Back to Presentations

Eliminating bias in forensic algorithms and statistics

M. W. Perlin, "Eliminating bias in forensic algorithms and statistics", American Academy of Forensic Sciences 73rd Annual Meeting, Interdisciplinary Symposium, 16-Feb-2021.


Talk

PowerPoint presentation of Dr. Perlin's talk.


PDFDownload Handout
PPTDownload PowerPoint

Abstract

The 2021 American Academy of Forensic Sciences meeting theme was One Academy Pursuing Justice through Truth in Evidence. To support and illustrate this theme, a four-hour Interdisciplinary Symposium was held – “To See or Not to See: Unbiased Answers to Forensic Questions.” The focus was the role of bias in targeted versus non-targeted forensic analysis, and efforts to overcome those biases.

Dr. Mark Perlin’s 55-minute talk addresses eliminating bias in forensic algorithms and statistics. He begins by discussing targeted forensic science approaches, showing different forms of human confirmation bias. He then introduces non-targeted approaches, illustrated with automated interpretation of DNA mixture evidence. The TrueAllele® computer method eliminates bias by decoupling data analysis and match comparison.

In the second half of the talk, Dr. Perlin presents a DNA case example – California v. Lopez. The defendant was accused of raping and killing his girlfriend’s toddler son. A rectal swab showed semen that matched the defendant’s DNA. The prosecution demanded the death penalty; the defense maintained his innocence. What was the truth? Could better science overcome the limits of confirmation bias? The forensic story is a cautionary tale.

Non-targeted statistical computing can eliminate human bias from forensic comparison. Past and current DNA cases must be reassessed with an accurate, objective and automated computer solution. This presentation shows how better science can deliver unbiased answers to forensic questions, achieving justice through truth in evidence.


Links