AI blindspots

How can you take social prejudices and structural inequalities into account before, during, and after the development of your data or AI application? 

When developing a new AI application, the focus is on improvement: your AI application is only successful when it provides a (better) solution to the problem you want to address. In doing so, (unforeseen) negative consequences of your AI application are sometimes overlooked. The 'AI Blindspots' card set helps you identify ‘AI Blindspots’ - ethical pitfalls - by reflecting on the decisions and actions you take in the development process of your data or AI application. This allows you to improve your data or AI application while addressing potential negative side effects of your application.

The card set can help you and your team to...

• create a culture of ethical and social awareness;
• identify challenges before they cause damage;
• improve the overall quality and acceptance of your product;
• communicate with your stakeholders about the ethical, social, and legal impact of your product and how you deal with it as a company.

The card set consists of 2 types of cards:

• The AI Blindspots cards are divided into three phases, namely the planning phase, the development phase and the implementation phase. Each card presents a blindspot and includes:
  • a set of questions that allow you to reflect on possible ‘blindspots’;
  • a ‘use case’ to demonstrate the importance of that ‘blindspot’;
  • some tools and tips that can help you detect/reduce the ‘blindspot’.
• The Ethical Dilemma cards are examples of situations where an ethical problem arises.

Below are six methods for using the AI Blindspots card set in your organisation.

To raise awareness, as a conversation starter:

• A conversation starter for you and your company to discuss the ethical, legal, and social aspects of data and AI applications.

To raise awareness, in a workshop:

• A workshop with your team in which you focus on one ‘Ethical Dilemma’ card of your choice and try to come up with solutions to it. Afterwards, you pull out the AI Blindspots cards for inspiration and see if you can come up with other solutions or make one of the solutions already thought up more concrete. > template available (see ‘Downloads’)
• A workshop with your team where the group is divided into 3 smaller groups. Again, you focus on one ‘Ethical Dilemma’ card of your choice. But each group looks at a different phase of the AI Blindspots cards (planning, development, implementation) and tries to come up with solutions to the dilemma using their cards. > template available (see ‘Downloads’)
• A workshop with your team in which you focus on your own identified ethical dilemma - write this down on the ‘Ethical Dilemma’ joker card. Then, in the group, devise one or more solutions to this dilemma. You can use the AI Blindspots cards for inspiration.
• A workshop with your team in which you come up with requirements for your data or AI application using a reverse brainstorm. A reverse brainstorm does not solve the problem, but actually makes it bigger. Use the AI Blindspots cards and ask yourself, ‘How can we ensure that this blindspot becomes a reality for our data or AI application?’. Now turn these (negative) unethical measures into (positive) ethical requirements for your data or AI application. > template available (see ‘Downloads’)

For evaluation:

• Use the AI Blindspots maps as a monitoring tool when planning, developing, or implementing your data or AI application. Have you taken these blindspots into account? If not, discuss with your team how you will deal with the blindspots.

This card set is an adaptation of the tool  'AI Blindspot' developed by Ania Caldeeron, Dan Taber, Hong Qu, and Jeff Wen from the Berkman Klein Center, een MIT Media Labs's 2019 Assembly Program. 

The AI Blindspots card set is available under a CC BY 4.0 Licence.