Photo by Dennis Maliepaard on Unsplash
Many network members are already exploring how AI shows up in our advocacy work and intersects with the issues we track, such as school safety, education, and child development. We have all heard much promise, especially from big tech companies, about how AI has the potential to propel positive change. But as many of us are already seeing, AI can also create new barriers for the communities that we care about, and reinforce entrenched power dynamics that unfairly hold people back.
Recently, the Partnership hosted a webinar with child advocates featuring Kevin De Liban, founder and President of TechTonic Justice, to help advocates understand the full scope of how AI is shaping the decisions that govern people’s access to housing, public benefits, child welfare services, and family stability. This blog is a summary of that webinar. You can also read the new report from TechTonic Justice here.
Read the report
Inescapable AI: The Ways AI Decides How Low-Income People Work, Live, Learn, and Survive.
Where is AI making decisions in the lives of people with lower incomes?
The short answer to this question is: Everywhere. Public and private agencies are using AI tools to make decisions around housing, employment, child welfare, domestic violence, language access, public benefits, and more.
As an example, within the realm of Public Benefits, government officials are using AI for:
- Eligibility determination
- Allocation – How much people qualify for
- Asset / Income limit enforcement
- Fraud detection
- Identity verification
- Chatbots
- Making hearing decisions (replacing hearing officers)
Public officials are turning to AI for these functions with various aims. Often, they see AI as a way to hide benefits cuts that might otherwise be unpopular. Other times, they hope that this technology can reduce the workload for under-resourced and under-staffed public agencies.
Even when acting in good faith, those same under-staffed agencies rarely have the capacity to oversee AI through its development, implementation, and use. As anyone who has used AI to plan a vacation can tell you, checking the computer’s work is essential – otherwise you run the risk of showing up at an attraction that closed in 2022. Likewise, in allowing AI to determine whether large numbers of people qualify for health coverage under Medicaid, for example, we run the risk of systematically denying health care to large numbers of people due to a technical glitch, misunderstanding of policy, bad data, or unsupported assumptions.
What happens when algorithms are allowed to make decisions about real people’s lives?
Kevin shared a particularly disastrous example from Michigan. From 2013-2016, the state implemented an automated tool called “MiDAS” to review how many unemployment claimants had committed fraud.
MiDAS found that 40,000 people in Michigan had supposedly committed unemployment fraud. As a result, unemployed Michiganders were assessed thousands of dollars in penalties. And the state enforced those penalties aggressively. Many had wages garnished, some losing their homes, cars, and jobs. People were forced to declare bankruptcy. Some people even engaged in self-harm. Eventually, after years of advocacy, the state reviewed the cases. Michigan’s Auditor General found that 93% of those fraud allegations were false. The State of Michigan had falsely accused around 40,000 people of fraud – devastating many of their lives.
The risk of error is far too high for real people to bear
Kevin shared that rather than making public systems more efficient, effective, or responsive, so far the most common result of using AI in public systems is to cut off people’s access.
AI tools are notoriously riddled with bias and inaccurate information. When public officials adopt AI tools to make decisions about whether a family should qualify for Medicaid or be allowed to live in public housing, it’s not the public official, or the tech company building that AI tool, who bears the risk of error. The true risk is borne by the family that may lose their health care or their housing.
It is worth asking whether the public agencies using AI tools actually understand the criteria that AI is using to make decisions on their behalf. It is nearly impossible to contest a decision when nobody knows what criteria were used to make that decision – not even the agency responsible for it.
The time for action is now
TechTonic Justice was founded to partner with advocates for low-income families and communities and to build long-term power among the people and communities that AI is leaving behind. It is not too late to take stock and correct course before AI becomes entrenched as a legitimate way to make key decisions impacting children and families.
If you are interested in learning more, read the new report or reach out directly to TechTonic Justice at techtonicjustice.org. You can request training for your organization or a community group, sign up for their newsletter, and otherwise join TechTonic Justice’s efforts to create a more just world where technology serves our collective well-being.
Acknowledgement
Many thanks to Kevin De Liban, Founder and President of TechTonic Justice for generously sharing his expertise and insights for this blog and during the Partnership Peer Exchange webinar.