Latest / Paths of Curiosity in an AI World / Q3 Recap - Data, Behavior & Human Systems: How Data Shapes What We Do, Think, and Become

Transcript

Q3 Recap - Data, Behavior & Human Systems: How Data Shapes What We Do, Think, and Become · Paths of Curiosity in an AI World

0:04
SPEAKER_02: Welcome to Paths of Curiosity in an AI world where we explore tomorrow's careers through the lens of AI.
0:22
SPEAKER_02: Okay, you get to pick.
0:24
SPEAKER_02: You can make a set of predictions yourself, or you can hand the job to an algorithm.
0:29
SPEAKER_02: Whoever you pick, you get paid for how accurate they are.
0:33
SPEAKER_02: And before you choose, you get to watch the algorithm practice.
0:38
SPEAKER_02: So you watch it, and it's wrong.
0:42
SPEAKER_02: A lot.
0:43
SPEAKER_02: It barely gets any of them right.
0:46
SPEAKER_02: But it is still better than you.
0:48
SPEAKER_02: It would earn you more money than you would earn yourself.
0:53
SPEAKER_02: So you picked the algorithm.
0:55
SPEAKER_02: But most people didn't.
0:57
SPEAKER_02: The people who never saw it work trusted it.
1:01
SPEAKER_02: The people who watched it make mistakes, fired it, met on themselves, and made less money doing it.
1:08
SPEAKER_02: They picked being wrong their own way over being less wrong the algorithm's way.
1:15
SPEAKER_02: That is the question this whole quarter.
1:17
SPEAKER_02: When you make a choice, how much of it is really yours?
1:21
SPEAKER_02: The feed you scroll?
1:22
SPEAKER_02: The ride a driver accepts, the score that decides your first credit card, the algorithm you trust or refuse.
1:30
SPEAKER_02: All of them look like free choices.
1:33
SPEAKER_02: Six guests told us why they usually are not.
1:51
SPEAKER_02: A financial policy expert who worked as the Deputy Assistant Secretary for Economic Policy at the United States Treasury and knows exactly who your credit score is quietly punishing.
2:04
SPEAKER_02: A sociologist who rode 5,000 miles with gig drivers, then went and worked inside Uber and came back out again.
2:13
SPEAKER_02: A political economist who wants to bring back the Leadites, but not the way you are picturing.
2:19
SPEAKER_02: A Microsoft researcher who spent 10 years building technology for the developing world and came home with a law about what technology cannot do.
2:29
SPEAKER_02: And a business school professor who figured out why you do not trust the algorithm, even when it is right.
2:36
SPEAKER_02: I'm Alora.
2:38
SPEAKER_02: And I'm Acadia, and this is the Data, Behavior, and Human Systems mini-series recap.
2:45
SPEAKER_02: Before we meet the guests, set it up for me.
2:48
SPEAKER_02: What ties these six conversations together?
2:51
SPEAKER_02: One question, six singles.
2:54
SPEAKER_02: How much of your life is a choice you made, and how much is a choice someone else made for you, and then dressed up to look like yours?
3:03
SPEAKER_02: That's a lot heavier than AI and behavior sounds.
3:07
SPEAKER_02: None of these six people think you are puppets.
3:10
SPEAKER_02: The setup is real, but once you can see it, you are not stuck.
3:14
SPEAKER_02: You just have to know where to look.
3:17
SPEAKER_02: So where do we start looking?
3:19
SPEAKER_02: Dr.
3:20
SPEAKER_02: Scion Bayant, behavioral economist at Swathmore and Wharton.
3:25
SPEAKER_02: And he started with something I did not expect.
3:28
SPEAKER_02: Tuberculosis.
3:29
SPEAKER_02: Which sounds like a strange place to start a conversation about choices.
3:33
SPEAKER_02: But a text message and a wristband made patients in Kenya two-thirds more likely to finish their treatment.
3:40
SPEAKER_02: Not a new drug, a reminder.
3:43
SPEAKER_04: Just essentially reminding them at the right time and following up with them is really what people need to do this very important activity.
3:49
SPEAKER_04: And so we should think about people more as human beings, flawed human beings, rather than these perfect rational robots, which economics sometimes makes it sound like everyone is out there just doing a bunch of math at the grocery store to figure out what to buy, but really people are people and our irrationality is kind through.
4:07
SPEAKER_02: This conversation starts by admitting people are bad at doing the obviously good thing.
4:13
SPEAKER_02: Can any of you name one person who has never put off something that takes four minutes?
4:19
SPEAKER_02: But once you accept that about people, nudging makes sense.
4:23
SPEAKER_02: His example is the cafeteria shelf.
4:26
SPEAKER_02: Put fruit at eye level and kids eat more fruit.
4:29
SPEAKER_02: Nobody took the chocolate away.
4:31
SPEAKER_02: It's still down there.
4:33
SPEAKER_02: You just made the fruit the easy grab.
