The Federal Trade Commission (FTC) has released a set of new AI guidelines on the use of artificial intelligence technologies and algorithms in autonomous decision-making. The guidelines addressed five general principles to ensure the proper use and protection of the consumer. The guidelines were contained in a blog post written by Andrew Smith, the director of the FTC Bureau of Consumer Protection. The new guidelines applied various studies, reports, legislations, and other sources that have been of recent interest in the regulation and management of AI and algorithms.
Five principles of the AI guidelines issued by the FTC
Transparency when applying automated tools was emphasized by the FTC AI guidelines. The FTC directs entities using AI to be open when collecting sensitive data to feed to AI models or be liable for possible FTC action. They should also consider notifying users when they are making automated decisions from information obtained from a third-party vendor. This requirement applies to a third-party vendor, who is considered as a consumer reporting agency under the Fair Credit Reporting Act (FCRA). An adverse action notice should be included in situations, for example, when denying somebody a loan or an apartment. The adverse action notice should inform customers of their right to view and possibly contest the information used to arrive at a decision when such information has been used to make an important decision.
Explaining the decision to the customer is also required by the new FTC AI guidelines. When businesses make decisions about consumers based on algorithms and data, they should explain to the customer what data was used and how the decision was arrived at. Businesses that calculate the scores of their customers should disclose the factors that influenced the score and their relative importance to other determinant factors. The businesses must also inform the customer when automated tools have the possibility of altering a deal based on the customer’s behavior.
The FTC also requires fairness in decision-making. The new AI guidelines require that decisions arising from the use of AI should not discriminate against protected classes of people. In case of a dispute, the FTC will analyze the input and the output to make a conclusion about whether anti-discrimination laws have been violated. Analyzing the input will allow the FTC to discover whether there are ethnic factors in the model that caused the discrimination. Similarly, analyzing the input will help the FTC understand whether a neutral model resulted in discrimination.
Another requirement demands that entities that use AI to make decisions to ensure their models are robust and empirically sound. This requirement will be legally binding for entities that operate AI models. The FTC will require consumer reporting agencies to ensure they implement proper measures to improve the accuracy of the information. Non-consumer reporting agencies will be considered as “furnishers” and be required to implement written policies that ensure the information is accurate and has integrity, according to the new FTC AI guidelines. This is because they provide data to the consumer reporting agencies, and providing inaccurate data would lead to errors upstream. Additionally, the FTC requires entities that apply AI tools in decision-making to validate their models and ensure they do not discriminate against a class of people.
Lastly, the FTC requires entities hold themselves accountable for compliance, ethics, fairness, and nondiscrimination. The FTC advises organizations that apply AI to answer the following key questions when deciding whether an algorithm could cause harm to consumers.
How representative is the data set?
Does the data model account for biases?
How accurate are the predictions based on big data?
Does a company’s reliance on big data raise ethical or fairness concerns?
Setting concrete steps for AI regulation
Compared to other AI guidelines, the FTC directives are more focused on addressing real use scenarios of AI and algorithms. The past guidelines by other agencies and advisory groups have been too broad and without specific applicable use cases. Contrarily, the FTC AI guidelines set the early concrete steps to definitively regulate AI technology.