AI has a strong business potential, but you have to take into account the risks and how to minimise them before implementing them.
There is a well underway battle in artificial intelligence. In recent years, the number of AI companies has risen by an incredible 270%, and even businesses that have still not taken the leap are still thinking about this.
However, when you are a CIO or business leader who wants to use the AI-whether you create your own internal technology or license it from a client-there are serious implications that you need to take into account.
The chance for prejudice is the most important thing to look at. Regrettably, several cases of AI biased against minority groups have been seen. This is not only unethical; for the client, too. If AI can’t work as it should for all, it doesn’t have a great advantage first of all.
There are three main questions to ask if you want to make AI part of your business strategy:
- Demand information.
AI systems are equipped and evaluated with vast quantities of data using machine learning. Such data should be diverse and inclusive of the various people and should use cases that it affects–it does not work properly otherwise. Begin by questioning where and how the data is collected and think critically of areas of data deficiency.
Even if you have specific, representative knowledge, bias can still occur when you do not have an attentive AI model training protocol. You want to ensure the data are matched with demographic information (for example gender, age, and ethnicity) and also with the aspect when training an AI algorithm: is a person wearing glasses, a hijab, or a face mask? It is essential that the algorithm is trained with significant data on every subset.
This analysis will continue, since you also support AI. Many times, people report the single accuracy score–e.g. “My AI is able to recognize people about a percentage of time” But you have to break it down and assess results on the basis of how well an AI works for different subgroups or populations–such as “AI works for men, some percentage of the time, but only so much for women.
- Please ask about the AI team.
Diverse teams are minimizing the bias— we construct what we learn, after all. Even with good intentions, a group of people who create algorithms may inadvertently introduce bias if they come from similar demography and backgrounds. It is only when we have diverse teams that we can say: “You know, I noticed that there are not enough details about those like me. Can we make sure that we include this?” Our Cairo data labelling team noted that at the time we had no data on hijab-carrying women, which was a big monitoring. We have therefore decided to add this to our data set.
Various teams can also consider new applications of technology inclusive of different groups and address problems for different groups of people. This is not only right, it is good for the business and is important for pushing the industry forward.
- Inquire how AI is going to be implemented.
It’s not just about developing reliable systems to solve the problem of AI bias. Equally important is how it is used. You must ensure that AI does not lead to prejudice or unintended effects in the real world.
For instance, take law enforcement. AI is developed by businesses to forecast the probability of a convict committing another crime in a criminal case and monitor sentences. However, reports show that this technology still disadvantages minority groups, with detrimental outcomes. Such cases of use should be avoided until the industry can ensure that AI systems are reliable and representative and implemented to prevent any partiality.
The bottom line: don’t wait for a bias question.
Bias defence cannot be a one-time thing. If your company uses AI, you have to reassess the protocols constantly and ask tough questions to make sure that it’s correct.
You can use money if you want to join AI firms, but you are worried about the risks. For example, the AI Collaboration brings together a number of international voices to research and develop best practices for AI technology.
The AI race just accelerates. Now, we will keep pace with our approach to risk management.