4:36
SPEAKER_02: That's a sneaky little thought.
4:39
SPEAKER_02: He agrees.
4:41
SPEAKER_02: He says flat out that is manipulation.
4:44
SPEAKER_02: His argument is that there is no version where you are not doing it.
4:49
SPEAKER_02: Somebody has to put the food somewhere.
4:51
SPEAKER_02: Once you know eye level wins, you cannot unknow it.
4:55
SPEAKER_02: So you might as well pick the option that is good for the person.
5:00
SPEAKER_02: Okay, now to the version that isn't in a cafeteria.
5:05
SPEAKER_02: So what about your phone?
5:08
SPEAKER_04: They call you users for a reason, um, because you're hooked, basically, right, to this product where we say there are drug users and there are social media users, and that's pretty much the main two industries that refer to the people who use their products as users.
5:21
SPEAKER_04: So not a great company to be in.
5:23
SPEAKER_02: Same exact mechanism as the fruit shelf, different intention.
5:28
SPEAKER_02: That is the thread for the whole quarter.
5:31
SPEAKER_02: Episode two From Nudges to Numbers Aaron Klein from Bookings, who used to be the chief economist of the Senate Banking Committee and helped write Dodd Frank at the Treasury Department.
5:43
SPEAKER_02: Topic Your credit score.
5:46
SPEAKER_02: The AI making the first call on your financial life has a name.
5:51
SPEAKER_02: It's FICO.
5:52
SPEAKER_02: It is a black box, and one of the biggest things it weighs is how long you have been in the system.
6:00
SPEAKER_02: Which for a 17-year-old is zero.
6:04
SPEAKER_02: Zero.
6:05
SPEAKER_02: You start behind for the crime of being new.
6:08
SPEAKER_02: And a lot of young people do not get a score at all.
6:11
SPEAKER_02: They come back unscored, which is its own problem.
6:15
SPEAKER_02: He compared the whole thing to an SAT score that is not very accurate, but somehow matters enormously.
6:23
SPEAKER_02: Great.
6:24
SPEAKER_02: And then there's the part that got me.
6:26
SPEAKER_02: There is a list of things a lender is flatly barred from using race, religion, national origin, sex, and marital status.
6:35
SPEAKER_02: Congress read those in the Equal Credit Opportunity Act in the 1970s.
6:41
SPEAKER_02: Okay, so that is handled.
6:44
SPEAKER_02: Except here's the study Klein pointed us to.
6:47
SPEAKER_02: Four researchers, one of them Manjo Piri at Duke, got data on about 250,000 online purchases at a German furniture retailer.
6:57
SPEAKER_02: People were buying on invoice, so the company had to guess who would pay.
7:02
SPEAKER_02: And the researcher is asked a simple question.
7:06
SPEAKER_02: How much can you predict from the things the website picks up for free?
7:10
SPEAKER_02: Just because you showed up?
7:13
SPEAKER_02: Such as whether the phone is an iPhone or an Android, what time you checked out, what your email provider is, whether your own name is in your email address.
7:24
SPEAKER_02: That is not financial information.
7:27
SPEAKER_02: None of it is, and it predicted default about as well as the actual credit score did.
7:33
SPEAKER_02: Slightly better in their data.
7:36
SPEAKER_02: Orders placed between midnight and six in the morning defaulted about twice as often as the average.
7:42
SPEAKER_02: People whose own name was in the email address defaulted less.
7:47
SPEAKER_01: Should you be able to use Mac vs PC?
7:50
SPEAKER_01: On the one hand, that's very correlated with race.
7:52
SPEAKER_01: On the other hand, I don't feel like Macs is going out trying to get just white people.
7:57
SPEAKER_01: I don't think it's a product with a legacy in history that way.
8:00
SPEAKER_01: I could be wrong, other people can disagree, but we need to be able to have those kinds of conversations.
8:05
SPEAKER_01: Why are Mac users better credit risks than PCs?
8:08
SPEAKER_01: Is what is that really telling us?
8:10
SPEAKER_02: And nobody tells you which one you failed.
8:14
SPEAKER_02: He said it is confusing by design.
8:16
SPEAKER_02: If you knew your email address moved your score, you would go and make a new email address and then it would stop telling anyone anything.
8:24
SPEAKER_02: Meanwhile, it is illegal to use your gender to price your credit card.
8:28
SPEAKER_02: However, in most states it is perfectly legal to use it to price car insurance for that same 16 year old.
8:35
SPEAKER_02: Same kid, same day, two answers.
8:40
SPEAKER_02: So somebody should build a better one and swap it in.
8:43
SPEAKER_02: That is what I said, and he answers with an oboe.
8:47
SPEAKER_02: With a what?
8:49
SPEAKER_02: An orchestra tunes to an obo's A.
8:51
SPEAKER_02: Not because the oboe is most accurate.
8:54
SPEAKER_02: The violin is.
