Polygraph-based deception detection and Machine Learning. Combining the Worst of Both Worlds?

Area of Science:

• Computational Science
• Forensic Science
• Legal Technology

Background:

• Critically examines the application of Machine Learning (ML) and Artificial Intelligence (AI) to polygraph screening data.
• Highlights the increasing challenges in ML/AI concerning fairness, transparency, and accountability.
• Questions the scientific validity of polygraph-based deception detection methods.

Discussion:

• Argues that applying ML to ambiguously labeled polygraph data is methodologically unsound.
• Contends that such research can lend false legitimacy to scientifically invalid techniques.
• Emphasizes heightened legal and ethical considerations for research in high-stakes environments like criminal justice and employment.

Key Insights:

• Polygraph screening lacks scientific validity, making ML application problematic.
• ML models trained on questionable data risk perpetuating pseudoscience.
• Ethical and methodological rigor must precede the application of advanced computational methods to sensitive data.

Outlook:

• Recommends addressing methodological concerns before validating polygraph techniques.
• Calls for greater scrutiny of research applying ML to forensic and legal contexts.
• Stresses the need for robust scientific foundations in legal and employment screening tools.