This week we look at the limits of expertise in innovation, and at the messy tradeoff between accuracy of AI and privacy of...you.

<<Support my work: book a keynote or briefing!>>

Want to support my work but don't need a keynote from a mad scientist? Become a paid subscriber to this newsletter and recommend to friends!

Research Roundup

Accuracy needn’t cost Privacy

Whether out of convenience or laziness, AI models of students or patients typically assume the data are all equally valid. If someone says they have a family history of cancer and another does not, then those data points are equally likely to be true (i.e., we assume the same noise parameter for everything). But what happiness when that assumption is both wrong and associated with protected data?

In an analysis of “77,836 adults with no history of colorectal cancer”, the “predictive value of self-reported family history was greater for White participants than for Black participants” due to different quality of medical records and family histories. As a result a “race-blind algorithm underpredicted risk for Black participants”, while an algorithm that took race into account “increased the fraction of Black participants” that should be called in for screening exams.

This finding mirrors work done on loan algorithms that shows increased lending rates for Black applicants when race was taken into account. But home loans and health are deeply private data. At The Human Trust, we believe that access to home and health should never require “selling” your private data. New research on differential privacy (DP), a method for training models without exposing data, finds that DP “can render…attacks impossible, while drops in performance are negligible”.

This leads the authors to claim that “not using DP at all is negligent when applying artificial intelligence models to sensitive data”. Better model, more accurate diagnoses, more privacy—All AI should be moving in this direction.

Science on the Balance Beam

Are experts better at predicting the future?

When evaluating projects within their area of expertise, “domain experts…are predictive of…success” but only in the sense that they can spot the projects that will fail. In other words, experts are very good at seeing where a plan runs counter to the known. For the unknown, however, they “make large errors” predicting which proposals will have the biggest impacts. Expertise is powerful in filtering out bad plans, but innovation seems to rely more on creativity and unique insights.

In fact, “the idiosyncrasy of conscious and unconscious decisions” makes the outcomes of research hard to predict. When teams of experts were tasked to “independently test the same prominent social science hypothesis” using identical data, the “teams reported both widely diverging numerical findings and substantive conclusions despite identical start conditions”. The teams’ “expertise, prior beliefs, and expectations barely predict” the wildly differing outcomes.

Expertise truly matters, particularly when dealing with the known. But it is creativity and exploration that breaks through the unknown. (btw, AI = known)

Weekly Indulgence

I’m hosting a party at the UN! But it has the worst theme: “Breaking Barriers in Women’s Mental Health” …or maybe it’s the best theme or at least one that has been ignored for too long. Please come support my work on postpartum depression, healthy online spaces, and even the transformative value of women’s entrepreneurship.

You are invited to a special luncheon on September 17th, 2024, at 12:00 PM, hosted at the United Nations Delegates Dining Room: “Breaking Barriers in Women’s Mental Health”. This event, organized by The Human Trust and Dionysus Health, is dedicated to addressing critical issues surrounding women’s mental health and exploring collaborative solutions to enhance mental health services globally.

We are honored to announce that Jesse Draper, founder of Halogen Ventures, will be our keynote speaker. Jesse’s unwavering support for women-led businesses and her commitment to mental health advocacy make her an empowering voice for this vital conversation.

This luncheon will bring together leaders from healthcare, policy, advocacy, research, and more, with the goal of sharing knowledge, identifying challenges, and developing actionable strategies to improve mental health outcomes for women worldwide. Please reach out to Conor Arevalo at carevalo@charitybrands.com, for details. We look forward to your presence and hope that you or a member of your team can join us for this impactful event.

Stage & Screen

Next up this month:

• September 8, Athens: ESOMAR
• September 17, NYC: Dionysus and The Human Trust are holding doing something amazing at the UN. Stay tuned to be apart of it.
• September 26, Wyoming: Tetons Leadership Counsel

See my full schedule and if your company, university, or conference just happen to be in one of the above locations and want the "best keynote I've ever heard" (shockingly spoken by multiple audiences last year)?

SciFi, Fantasy, & Me

Scifi time travel thriller with flecks of neuroscience…how could I not enjoy Recursion. If you had wished The Adam Project was less explosiony and more thoughtful you might enjoy Recursion.

P.S., Congratulations to Starter Villain, a classic bit of genre comedy from John Scalzi, for winning the Dragon Award. I enjoyed the story and love the cover.

Vivienne L'Ecuyer Ming