8:55
SPEAKER_02: They tune to the oboe because the oboe is the hardest thing in the room to tune.
9:00
SPEAKER_02: So if the oboe is a little flat, the whole orchestra goes a little flat on purpose.
9:06
SPEAKER_02: Because the goal isn't being right.
9:08
SPEAKER_02: The goal is matching.
9:10
SPEAKER_01: And so I've often said that the entire financial system is tuned to FICO, these credit scores, which are like an out-of-tune oboe.
9:22
SPEAKER_01: And so, you know, what's standing in the way is it's very difficult to get everybody to retune.
9:27
SPEAKER_01: And if a few people start doing things a little more accurately, the rest of the system isn't going to incorporate that.
9:34
SPEAKER_02: So a better algorithm loses to a worse one everybody already uses.
9:40
SPEAKER_02: And here's the kicker.
9:41
SPEAKER_02: He thinks if FICO were invented today, it would fail every fairness test we have and never get approved.
9:48
SPEAKER_02: It only survives because it is old, baked into the system.
9:52
SPEAKER_02: New information gets held to a higher standard than the bias information it would replace.
9:58
SPEAKER_02: Which quietly guarantees the old thing wins forever.
10:03
SPEAKER_02: Which is exactly the kind of problem that needs people to understand the tech and the policy.
10:09
SPEAKER_02: Episode 3.
10:10
SPEAKER_02: From your bank to your car.
10:26
SPEAKER_02: She found that the app is the boss, not a metaphor.
10:30
SPEAKER_02: The app assigns your work, rates your work, and decides whether you keep working.
10:35
SPEAKER_02: She has a name for that.
10:36
SPEAKER_03: Instead of having direct control over how people do their jobs, you have a system of algorithmic management that becomes an intermediary for how to control the labor process people subscribe to without necessarily telling them exactly or precisely what they have to do to achieve their performance metrics.
10:57
SPEAKER_03: That has real implications for how workers are classified.
11:01
SPEAKER_03: If you can achieve control without that direct relationship, employers can elide some of the rules designed to protect workers.
11:11
SPEAKER_02: Say the last part again, in normal words.
11:14
SPEAKER_02: If you control a worker through a person, you are an employer.
11:18
SPEAKER_02: And a hundred years of law comes with that.
11:21
SPEAKER_02: Minimum wage, unemployment, the right to organize.
11:25
SPEAKER_02: However, if you control the workers through an app instead, you are just the platform they choose to use?
11:33
SPEAKER_02: Same control, none of the obligations.
11:37
SPEAKER_02: That is the whole move.
11:38
SPEAKER_02: And it is not that the control is loose.
11:41
SPEAKER_02: It is tighter in some ways because it never sleeps.
11:44
SPEAKER_02: Okay, so how does that actually reach the driver?
11:48
SPEAKER_02: Through what they are not told.
11:51
SPEAKER_03: Well, drivers never received, and it's not just super, drivers from Lyft from others don't receive an employee handbook saying these are what the rules are and this is what your expectation is.
12:02
SPEAKER_03: They generally have some community guidelines and they have instead performance metrics that are very changeable.
12:11
SPEAKER_02: So what do you do if you want to keep your job?
12:14
SPEAKER_02: You guess.
12:15
SPEAKER_02: That is her word.
12:17
SPEAKER_02: Drivers have to guess at what the right behavior is to please the algorithm.
12:22
SPEAKER_02: And the right behavior changes, and it changes by city.
12:25
SPEAKER_02: She watched drivers in Seattle opt out of carpool rides while drivers in the Northeast City could not.
12:32
SPEAKER_02: Same company, same week.
12:34
SPEAKER_02: So you're being graded on a rubric nobody will show you.
12:38
SPEAKER_02: And the rubric is different in the next town over.
12:42
SPEAKER_02: And you learn the rules from other drivers, and forums, from rumors, from watching who disappears.
12:49
SPEAKER_02: Crowdsourcing your own handbook.
12:52
SPEAKER_02: Here's the sharpest version of it.
12:54
SPEAKER_02: For years you could not see where a trip was going before you accepted it.
12:59
SPEAKER_02: So you do not know how far it is, you do not know what it pays, and you do not know if it strands you somewhere with no ride back.
13:08
SPEAKER_02: So it's a choice with the information removed.
13:11
SPEAKER_02: And the information is not missing by accident.
13:15
SPEAKER_02: The company has it, and if you decline, your acceptance rate moves, and if it moves too far, you can be deactivated, and nobody will tell you where that line is.
13:26
SPEAKER_02: She said drivers never have clear insight into how often deactivation actually happens.
13:32
SPEAKER_02: That is information only the platform holds.
13:36
SPEAKER_02: And if it happens to you?
13:38
SPEAKER_02: That's the part she said mattered most.
13:40
SPEAKER_02: Some of these drivers have taken out a lease on their car.
13:43
SPEAKER_02: They have to put real money and real time into learning the platform.
