AI algorithms can reflect the biases inherent in the data they are trained on, leading to discrimination against certain groups. For example, if a facial recognition algorithm is trained on data sets consisting mostly of white faces, it may fail to recognize faces of other races accurately. Similarly, if an algorithm is trained on data sets that reflect gender, racial or other social biases, it may result in unfair outcomes.

Best Practices:

  • Diversify Data Sources: Accessing data from a variety of sources can help to reduce bias in AI algorithms or finding ways to include balanced information of demographics.
  • Regular Auditing: Regular audits and testing of the AI algorithm can help identify potential biases and suggest improvements.
2: Fairness

Fairness in AI development refers to ensuring that AI systems do not systematically favour one group over another. AI systems should operate in a way that is unbiased and equitable to all individuals and groups. Ensuring fairness in AI development is paramount to prevent discrimination and injustice.

Best Practices:

  • Diversity of Development Team: Ensuring that the development team is diverse can help to reduce inherent biases.
  • Algorithmic Fairness: Promoting the use of algorithms that are designed explicitly to minimize unfairness can help prevent the deployment of unfair AI systems.
3: Transparency

AI systems must be transparent to ensure that their decision-making processes are understandable and explainable. When AI systems operate transparently, individuals and organisations can verify the decisions made by the system.

Best Practices:

  • Open Source: Using open-source models can help ensure transparency in the development process.
  •  Easy to Understand: The output of the AI algorithm should be easy to understand by the users.

A subsequent prompt added “Human Oversight” to the list as a great reminder that whatever the degree of “artificial” intelligence you employ in your project, you can never abdicate your own responsibility and should never step back from applying your own intelligence. Chat GPT went on:

“​​Adopting an ethical approach to generative AI technology means ensuring human oversight in its development and functioning. Although the technology can be very autonomous, you must always have a human in the loop and actively monitoring the technology's performance. This provides an opportunity to address potential issues and adjust the technology to comply with ethical guidelines.”

Once begun, your project needs constant ethical oversight - as our robotic friend has reminded us. So I urge everyone to get out in front of this potentially hazardous pitfall before you even begin, but even then that is no time to sit back and relax. You must remain vigilant throughout. This is the best way to protect yourself from unforeseen barriers to success and to ensure that you achieve the outcomes you want.