Using AI to advance automation is a legitimate business practice. However, it’s worth considering where liability lies for incorrect decisions or inappropriate communications. AI models are approximations and do not typically carry liability for any model errors. Employers need to be very deliberate about the role of the model and the role of humans, and have processes in place to rectify any errors that are made.
AI models must also be explicitly tested for biases. Model providers and employers must agree up front what protected categories need to be considered. Bias metrics for protected categories should be recorded and presented with models on initial training and through all future increments of the models.
Response Provided by: Barry Duffy, Solution Architect at FINEOS