13:47
SPEAKER_02: And then one day the account is off.
13:50
SPEAKER_03: But where the rubber really hit the road centered on accountability.
13:54
SPEAKER_03: If you're a driver who's taken out a lease or you've invested significant capital or time in learning these platforms and you're suddenly deactivated one day, what's your recourse?
14:04
SPEAKER_03: What do you do when things go wrong?
14:07
SPEAKER_02: The answer is that there is no manager to appeal to you because the company's position is that it was never your manager.
14:14
SPEAKER_02: I want to ask the obvious thing.
14:16
SPEAKER_02: Do drivers hate this?
14:18
SPEAKER_02: No, and I think that is important because it would be easy to tell this as a story about miserable people.
14:24
SPEAKER_02: A lot of the drivers she rode with genuinely valued the flexibility.
14:28
SPEAKER_02: She interviewed a woman in New Orleans whose son had bipolar disease who needed to be able to drop a shift when something happened.
14:36
SPEAKER_02: And she said some drivers actually prefer being told what to do by a company over being told by a human supervisor.
14:44
SPEAKER_02: So the flexibility is real.
14:47
SPEAKER_02: The world that is doing the damage is a different one.
14:50
SPEAKER_03: But the difference is these drivers were being valorized through corporate rhetoric about how they were entrepreneurs.
14:57
SPEAKER_03: And an entrepreneur, or even just an independent contractor, is supposed to be able to make informed choices about their decisions.
15:05
SPEAKER_03: And they really couldn't make those decisions in their own best interest under those limited and constrained circumstances.
15:12
SPEAKER_02: That is the trick.
15:14
SPEAKER_02: You get called the boss of your own business, and then you get handed a job where you cannot see the pay before you take it.
15:21
SPEAKER_03: So the sum up of it is, I would say, after all of these years, is that they have a lot of freedom to make really constrained choices.
15:28
SPEAKER_03: And they operate in that space, I think, to this day.
15:32
SPEAKER_02: And she's finding the same shape in hospitals now.
15:36
SPEAKER_02: Same shape.
15:37
SPEAKER_02: She's auditing the forms you fill out at the registration desk.
15:42
SPEAKER_02: And what she keeps finding is a page that tells you that you have three choices about your data and then one button at the bottom.
15:50
SPEAKER_02: Let me guess what it says.
15:52
SPEAKER_02: I accept.
15:54
SPEAKER_03: If you have the ability or the right to say no, can you?
15:58
SPEAKER_03: And the answer is kind of shocking because often it's basically impossible.
16:02
SPEAKER_02: So the illusion of free choice is not that somebody took the choice away.
16:07
SPEAKER_02: They left it right there in front of you.
16:09
SPEAKER_02: They just took away everything you would need to use it.
16:12
SPEAKER_02: Episode 4.
16:13
SPEAKER_02: Jathan Sigowski, political economist at Monasha in Australia.
16:19
SPEAKER_02: He wrote Too Smart and the Mechanic and the Luddite and hosts a podcast called This Machine Kills.
16:26
SPEAKER_02: And he wants the word Luddite back.
16:28
SPEAKER_02: Here's what I did not know.
16:30
SPEAKER_02: When the Luddites broke into a workshop, they did not wreck everything in it.
16:34
SPEAKER_02: Some machines got destroyed, some got left alone.
16:37
SPEAKER_02: Same room.
16:38
SPEAKER_02: How did they pick?
16:40
SPEAKER_02: By who owned it and how that owner treated people.
16:43
SPEAKER_02: So they were aiming at something other than the machine.
16:47
SPEAKER_02: So this is a sentence I keep coming back to.
16:50
SPEAKER_06: Crashing machines, again, not because they were afraid of technology, but because they saw the technologies, the machines as tools of larger political, economic, and social systems that were oppressing them, exploiting them, extracting their labor and value in various ways.
17:07
SPEAKER_02: The machine is not the thing doing it to you.
17:10
SPEAKER_02: That is the system.
17:11
SPEAKER_02: The machine is part of the thing you can reach.
17:14
SPEAKER_02: Somebody decided to cut the rate.
17:17
SPEAKER_02: Somebody decided the cheap goods were good enough.
17:20
SPEAKER_02: You cannot get at a decision.
17:22
SPEAKER_02: You cannot get at a frame.
17:23
SPEAKER_02: So the hammer is aimed at pass the machine at whoever put it there.
17:28
SPEAKER_02: And once you see it that way, the label everyone uses is backwards.
17:32
SPEAKER_06: You know, the Lovites, I always say it was a labor movement, right?
17:36
SPEAKER_06: It was not an anti-technology movement, it was a pro-labour movement.
17:40
SPEAKER_02: And they tried everything else first.
17:43
SPEAKER_02: They negotiated, they petitioned, they protested, and they were dismissed every single time.
17:49
SPEAKER_02: He gave an example of that today.
17:52
SPEAKER_02: In schools.
