A few years ago, I posited that Social Security Administration (SSA) may someday implement automated decisions. Today, my observation of recent trends in artificial intelligence (AI) confirms that we are moving closer to that day.
SSA has long used data analytics and computer systems to improve its ability to process large volumes of claims efficiently. Data analysis of early court decisions on the use of vocational expert evidence led the agency to develop medical-vocational guidelines (also known as “Grid Rules” to direct conclusions based on factors of a claimant’s age, education, past work experience, and residual functional capacity. SSA has also implemented several electronic case management systems (eCMS) to document case development and organize medical evidence. Since 2018, adjudicative staff at the hearing levels now use an array of decision support tools known as Insight to identify inconsistent decisions and generate templates for written decisions. Notwithstanding, substantive findings and rationale must be manually entered, and the final decision reviewed and signed by an administrative law judge.
While some theorize that AI is most useful for low level tasks with minimal discretion, the rising prominence and rapid innovation of technology likely points to a more prominent role for AI in complex adjudication. In fact, an executive order in 2020 promotes and accelerates the adoption of trustworthy AI in federal agency services.
As with any new technology, accelerated adoption of AI will also be met with a healthy dose of skepticism. Alarmists and popular science fiction writers have long warned of the potential dangers of AI in government – from programming bias and error to depersonalization of due process and a loss of human empathy. Such risks are inherent in any technological advancement and must be checked and mitigated by careful human review.
It is clear that AI is on the rise in federal government and must be used to advance, not undermine, due process.
 Exec. Order No. 13960, 85 Fed. Reg. 78939 (Dec 8, 2020)