'Hold yourself accountable' on keeping AI unbiased, or the FTC might do it for you
The U.S. Federal Trade Commission emphasized the importance of using artificial intelligence in an equitable manner in a blog post earlier this month.
As FTC attorney Elisa Jillson noted in the post, so-called neutral technology can produce troubling outcomes, including racial discrimination or the replication of other biases.
"As your company launches into the new world of artificial intelligence, keep your practices grounded in established FTC consumer protection principles," Jillson advised.
WHY IT MATTERS
AI has the potential to advance innovation in the healthcare industry. But, as Jillson (and others) points out, tools are only as unbiased as the data they're trained on – and too often, that data represents the bias of its creators.
In her blog post, Jillson draws attention to three laws important to AI developers and users: Section 5 of the FTC Act, the Fair Credit Reporting Act and the Equal Credit Opportunity Act.
Jillson also offers a number of recommendations to try to ensure AI is not replicating inequity:
- Start with the right foundation, including improving data, thinking about ways to account for gaps, and limiting where or how the model is used.
- Watch out for discriminatory outcomes, making sure to test your algorithm before using it and periodically after that.
- Embrace transparency, such as by conducting and publishing the results of independent audits or by opening your data to outside inspection.
- Don't exaggerate your algorithm – under the FTC Act, statements to business customers and consumers must be truthful, non-deceptive and evidence-based.
- Be truthful about data use, and about whether users can control how their information is shared.
- Do more good than harm. If the injuries it causes are not reasonably avoidable or outweighed by benefits, the FTC can challenge that model as unfair.
Jillson encourages companies that rely on AI to take these recommendations to heart.
"As we’ve noted, it’s important to hold yourself accountable for your algorithm’s performance. Our recommendations for transparency and independence can help you do just that," she wrote.
"But keep in mind that if you don’t hold yourself accountable, the FTC may do it for you," she warned.
THE LARGER TREND
AI discrimination can have tangible – and sometimes deadly – effects on already vulnerable people.
This past August, an article in the Journal of the American Medical Informatics Association found that AI bias may worsen COVID-19 health disparities for people of color. Researchers flagged the danger of regarding AI as intrinsically objective, especially when building models for optimal allocation of resources.
As former FCC Commissioner Mignon Clyburn told Healthcare IT News this February, tools aimed at addressing inequity must address foundational issues in the health industry.
"Some of your [AI] verifications don't see my face as an African American woman, but it's supposed to provide an on-ramp? I think not," Clyburn said.
ON THE RECORD
"While the sophisticated technology may be new, the FTC’s attention to automated decision making is not," wrote Jillson.