17:53
SPEAKER_02: Districts were sending laptops home with monitoring software on them.
17:57
SPEAKER_02: Keystrokes, microphone, camera.
18:00
SPEAKER_02: They were cases where families went into the settings, turned off recordings, and the laptop recorded them at home anyways.
18:09
SPEAKER_02: And the laptop is the frame.
18:11
SPEAKER_02: Nobody is being surveyed by a laptop.
18:14
SPEAKER_02: Somebody signed that contract.
18:16
SPEAKER_02: Which is the part everyone skips.
18:19
SPEAKER_02: And that is his sharpest point in the whole conversation.
18:22
SPEAKER_06: And so I didn't deny your job application, an algorithm did, right?
18:27
SPEAKER_06: Or I didn't deny your promotion, an algorithm did.
18:30
SPEAKER_06: Or I didn't do this to you or that to you, an algorithm did, right?
18:34
SPEAKER_06: But that's deception, right?
18:36
SPEAKER_06: That is lying.
18:38
SPEAKER_06: It's cynical because at the end of the day, the algorithm is never going to tap you on the shoulder and tell you to fuck up.
18:45
SPEAKER_06: This is my office now, right?
18:47
SPEAKER_06: It's always a boss behind the algorithm or behind the robot who has made a decision to use that technology to downsize, to reduce head count, to lower wages, to degrade the quality of working conditions, right?
19:03
SPEAKER_06: It is always a boss, uh a consultant, uh, a company behind those decisions.
19:10
SPEAKER_06: The technology just becomes used as a shield for it, right?
19:13
SPEAKER_02: 200 years apart, and it's the same sentence.
19:17
SPEAKER_02: The machine is standing in front.
19:19
SPEAKER_02: So what is the question?
19:21
SPEAKER_06: We should hold up the technology and say, not does this thing spark joy, but does this thing make my life better?
19:29
SPEAKER_06: Does it make me a better person?
19:31
SPEAKER_06: Does it contribute to social welfare in some way, right?
19:34
SPEAKER_06: Making a better society.
19:36
SPEAKER_06: And once we you know hold up and compare technology through that very simple criteria, I think you'd be surprised at how many of them failed to pass that simple test.
19:48
SPEAKER_02: Instead of does this spark joy, does this make my life better?
19:53
SPEAKER_02: And it is the same question those workers were asking in a workshop.
19:57
SPEAKER_02: Not is this a good machine, who is it good for?
20:01
SPEAKER_02: Episode 5 zooms all the way out.
20:04
SPEAKER_02: Kentaro Toyoma, ten years at Microsoft Research, co-founded their research lab in India, now the University of Michigan, wrote Geek Heresy.
20:14
SPEAKER_02: And he is the one who says technology does not fix broken things.
20:19
SPEAKER_00: In the end, what technology does primarily is not necessarily to improve things, but to amplify whatever underlying human forces there already are.
20:29
SPEAKER_00: And so what that means is if the human forces are well-intentioned and capable, that's great.
20:34
SPEAKER_00: Technology makes things better.
20:36
SPEAKER_00: Uh but in so many of the challenges that we were trying to address, the underlying human forces were, you know, were indifferent or dysfunctional or sometimes outright corrupt.
20:46
SPEAKER_00: And in those situations, even the best design technology doesn't uh fix the problem.
20:52
SPEAKER_02: And he learned that from a project that worked.
20:56
SPEAKER_02: His case study is this.
20:58
SPEAKER_02: There were rural schools in India, 30 kids, a handful of computers, so five or ten kids crowded one machine, and one confident kid takes the mouse and everybody else watches.
21:09
SPEAKER_02: So the fix seems simple and fruitful.
21:11
SPEAKER_02: His team plugs in a mouse for every kid and rewrites the software so everyone gets their own colored cursor.
21:18
SPEAKER_02: Five kids on one computer learn as much as five kids with five computers.
21:24
SPEAKER_02: And then it leaves the lab.
21:26
SPEAKER_02: Principals locked the computers in closets because they were scared they would break.
21:31
SPEAKER_02: Teachers would not touch them because they felt insecure about their own skills.
21:35
SPEAKER_02: Almost no school had anyone to fix anything.
21:39
SPEAKER_02: None of which is a technology problem.
21:42
SPEAKER_02: None of it.
21:43
SPEAKER_02: So his rule is this follow the power, look at who controls the technology, and that tells you what it is going to amplify.
21:51
SPEAKER_02: Which right now, for AI, is a very small number of companies.
21:56
SPEAKER_02: And then there is the detail.
21:58
SPEAKER_02: If the service is free, You are the product.
22:01
SPEAKER_02: And his version goes even further.
22:03
SPEAKER_02: If the AI guidance is free, your brain is the product.
22:07
SPEAKER_02: And his reason for this is kind of devastating.
22:11
SPEAKER_02: Your attention is your time, and your time is literally your life.
