How AI could help doctors detect patterns | Christopher Petrilli, MD | TEDxNYU Langone Health

Key Concepts:

• Pattern Recognition
• Apophenia (finding patterns in randomness)
• Cognitive Bias (e.g., Availability Bias)
• Numeracy (understanding and working with numbers)
• Probability (likelihood of an event)
• Artificial Intelligence (AI)
• Human Intelligence (HI)

1. Pattern Recognition and Apophenia:

• Humans are natural pattern seekers, a trait crucial for survival (e.g., identifying a shadow as a potential shark).
• Apophenia is the tendency to perceive meaningful patterns in random data (e.g., seeing shapes in clouds, the Virgin Mary in toast).
• Apophenia can be harmless but also dangerous, leading to misdiagnosis in medicine.

2. Cognitive Biases:

• Cognitive biases are systematic errors in thinking that affect decision-making.
• Availability bias: Over-reliance on readily available information (e.g., a doctor assuming a patient has the flu during flu season).
• Availability bias can lead to life-threatening mistakes, even by skilled professionals.

3. The Role of AI:

• AI excels at pattern detection and data processing, surpassing human capabilities in speed and accuracy.
• AI is likened to an "overeager intern" – enthusiastic but lacking context and understanding.
• AI is not immune to biases because it is trained on human data, which contains existing biases.
• AI can amplify existing biases if left unchecked, reflecting past flawed decisions.

4. Numeracy and Probability:

• Numeracy is essential for understanding probability and making informed decisions.
• Emotions can impair our ability to apply numeracy objectively.
• Example: Rolling dice – understanding the probability of rolling a seven versus a twelve.

5. Case Study: Sally Clark and SIDS:

• Sally Clark was wrongly convicted of murdering her two children who died from SIDS.
• An expert witness incorrectly calculated the probability of two SIDS deaths in the same family as one in 73 million.
• The correct probability was closer to one in 60, considering the increased risk after the first SIDS death.
• The jury's emotional response to the perceived pattern (two dead babies) led to a wrongful conviction.
• Sally Clark's life was destroyed, highlighting the devastating consequences of misinterpreting statistical data.

6. AI and Human Collaboration:

• AI can objectively evaluate symptoms and suggest potential diagnoses that humans might overlook.
• AI is not a replacement for human doctors but a tool to augment their abilities.
• The speaker's friend was diagnosed with acute myeloid leukemia after Chat GPT suggested it as a possibility, which the urgent care doctor didn't consider.
• The diagnosis was made possible through a partnership between AI and a human doctor.

7. Conclusion:

• While apophenia is a real phenomenon, our brains' ability to find patterns is valuable.
• High-stakes decisions require collaboration between human intelligence and artificial intelligence.
• Combining human intuition and AI's analytical capabilities can lead to safer and better outcomes.
• AI and HI working together can create a world far safer and better than either of us could create alone.