Automated Decision Making
It is vital to distinguish between full automation and systems designed to assist humans.
| Feature | Automated Decision Making | Augmented Decision Making |
|---|---|---|
| Human Role | None (at the point of decision) | Human-in-the-loop (final approval) |
| Speed | Near-instantaneous | Limited by human reaction time |
| Accountability | Often blurred or systemic | Resides with the human operator |
| Complexity | Handles massive data volume | Handles nuanced, contextual edge cases |
Identify the Stakeholders: When analyzing a scenario, always consider three groups: the decision-maker (organization), the subject (individual affected), and the developer (algorithm creator).
Check for Bias Sources: Look for 'proxy variables' where data that seems neutral (like a zip code) might actually represent a biased category (like race or income level).
Evaluate Transparency: If a question asks about the 'Black Box' problem, focus your answer on the lack of interpretability—the inability to explain why a specific output was generated.
Verify the Feedback Loop: Always check if the system has a mechanism to learn from its mistakes; without a feedback loop, errors in the initial data will be amplified over time.
The Neutrality Myth: A common misconception is that algorithms are inherently objective; in reality, they often mirror and amplify the human biases present in their training data.
Context Blindness: Automated systems often fail to account for 'edge cases' or unique human circumstances that fall outside the statistical norm of the training set.
Over-reliance (Automation Bias): Humans tend to trust automated outputs more than their own judgment, leading to a failure to intervene even when the system is clearly malfunctioning.