22:16
SPEAKER_02: Every video or ad is a piece of your life you donated so a company could sell it.
22:23
SPEAKER_02: And he thinks AI mostly follows the law of amplification more than ever.
22:29
SPEAKER_02: But with a slight caveat.
22:31
SPEAKER_02: AI is the first technology we treat like a person, and research is starting to show it can be better than humans at changing your mind.
22:40
SPEAKER_02: Once a technology changes what you want, that is not just amplifying anymore.
22:46
SPEAKER_02: Episode 6.
22:47
SPEAKER_02: The most personal question of the quarter Why do we not trust algorithms even when they are better than us?
22:54
SPEAKER_02: Berkeley divorced at Chicago Booth who coined the term algorithm inversion.
23:00
SPEAKER_02: And his own experiment surprised him.
23:03
SPEAKER_02: Completely, people made predictions for real money.
23:07
SPEAKER_02: Some watched an algorithm work first, some did not.
23:10
SPEAKER_02: He expected the people who saw it perform well would use it, and those who did not would opt not to.
23:17
SPEAKER_02: But the opposite thing happened.
23:20
SPEAKER_02: The people who never saw it just assumed it was great and used it.
23:24
SPEAKER_02: The people who watched it work saw it miss a bunch, decided it looked bad, and trusted themselves instead, even when the algorithm was still more accurate than the human.
24:02
SPEAKER_05: And so people, once they saw that's how well the algorithm performed, they just decided I don't want to use that, that doesn't look good enough.
24:10
SPEAKER_02: They gave them money to avoid watching something be imperfect.
24:15
SPEAKER_02: Real money.
24:16
SPEAKER_02: And the fix was not making it more accurate.
24:19
SPEAKER_02: The fix was letting people tweak it a little.
24:38
SPEAKER_05: So you're kind of almost giving in and saying, Alright, I'm okay with being wrong a lot.
24:44
SPEAKER_05: And people don't like that.
24:45
SPEAKER_05: People don't like adopting a system where they know a lot of the time it's going to be wrong, and I have to be okay with that.
24:53
SPEAKER_05: And so when you have just a little bit of your own judgment and you can change the predictions, it kind of gives you this feeling of I'm not accepting that I'm going to be wrong every time.
25:05
SPEAKER_02: Then he separated the two things everybody smooshes together.
25:09
SPEAKER_02: This is the part that ties the whole quarter in a bow.
25:14
SPEAKER_02: Whether apps say 20% chance of rain during a concert you want to go to.
25:19
SPEAKER_02: That prediction can be flawless and still not tell you what to do.
25:24
SPEAKER_02: Because it does not know how much I hate being rained on.
25:27
SPEAKER_02: Or how much you want to see that band.
25:30
SPEAKER_02: Only you know that.
25:31
SPEAKER_05: Algos are the predictions, and humans are the decisions.
25:34
SPEAKER_05: So those predictions are just part of the decision.
25:36
SPEAKER_05: Then I need to integrate that prediction with how I feel about different outcomes.
25:40
SPEAKER_02: The machine can be better than you at predicting, but it's still your responsibility to decide what to do with that.
25:47
SPEAKER_02: Careers.
25:48
SPEAKER_02: Because everyone assumes there is one obvious way into this field.
25:52
SPEAKER_02: There is not.
25:54
SPEAKER_02: Not one of these six people started where they end up.
25:57
SPEAKER_02: A chemist, a physician, a financial manager, a historian, an activist teacher, and a woman who followed drivers into the app company.
26:07
SPEAKER_02: Zero of them plan this.
26:09
SPEAKER_02: So what do you do with that if you're 15 and everyone is asking what your thing is?
26:14
SPEAKER_02: That is the thing I want people to sit with.
26:17
SPEAKER_02: This field does not reward picking the right lane at 16.
26:21
SPEAKER_02: It rewards being curious about two things at once until they crash into each other.
26:26
SPEAKER_02: Dr.
26:27
SPEAKER_02: Bayant wanted to do foreign policy and diplomacy.
26:30
SPEAKER_02: He got to grade school right after the Iraq and Afghanistan war started.
26:35
SPEAKER_02: Everybody kept pushing him towards national security, which he did not really care about.
26:40
SPEAKER_02: So he took other classes and fell into microeconomics.
26:44
SPEAKER_02: A year working in Kenya after college, partly at a refugee camp, is what showed him what he was best at.
26:50
SPEAKER_02: He describes the whole path as surfing.
26:53
SPEAKER_02: You steer the board, the ocean is stronger than you.
26:57
SPEAKER_02: Aaron Klein?
26:58
SPEAKER_02: He was a high school kid at a beat up school in a rich county.
27:02
SPEAKER_02: Holes in the wall, no athletic field, no parking lot.
27:06
SPEAKER_02: The county could only afford to rebuild one school that year and they picked the wealthier one.
27:11
SPEAKER_02: And he organized.
