IBM recently announced that it would stop developing facial recognition technology, out of fear that it could lead to bias, and does not support its sale to law enforcement agencies.
The past few years have seen tremendous investment in facial recognition applications. AI has proven to be quite adept at processing images to determine all manner of things, including identify verification (e.g., finding criminals in security cameras, unlocking your phone), disease identification, and more. Some researchers claim facial recognition can even predict personality factors and sexual orientation, although these claims have not been widely replicated.
There are undoubtedly beneficial uses of this technology – reuniting kidnapped children with their families, for example. On the whole, however, many uses of facial recognition are unproven, may result in significant bias, and encourage the widespread storage and use of our personal data in ways we may not realize or agree to. IBMs decision to withdraw from this market is responsible and appropriate.
In hiring, Modern Hire has long taken the stand that it is inappropriate to use facial recognition technology in hiring decisions. Further, we believe that the only information that should be used to make hiring decisions is that which is expressly provided by the candidate. This includes responses to interview questions, test responses, resumes, and all the other voluntarily provided information, making up the typical hiring processes. It does not include using AI to predict behavior based on how a person looks or talks or analyzing their social media activity. This type of data is likely to contain bias, is difficult to interpret, is invasive, is not job-relevant, and to boot, has not been shown by rigorous peer-reviewed research to add predictive value to hiring decisions.
When considering AI applications for use in screening and selecting candidates, one always has to start with the criteria of job relevancy and ‘face validity.’ Job relevancy involves pointing back to the job itself and showing that what you measure links to the pieces of knowledge, skills, and abilities required to succeed. For face validity, you have to ask whether the evaluation seems related to the job and fair to candidates experiencing it. Facial recognition does not meet either of these requirements.
In this era of ever more powerful AI, it is essential to remember that all AI should serve to improve the human condition. At Modern Hire, we research and strongly support AI, but only to the extent that it is scientifically validated, reliable, and fair.
For more information on Modern Hire’s Ethical AI Position click here.