Home Artificial Intelligence AI Bias: Good intentions can lead to nasty results | by Cassie Kozyrkov | Sep, 2023

AI Bias: Good intentions can lead to nasty results | by Cassie Kozyrkov | Sep, 2023

by Joey De Leon

Why fairness through unawareness is a pretty idea with ugly consequences

AI has become an integral part of our lives, from powering virtual assistants to making critical decisions in various fields. However, it is essential to remember that AI is not magic. Its abilities are a result of pattern recognition and the safety nets built into it by its programmers.

When we develop an AI solution, it is crucial to acknowledge that the results we get are not based on the outcomes we hope for but rather on the patterns present in the data we feed it. AI systems are built entirely on the patterns they identify in the examples they are given, and they optimize their behaviors based on the objectives we set for them.

Therefore, it should not come as a surprise when an AI system utilizes the patterns it finds within the data, even if we try to hide them. The system will use the available information to make decisions and judgments, regardless of our intentions.

To ensure AI safety and mitigate potential negative consequences, it is imperative to incorporate policy layers into AI systems. Policy layers serve as the AI equivalent of human etiquette, providing a set of guidelines and filters for the system’s output.

Just as society teaches us manners and filters our speech in public, policy layers act as a set of rules and checks for AI systems. They sit on top of the machine learning/artificial intelligence system and determine how the output is filtered and what actions should be taken.

Policy layers offer an additional layer of protection and accountability, helping to prevent biased or harmful decisions made by AI systems. They act as safety nets, ensuring that the system’s outcomes align with ethical and moral standards.

In my previous article on AI system reliability, I highlighted the importance of policy layers. They serve as the heroes, ensuring that AI systems are reliable and safeguarding against foreseeable errors. Additionally, policy layers offer the advantage of being relatively easy to modify without the need for extensive system retraining, allowing for quick reaction times in the face of unforeseen mistakes.

The concept of fairness through unawareness suggests that by hiding certain patterns from AI systems, we can eliminate biases and ensure fair outcomes. However, this approach may lead to undesirable consequences. Instead, implementing policy layers provides a more practical and effective solution to promote fairness and ethical decision-making.

AI technology continues to advance, and the responsible implementation of policy layers and ethical guidelines should be a priority. By doing so, we can harness the power of AI while ensuring that it aligns with our values and serves the best interests of society as a whole.

You may also like