27:20
SPEAKER_02: The vote flipped from five to four against to six to three in favor.
27:25
SPEAKER_02: The new building opened years after he graduated.
27:29
SPEAKER_02: His daughter goes there now.
27:31
SPEAKER_02: That whole fight is why he went into public service.
27:35
SPEAKER_02: And he was on the Senate Banking Committee before he turned twenty five.
27:39
SPEAKER_02: Alex Rosemand?
27:41
SPEAKER_02: History and sociology degrees.
27:43
SPEAKER_02: Seven years of field work with drivers.
27:45
SPEAKER_02: Then she went and worked inside the company she had spent seven years studying from the outside.
27:51
SPEAKER_02: She left and went back to investigating.
27:54
SPEAKER_03: So having experienced all of that and having written broadly about what it was like for drivers to have an algorithmic boss a couple of years later, I was looking at what else could I nowhere learn or do to try to improve the outcomes or experiences for drivers.
28:11
SPEAKER_02: And Jason Sadowski.
28:13
SPEAKER_02: He was halfway through a polymer chemistry degree at a tech school, working in the lab, fully on the scientist track since childhood.
28:21
SPEAKER_02: He just needed to fill an elective.
28:24
SPEAKER_02: One elective.
28:40
SPEAKER_06: And so I did.
28:41
SPEAKER_02: He calls those sliding door moments.
28:44
SPEAKER_02: And he is honest that without that one class, he would probably be in a lab right now, living an entirely different life.
28:52
SPEAKER_02: Kintaro Tayoma?
28:54
SPEAKER_02: Physicist at Harvard, then a computer scientist PhD at Yale, then Microsoft.
28:59
SPEAKER_02: He only stopped building AI and started studying its effects.
29:03
SPEAKER_02: After a decade of watching his own well-built projects die for reasons that had nothing to do with the code.
29:10
SPEAKER_02: And Berkeley Dietforced?
29:12
SPEAKER_02: Started as a financial major who thought predicting which stocks go up sounded amazing.
29:18
SPEAKER_02: Took a couple classes on human decision making by accident and realized what he loved was not making the predictions.
29:25
SPEAKER_02: It was figuring out how people make them at all.
29:28
SPEAKER_02: There's something that connects all six.
29:31
SPEAKER_02: They all found the crossing point between two things.
29:34
SPEAKER_02: Bayant is a would-be diplomat who became a field experimentalist.
29:39
SPEAKER_02: Klein is a teenage organizer who became a financial economist.
29:44
SPEAKER_02: Rosebland is a historian who became a technology investigator.
29:49
SPEAKER_02: Sidowski is a chemist who became a political economist.
29:54
SPEAKER_02: Tayoma is a physicist who became a critic in his own industry.
29:58
SPEAKER_02: Diavorced is a finance major who became a decision scientist.
30:03
SPEAKER_02: None of them are just one thing, and the reason they are valuable is exactly because they are not.
30:09
SPEAKER_02: Do not prune your interests too early.
30:11
SPEAKER_02: The combination that feels random right now may be the thing that makes you irreplaceable later.
30:18
SPEAKER_02: Concrete skills.
30:19
SPEAKER_02: What do you actually need to build?
30:21
SPEAKER_02: Dr.
30:21
SPEAKER_02: Bayant's answer is to go read real abstracts.
30:33
SPEAKER_04: Read every abstract.
30:35
SPEAKER_04: It'll probably take you 20 minutes.
30:37
SPEAKER_04: And I'm sure that several of them will jump out to you as like, ooh, that's kind of cool.
30:42
SPEAKER_02: And his surprise skill?
30:44
SPEAKER_02: Charm.
30:45
SPEAKER_02: He was careful about the word.
30:47
SPEAKER_02: What he meant is not being insufferable.
30:50
SPEAKER_02: He walks into a government office to tell someone their forms are too complicated.
30:54
SPEAKER_02: But at the end of the day, he knows less about that program than the person who runs it every day.
31:00
SPEAKER_02: And he said the more education people have, the worse they usually are at remembering that.
31:11
SPEAKER_01: Cold email people.
31:13
SPEAKER_01: Just reach out to people.
31:15
SPEAKER_01: People will respond.
31:16
SPEAKER_01: You reached out to me, I didn't know you, but you know, I said this is an opportunity.
31:20
SPEAKER_01: Uh, and I was impressed by the fact that you reached out to me.
31:24
SPEAKER_01: And so people and networks make a massive difference in your life.
31:29
SPEAKER_02: Which is a little funny given how this episode happened.
31:33
SPEAKER_02: It is the whole proof of the concept.
31:35
SPEAKER_02: I simply reached out.
31:37
SPEAKER_02: Alex Rosenblatt said, ethnography.
31:40
SPEAKER_02: Go sit with people a system happens to, instead of only measuring them from a spreadsheet.
