1. We have a new paper out in PNAS today, in which we address the harm wrought by dramatically restructuring human communication of the span of a decade, with no aim other than selling ads.

It might be the most important paper of my career.

pnas.org/content/118/27…
2. This thread describes the paper and the backstory leading to it. I'll be posting over the course of the day as I can find the time.

Three years ago, @uwcip postdoc @jbakcoleman organized a summer meeting at Princeton. I attended, and it changed the direction of my research.
3. Think about how you receive information today, compared to fifteen years or twenty years ago.

Social media, internet search, click-based advertising: innovations in information technology and new mechanisms for monetizing information have rewired human communication.
4. The problem is that this enormous transformation has taken place not as a stewarded effort to improve information quality or to further human well-being, but more or less simply for the purpose of keeping people engaged online in order to sell ads.
5. We don't have a theory for how human decision-making operates in an algorithm-driven online network of comprising billions of souls—and we argue there is no reason to expect some sort of invisible hand is going to bail us out and ensure that good information floats to the top.
6. It is difficult to overestimate the stakes. If these technologies so effectively sow mis- and disinformation worldwide, how we can hope to solve problems such as global warming, extinction, war, food security, pandemics? How can we prevent democracies from crumbling?
7. Aside: The paper has been in the works for over two years. A 2019 draft said something along the lines of "Imagine that pandemic hit and people wouldn't follow public health advice because of misinformation spread online. We'd be *really* screwed then."
8. So what is so radically different now compared to twenty years ago, and why does it matter? In the paper, we explore a few factors.

First, scale. We've gone from small face-to-face communities to a global network of 3.6 billion social media users in the blink of an eye.
9. We write "Expanding the scale of a collectively behaving system by eight orders of magnitude is certain to have functional consequences. Not only are societies at the scale of ours rare in the natural world; they also are often ecologically unstable where they do form."
10. Research in opinion dynamics, animal behavior, network epistemology, and statistical physics reveals that the ability to come to collective decisions depends strongly on the size of the group. Bigger does not mean wiser or better-functioning.
11. Second, network structure. Face-to-face networks limit the scope of influence; we can only talk to so many people in a day. On social media it's different. IRL I couldn't tell 200,000 people about this paper in an afternoon. More than that have clicked on this thread already.
12. Moreover, preferential attachment processes (often abetted by algorithms, more on that later) accelerate the inequality of influence online.

Network structure differs as well, with more "long ties" online that accelerate the spread of information and disinformation alike.
13. Third, the ease and fidelity afforded by digital communication. Online, messages can be forwarded, re-forwarded, and re-re-forwarded time and again without a loss of resolution, without breaking down in gibberish like in a kids' game of telephone.
14. All this takes place at essentially zero cost at staggering speed. Someone pens a piece of disinformation after lunch. It gets amplified a few times, and reaches e.g. the leader of the free world later that afternoon. He retweets it to 90 million people, who amplify further.
15. Suddenly a deception is cascading through every corner of the online world. It's even hard to triangulate to figure out if it's true, because by this point it's coming at you from all sides and seemingly from a plenitude of sources.

The very same day it was crafted.
16. Think about the downside of frictionless communication.

Why isn't your postal mailbox buried in 50 pounds of junk mail every day? Because stamps cost money.

Why didn't you get long-distance phone scammers in the 1990s? 25 cents a minute back then, that's why.
17. I want to correct a misperception I'm seeing. Our message is not "ads are evil."

It's that social media etc. been designed largely to sell ads, which means it is not designed with care to facilitate the spread of reliable information, let alone improve human well-being.
18. The fourth and final development we discuss is the role of algorithms and algorithmic feedback.

The posts you see on social media—including this thread, I fear—are fed to you by algorithms designed to maximize your engagement, not the veracity of the content you consume.
19. Perhaps even worse, who you *know* on social media is in some large part a function of whom algorithms wanted you to know.

You may like your online friends because they're cool or whatever, but you *met* many of them because some algorithm thought that would keep you online.
20. And these algorithms know so much about us. The amount of data is staggering. There are hundreds of simultaneous A-B experiments ongoing at any time to figure out what makes us click, collectively....
21. ...and to build up a detailed enough profile to predict what makes you click, specifically.

Sure, the results are sometimes risible. But sometimes they're not. Machine learning is a powerful beast when competently implemented and fed with an endless supply of data.
22. Algorithms create filter bubbles at times. They recommend that I connect with people who thinking like I do, and even if I seek out divergent viewpoints, other algorithms may learn that I don't engage with them and start to down-rank them in my feed.
23. Other times, algorithms may learn that agreement is boring and conflict is engaging. In the struggle for attention, righteous indignation outcompetes thoughtful discussion. So that's what we get. In this way they regulate the emotional tenor of the online worlds we inhabit.
24. Now add in the fact that these algorithms are opaque, sometimes even to their creators, and mercurial. We don't know what they're doing, or what effects they are having, because the only people who have the data to measure it consider that information too valuable to share.
25. "In sum, we are offloading our evolved information-foraging processes onto algorithms. But these algorithms are typically designed to maximize profitability, with often insufficient incentive to promote an informed, just, healthy, and sustainable society."

