AI May Be Reproducing Historical Inequalities

Being data-driven doesn't mean being bias-free.

In a recent episode of The New Mainstream Podcast, ThinkNow Co-Founder and host Mario Carrasco spoke with Tim Turner, Director of Insights and Analytics at Thresholds, about a growing concern in the age of artificial intelligence: AI systems may be reproducing the same historical inequalities embedded in the data used to train them.

When Data Reflects the Past, AI Can Repeat It

Artificial intelligence is often perceived as objective. After all, algorithms rely on data rather than personal opinions or individual judgment. But what happens when that data reflects decades of inequality, underrepresentation, or systemic bias?

As Turner explains, historical data isn't necessarily neutral. It reflects the decisions, practices, and social conditions under which it was collected.

Consider predictive models trained on historical incarceration records. If certain communities have experienced disproportionate policing, those patterns may become embedded in the datasets used to train AI systems. Without careful evaluation, the resulting predictions could reinforce existing inequalities rather than challenge them.

Being Data-Driven Doesn't Mean Being Bias-Free

One of the most important insights from the conversation is that data itself is a human construct.

People determine what information to collect, how to categorize it, which variables matter, and how to interpret the results. This means that even the most sophisticated analytical models can reflect the assumptions and limitations of the people and systems that developed them. With that in mind, it's worth asking:

Are we using technology to make better decisions, or are we simply automating the biases of the past?

Being data-driven isn't enough. Organizations must also understand where their data comes from, whose experiences it represents, and which perspectives may be missing.

Responsible AI Starts with Responsible Data

At Thresholds, a Chicago-based nonprofit providing community mental health services, Turner approaches data and technology from a human-centered perspective.

His organization evaluates opportunities to implement artificial intelligence based on three fundamental principles:

• Align with the mission. Technology should address real organizational needs rather than be adopted simply because it's available.
• Maximize value. Successful implementation requires training, strategic prioritization, and tools that genuinely support employees and the people they serve.
• Minimize risk. Strong governance, privacy protections, and clear policies are essential, especially when working with sensitive information.

Representation is a Responsibility

This conversation highlights a particularly important responsibility: ensuring that the people represented in datasets reflect the diversity of the populations being studied.

When certain communities are missing, underrepresented, or inaccurately characterized, the resulting insights can offer an incomplete picture of reality.

The Future of AI Still Depends on Human Judgment

Technology cannot eliminate the need for human responsibility.

For Turner, the ultimate purpose of data is to empower people. At Thresholds, analytics supports clinical teams by providing relevant information to help them deliver more timely and personalized care.

AI may be transforming how organizations make decisions, but people remain responsible for ensuring those decisions are well-informed, equitable, and have a positive impact.

Listen to the and discover how data equity, AI governance, and human-centered decision-making are connected.