Employee Training Resource

picture4.png

Noting potential AI bias

Individuals should participate in the assessment of output resulting from the application of AI for detecting hidden bias. They can determine if there is stereotyping or an unbalanced situation. A method for determining these issues is to assess whether their discussion includes both assumptions related to identity and abilities. This will assist in eliminating unfair content before it appears in front of users.

picture5.png

How to Respond to Concerns

Employees are not allowed to use, copy, or share the potentially biased content. Each employee should record the potential problem by capturing both the question being asked and the response produced by the AI system. By doing this, there will be an opportunity to develop accountability within the organization. Thereafter, they need to report their findings to a designated manager, supervisor, or department based upon the organizational policy. Reporting this issue will allow all levels of management to address the situation responsibly and uniformly. Ultimately, reporting these issues will enhance how efficiently an organization can utilize AI technology.

picture6.png

Applying Safeguards to Minimize Harm

One of the most important things to do when working with an artificial intelligence is to take several actions so that the potential for biased responses from the AI is minimized. First, it is very important to regularly update the AI's output in order to provide equal opportunities. Second, ensure the questions to the AI are open-ended; clearly defined; and unbiased. This will allow the AI to respond fairly. Third, review the outputs of the AI with additional perspectives such as those of humans and business policies in order to determine if the outputs are fair. Fourth, test the outputs of the AI for fairness in determining whether the responses would be just as good or bad under a different role or identity.

Ethical AI Use Checklist

Ensure you

• Examine the output of the AI: Check whether the outputs of the AI contain stereotypes or imbalances in the content.
• Assess Assumptions: Evaluate the outputs provided by the AI to the extent of assumptions of who, with what ability, or background.
• Document Concerns: Record concerns regarding the existence of bias, such as the outputs produced by the AI.
• Issue Reports: Issue with the supervisor/manager regarding the possibility of biasness as prescribed by the company policies.
• Use Safeguards: Implement Safeguards, through the application of open-ended questions, updates, and perspectives.