31:46
SPEAKER_03: It's the fact of doing qualitative research that gives you a grounding for how big promises are actually landing on people affected by them, the distribution of harms amongst people who are affected by those same technologies, as well as the benefits.
31:59
SPEAKER_02: Jathan Sadoski said critical reasoning straight up.
32:03
SPEAKER_02: What does this do?
32:04
SPEAKER_02: How does it work?
32:05
SPEAKER_02: Who does it work for?
32:07
SPEAKER_02: Why does it exist?
32:09
SPEAKER_02: And his point was that the product asking for the most trust are specifically designs you never ask.
32:15
SPEAKER_02: And Kentarotoyama's answer cuts against everything else in this quarter.
32:20
SPEAKER_02: Do not assume the technology is the fix.
32:23
SPEAKER_02: Ask who has power over it and what they want.
32:26
SPEAKER_02: And he explained why that is so hard to hear from inside the industry.
32:31
SPEAKER_02: It is difficult to get a man to understand something when his livelihood depends on him not understanding it.
32:38
SPEAKER_02: And Berkeley Dietfors said read the actual research instead of someone's summary of it.
32:43
SPEAKER_02: And his surprise skill was the least cool one on this list.
32:48
SPEAKER_02: Time management.
32:49
SPEAKER_02: He said once you are a PhD student and then a professor, literally no one tells you what to do on any given day.
32:57
SPEAKER_02: The skill is taking something enormous and breaking it into things you can do this afternoon.
33:03
SPEAKER_02: Read carefully.
33:04
SPEAKER_02: Talk to real people.
33:05
SPEAKER_02: Ask who benefits before you believe the story.
33:08
SPEAKER_02: Aaron Klein said request your own free credit card report and read it, because the more common your name is, the more likely it is carrying someone else's mistake.
33:18
SPEAKER_02: He has an Aaron Klein in New Jersey who messed up his.
33:21
SPEAKER_02: He said open a bank or credit union account early and save something, even one or two percent because the habit matters more than the amount.
33:30
SPEAKER_02: That one is unusually actionable for a 17-year-old.
33:34
SPEAKER_02: Most career advice is not a thing you can do on the weekend.
33:38
SPEAKER_02: And Jaton Sadowski's test is the other one you can run today.
33:42
SPEAKER_02: Pick one app on your phone.
33:44
SPEAKER_02: Does it make my life better?
33:46
SPEAKER_02: Does it make society better off?
33:48
SPEAKER_02: That is the whole exercise.
33:50
SPEAKER_02: And now for what type of media our guest suggests if people want to learn more.
33:55
SPEAKER_02: Dr.
33:55
SPEAKER_02: Bayant said start with the journal of economic perspectives, any issue in the last twenty years.
34:01
SPEAKER_02: It is written for normal humans so you can follow the ideas without drowning in math.
34:07
SPEAKER_02: Find the topic that grabs you.
34:08
SPEAKER_02: Alex Rosemand said email her.
34:11
SPEAKER_02: She will send reading recommendation based on what you actually care about.
34:15
SPEAKER_02: And Uberland is the full story behind episode three.
34:20
SPEAKER_02: Jathan Sadowski pointed to his podcast, This Machine Kills, in both his books, which he said he read specifically so that people brand new to this stuff could get into it.
34:30
SPEAKER_02: Kintao Tayoma pointed to Evgeny Morozov's books for anyone who wants to go deeper on why technology alone does not save the world.
34:41
SPEAKER_02: And he says to Chase whatever you find genuinely interesting, because the world rewards experts and you may well be an expert in something you like.
34:50
SPEAKER_02: And Berkeley Dietforce said read research papers.
34:55
SPEAKER_02: He suggested accepting error to make less error and the robust beauty of the improper linear models and decision making.
35:04
SPEAKER_02: Bayant said if the choice environment is designed, you get to ask who designed it and why.
35:10
SPEAKER_02: Klein said a person decides to weight the algorithm that way and a decision can be challenging.
35:17
SPEAKER_02: Rosenblatt said a constrained choice is still a choice, and writing down what is happening is how it will eventually change.
35:25
SPEAKER_02: Sadaski said the does this make my life better test is something you can run right now, today on anything in your pocket.
35:33
SPEAKER_02: Toyama says amplification only pushes in the direction of the underlying systems, which means changing who holds power changes what gets amplified.
35:43
SPEAKER_02: And the Antforce said the machine can outpredict you and still will never make the decision for you.
35:50
SPEAKER_02: That part is yours.
35:52
SPEAKER_02: So the illusion is not that the choice is fake.
35:55
SPEAKER_02: The choice was never fully free and was never fully taken.
35:59
SPEAKER_02: There is a real decision buried in there somewhere, and finding it is the key.
36:04
SPEAKER_02: And that work is wide open.
36:08
SPEAKER_02: This has been passive curiosity in an AI world.
36:12
SPEAKER_02: I'm Acadia, and I'm Alora.
36:15
SPEAKER_02: Until next time, stay curious.