To be continued...
26. I'll finish up this thread tomorrow. In the meantime, check out lead author @jbakcoleman's feed and those of my coauthors.

@WolframBarfuss
@icouzin
@JonathanDonges
@andy_gersick
@jenniferjacquet
@albert_kao
@RachelEMoran
@kaiatombak
@jayvanbavel
@elkeweber
@PRomanczuk
27. In the paper, we argue that the human collective behavior, and the way that emerging communication technologies are changing it, needs to become a crisis discipline.

What is a crisis discipline?
28. In my view, a crisis discipline is one in which we have an urgent need to act to steer a complex system in ways that minimize harm, yet we lack a complete understand of how that system operates.

This was the case for climate change, for example.
29. We dealt with something like this around COVID in Feb/March 2020 as well. So many things we didn't know: what was the infection fatality rate, what was R0, how did it spread, etc. But (arguments from some prominent researchers notwithstanding) we couldn't wait to act.
30. COVID was spreading exponentially around the globe. We needed to make our best guesses about how to halt or at least slow that spread, and to act upon them immediately. At the same time, we needed to work at rapidly to answer the questions to which we lacked answers.
31. We don't have the theory in place to understand how the design of social networks drives network structure and function and how that in turn drives the flow of information or disinformation through communication networks. Yet we can't wait to figure it out.
32. We need to be acting and learning in parallel.

We also need all hands on deck. Crisis disciplines are radically transdisciplinary. We need to integrate a breadth of perspectives and we cannot predict from where the crucial insights will emerge to solve the problems we face.

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Carl T. Bergstrom

Carl T. Bergstrom Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @CT_Bergstrom

24 Jul
There's a new paper reporting on a randomized controlled trial of daily testing instead of quarantining for students exposed to Covid cases.

The study failed to find significant differences between the arms, even in the number of missed days of schools.

modmedmicro.nsms.ox.ac.uk/wp-content/upl… Image
OK fair enough, but look how the @guardian reports on it as if it demonstrated an effect.

theguardian.com/world/2021/jul… Image
How could @theguardian have gotten it so mixed up?

If you've been reading my feed for while, you know where I'm going.

The @UniofOxford press release. Look at this.

ox.ac.uk/news/2021-07-2… Image
Read 6 tweets
17 Jul
After Biden's criticism of @Facebook for their inaction on antivax propaganda, @fbnewsroom is going on the offensive.

They issued a press release that is a textbook example of bluster, obfuscation, and agnatogenesis. Here's the link. Let's take a look.
about.fb.com/news/2021/07/s…
"Facts, not allegations."

It would be nice to be able to rely on facts. Unfortunately, Facebook has exclusive access to the data we need to know what the facts are, and is not forthcoming with those facts. See e.g. the ongoing CrowdTangle debacle. nytimes.com/2021/07/14/tec… Image
"Vaccine acceptance among Facebook users in the US has increased."

This is a red herring, given that vaccine acceptance among everyone in the US has increased—dramatically. Look at the decline in those truly hesitant: the "wait and see" group. Source: kff.org/coronavirus-co… ImageImage
Read 24 tweets
12 Jul
Damn close to false lights by @TheSun.

The lede: expert is warning of 100K cases/day
First sentence: Carl Bergstrom says...

I'm getting hate mail. But I'm not the expert they refer to. And I never spoke to them. They're quoting an old, unrelated tweet.
thesun.co.uk/news/15556185/…
This indubitably creates the false impression that I am predicting 100,000 cases per day.

It's criminally poor journalism, and it leads to harassment as well as reputational harm.
Here's another version. What the hell are these people trying to do to me?
Read 6 tweets
7 Jul
1. Kevin Gross and I have worked together for a number of years, using mathematical models to understand how science works and how to make it work even better.

We just posted our latest paper to the arXiv.

arxiv.org/abs/2106.13282

Here is a short thread about that paper.
2. We begin from the perspective I describe on the page below: the norms and institutions that comprise science create incentives for researchers to pursue certain questions and research strategies, shaping in turn our scientific knowledge of the world.

ctbergstrom.com/understanding-…
3. In this paper, we look at the institution of peer review. Scientists confront peer review at multiple stages of a research project. Before doing a study, they face *ex ante* review when writing a grant proposal to persuade a funder that the research project merits funding.
Read 19 tweets
4 Jul
This morning we lost a great scientist and great human being.

I don't know quite how to speak to his memory just yet. There's no single story that captures what I would like to say, no single piece of his work that comes close to summarizing the whole.
He was a mentor to me as an undergraduate student. Over the past 18 months as I've struggled to navigate my dual role as scientist and citizen, I've thought of his words on a daily basis and tried to live up to his advice. He will be sorely missed.
I feel ready to write a bit about Dick's influence on my career.

As an undergrad, I was amazingly fortunate to take Bio 17, the evolution course that he taught with Stephen Jay Gould.
Read 19 tweets
30 Jun
Brilliant.

What other topics need to be on Andy's STATS 102 syllabus?
Probably a more advanced topic but

- Yes, it's counterintuitve. No, it's not Berkson's paradox.
- Unless you're a professional or moving on very soon, don't do your own electrical or your own stats.

h/t
Read 5 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!

Follow Us on Twitter!

:(