Bacon is Salt

Wake Up and Smell the Coffee

My Recent Podcast

My most recent episode of Brave New World features Alex Wiltschko, CEO of Osmo.ai. Osmo is a Google spinoff whose mission is to enable computers to smell.  Alex has a fascinating background as a serial entrepreneur and machine learning engineer at Twitter and Google.

Unlike vision and language, smell is still a very poorly understood sense, but it’s a critical one that has played a major role in the evolution of species. You might not realize this, but you wouldn’t be able to taste much without your sense of smell. Bacon would taste like salt, coffee would just taste bitter, and a pizza would be relatively tasteless. Equally importantly, we are learning that smell is deeply connected to emotion, and an essential part of our well-being. Its potential applications are endless, and include medicine, supply chains, security, mental health, and more.

Osmo has made impressive progress already by mapping molecular structure onto smell. This means that if you input a molecule, Osmo’s algorithm predicts its smell. Smell is described in terms of “smell percepts,” such as “smoky,” “floral,” etc. What’s interesting is that some are “closer” to others, but the “geometry of olfaction” is just beginning to emerge.

So, check out my conversation with Alex about this fascinating sense of ours:

https://bravenewpodcast.com/episodes/2024/04/04/episode-81-alex-wiltschko-on-the-sense-of-smell/

Kahneman’s Noise

I was saddened to hear that Daniel Kahneman passed away last week. Danny, as everyone called him, was my early podcast guest on Brave New World, where we discussed his latest book, Noise. Here’s a link to our conversation:

https://bravenewpodcast.com/episodes/2021/09/16/episode-21-daniel-kahneman-on-how-noise-hampers-judgemenst/

Kahneman changed how the world thinks about decision-making. He made us aware of the limitations of the human mind as a measuring instrument. These limitations give rise to “bias” and “noise” – two independent kinds of errors that arise in decision-making. Using a shooting analogy, bias is a systematic tendency for shots to be shifted in a specific direction relative to the bullseye, whereas noise is the scatter, regardless of where they lie relative to the bullseye. Kahneman defines noise as the undesirable variance among judges on identical cases, or the inconsistency of a single judge on identical cases on different occasions. Two felons who receive sentences of three years and seven years by different judges when they should both receive five is an example of noise in the system. For such a crime, a judge whose sentences are systematically under five years versus another judge who is systematically over five years is an example of bias in the judicial system.

During our conversation, Danny argued that while bias gets most of the attention, noise is often the more serious culprit and is overlooked. Unlike bias, which is hard to quantify because we first need to define the truth, noise is everywhere and easier to measure. A main reason for writing the book was to redress that imbalance between bias and noise through concrete examples. As Danny put it, the variance among underwriters in an insurance company was four to five times larger than they had expected. It’s a pervasive problem across all areas of our lives involving human judgment: justice, health, child custody, immigration, hiring, patents, forecasting, insurance, and more.

I reviewed Noise for the LA Times Review of Books. Kahneman and his co-authors use lots of compelling examples and crisp measurements to demonstrate why noise is typically the bigger culprit in errors of judgment. You can find my review here:

https://lareviewofbooks.org/article/dissecting-noise/

We had a wonderful lunch in Eataly after our podcast conversation, where we discussed AI, and its impact on business, government, democracy and liberalism. We continued to communicate, and in my last message to him a few months ago – in preparation for a conversation on free will with my most recent guest Kevin Mitchell – I asked him whether free will might be related to noise. I didn’t realize he was on his way out.

It was great to get to know you Danny. You changed the way I think about decisions, especially the contrast between human and algorithmic decision-making, and the challenges in combining the two.

What Resonated With Me

I will share three nuggets from our conversation that have stayed with me, and that were unique to our conversation.  

I can appreciate that variability in sentencing decisions is undesirable, I asked Danny, but why does noise arise in the first place? I’ll summarize his takeaways using three excerpts from our conversation.

First, people are shockingly different:

The main source of noise is that people look at the world and they see the world differently. And they see every case differently from the way others see it. If you took the judges, and you gave them say 15 different crimes, and you ask them to rank the crimes, from the most severe to the least severe, they would not agree on their ranking. It's not only that one of them would set higher sentences than the other, the ranking would be different. And that is it turns out the biggest source of noise, and it's the most mysterious. We know that people are different from each other. But we're not quite prepared to recognize the extent to which people see the same situation differently from one another.

Second, we are unaware of our “objective ignorance” about things, and don’t realize how unpredictable things are:

An error in prediction is not necessarily a mistake, because many events that we try to predict or forecast are simply not completely predictable.  I use the term objective ignorance for exactly that. You've described predictability, where there is a limit, an objective limit to what we can accomplish. And we'd better be aware of it, because we tend to blame people for having failed to forecast events. We blame them in hindsight, because in hindsight, we understand the events. But in fact, because of objective ignorance, they could not have predicted the event.

Third, we understand the world in causal terms after the fact by filling in the blanks conveniently, even though we had no idea ex-ante how things would unfold:

We live by making sense of what happens as it happens. And we have causal interpretations. And after the fact it seems obvious that it had to happen that way, because there is an inevitability in hindsight to what did happen. And that gives us the sense that we are in in touch with reality, that we understand the world. And that is to a significant degree an illusion.

These points really resonated will me. Experts tend to offer causal explanations after the fact with great confidence, which provides a false illusion of predictability.

Considering that noise undermines our trust in a system, I asked Danny how should we think about merging noisy humans and AI in various systems such as justice, health and business to make better decisions. Should we strive for consistency and let machines make decisions when they are better than those of humans? Or is every situation completely different and in need of human judgment?  He replied:

In many situations, when you can substitute a better instrument for human judgment, I think you should. This sometimes raises difficult moral questions.

He also mentioned that the “combination of humans and machines can be unstable.”

This is perhaps one of the most vexing problems at the intersection of human decision-making and AI, that is, how we reconcile our need for uniqueness with our desire for consistency and higher quality decisions. 

The Anxious Generation

Last week, I walked into class wearing my headphones, waiting for the song to finish while I set up. A student asked me what I was listening to.

Pink Floyd, I said. Anyone listen to Pink Floyd?

Zero hands went up. I was shocked that they were unaware of such an iconic rock and roll band. It reminded me of my conversation with Jonathan Haidt, who mentioned that the majority of content consumed by kids today was produced in the last month!  In contrast, previous generations grew up on materials – cartoons, movies, books – created by the previous generation. Jon has a new book out this week called The Anxious Generation, which you should check out. It’s about the havoc that social media platforms have wreaked on our kids.

The last Pink Floyd concert I attended was at the Meadowlands in New Jersey when I was in my thirties. I was sitting next to a fifteen-year-old who was puffing away at a joint. When the band started, I asked him which album the song was from.

He looked at me up and down and responded “They’ll play your stuff after the break!”