1. Include multiple stakeholders in your AI programming decision-making.  

Begin by pulling together a small committee – with different functions and points of view.  The group should include potential users, customers, your development team, and management.  Make sure you bring in someone impartial concerning your company’s product and services.  A diverse group of individuals will pay dividends toward making informed, critical decisions.  

2. Clearly define

Clearly define intended uses of your AI software – rank and determine how important each is and how they benefit and potentially harm the people using it.  Start with a proper strategy and develop use/abuse cases and failure mode analysis.   

For example, facial recognition technology may reduce crime rates, but it also infringes on individuals’ freedoms, sometimes falsely profiling and harming people of color.  As a goal, determine what standards you will measure because you don’t want to have a partial or biased opinion of what matters most.   

3. Don’t deploy your AI product before you have rigorously tested it

This is the idea known in the developer community as Shift Left.  A biased product could potentially harm individuals, so thoroughly vet it before using it.  That means determining what metrics are acceptable and when you have met those conditions.   

Better yet, get an opinion from an unbiased party like Virtual AI https://www.virtuousai.com and make sure you have their approval to implement or whether your product needs additional fine-tuning due to unintended racial bias. 

4. Look holistically at your company’s AI ethics.

That means testing at multiple critical areas – from UX design to your feedback mechanism.  Many companies skip the process but consider the details before implementing a properly working system. 

In your testing, you may find that some minority clients may have a propensity to use your product differently than you might expect.  So, you must have an accurate understanding of what you are doing from a UX perspective, including how you have arranged elements on a page and whether your design has the effect you intend. 

5. Stay current on trends – and follow the news.

You can subscribe to our free newsletter at https://www.virtuousai.com/subscribe-now/ for the latest news and recommendations.  One caution: many popular trends are not the best.  For instance, many language models – including BERT – that have racist biases and can cause unintentional harm to certain minorities.  So, it’s essential to be aware of the landscape – and know if there is updated information about the algorithms you are using and potential ways to minimize the potential for unethical AI.   

Interested in learning more about how to develop ethical AI? Our firm can help you put best practices in place to better serve your customers?  Contact us! Quickly develop ethical AI that is explainable, equitable, and reliable with help from our complete AI IaaS. Sign up for FREE diagnostics.