In recent years, the implementation of artificial intelligence (AI) in the administration of the United States' social safety net programs has sparked significant debate. While AI technologies hold the promise of improving efficiency and speed in processing applications for public assistance, their actual utilisation has reportedly created hurdles for many applicants. Instances abound where algorithmic systems, implemented to streamline operations in various states, have instead excluded individuals from vital services.
One of the prominent cases highlighting this issue occurred in Indiana. In 2016, the state moved towards automating eligibility determinations for public assistance programs. The result was a staggering one million applications denied over three years, marking a 54 percent increase in rejections. Among those affected were a young child with cerebral palsy and a woman unable to attend a caseworker appointment due to terminal cancer. These situations underscore the potentially detrimental impact of AI systems when inadequately designed or executed.
The situation in Arkansas further illustrates these challenges. There, a Medicaid assessment algorithm aimed at determining eligibility for home health care resulted in significant cutbacks, reducing hours of care for people with disabilities. Similarly, Michigan adopted an automated system to detect fraud in unemployment insurance, leading to a dramatic increase in fraud accusations—five times more than under the prior system. Astonishingly, this led to about 40,000 false fraud accusations against beneficiaries.
These developments are not new but represent a modern twist on longstanding issues. Historically, the US has set high administrative barriers for accessing social assistance, a trend rooted in the New Deal era. During that period, a distinction arose between "earned" benefits like Social Security and welfare programs, creating differential access often to the detriment of the most needy. AI tools, as documented by scholar Virginia Eubanks, seem to be enhancing these barriers, potentially obstructing access rather than facilitating it.
In response to these challenges, the Biden-Harris administration has initiated efforts to reform the role of AI in public assistance. President Joe Biden directed substantial federal agencies dealing with the social safety net to enhance transparency and human review within algorithmic tools and to address inherent biases. The administration's actions underline a commitment to ensuring AI aids rather than hinders access to services like Medicaid and nutritional aid.
Looking ahead, there lies an opportunity for future administrations to continue these efforts, ensuring AI technologies widen rather than restrict access to benefits. Addressing the "take-up problem," or low utilisation of benefits by eligible individuals possibly deterred by automated systems, is critical. This involves publishing detailed analyses of how AI influences both incorrect benefit denials and non-uptake of available programs.
Furthermore, increasing the involvement of those who rely on these services can shape the deployment and adjustment of AI technology. Establishing consultative councils that include recipients of social safety programs could provide valuable insights into the impacts of these technologies. Agencies, following the current progress in promoting participation, should consider compensating these individuals for their contributions.
Critically, individuals must also have clear avenues to contest AI-influenced determinations. Simple pathways to challenge decisions and connect with legal aid will empower those wrongly affected. The U.S. government's existing equal opportunity offices could serve as a base for establishing these mechanisms, complemented by collaborations with organisations already addressing AI-related injustices.
While the role of the presidency is central, legislative backing from Congress is fundamental to enacting enduring reforms. Ensuring equitable access to public benefits in an AI-driven era will call for statutory changes and adequate resource allocations at all levels of government. These steps represent initial yet significant movements towards securing a more inclusive and effective social safety net.
Source: Noah Wire Services