Blog

Jan 22, 2016 · interview

Eli Pariser on the Ethics of Algorithmic Filtering

image

Chiefly his reflection, of which the portrait / Is the reflection, of which the portrait / Is the reflection once removed. / The glass chose to reflect only what he saw / Which was enough for his purpose: his image / Glazed, embalmed, projected at a 180-degree angle.

— Parmagiano and John Ashbery - Self-Portrait in a Convex Mirror

Personalizing experiences and recommendations for consumers is the goal of many data science efforts, so much so that the NYC Media Lab created Personalizationpalooza to unite technicians and media leadership around the topic. And while studies have shown that recommendation algorithms promote happy consumers, it’s questionable that also they promote healthy citizens.

Zeynep Tufecki, for example, recently wrote a Medium post explaining how Facebook’s focus on “engagement” as a metric for deciding which content to share on user news feeds can happen “to the detriment of the substantive experiences and interactions” she (and others) wants on Facebook. Zufecki illustrates her point with an instance where users felt morally unable to “like” a video with child refugees surrounded by dead bodies. She argues that the platform needs different mechanisms and metrics to ensure that content worth viewing can compete with the content we’re more apt to “like.”

Tufecki is not alone in her concerns. In 2011, Eli Pariser, CEO of Upworthy, wrote the first monograph about the risks algorithmic filtering poses to citizenship and society. The Filter Bubble is still worth reading today, as it provides a framework for understanding how filtering algorithms work and what personal, social, and political consequences they may have. It’s critical we understand these tools now that the 2016 election season is in full swing.

We recently interviewed Pariser to see how his thinking has evolved in light of recent algorithmic progress. Keep reading for highlights.

Upworthy has received a lot of attention for using clickbait to engage readers. While you recently pivoted your content strategy (and redesigned your logo), what problem was the site originally trying to solve?

A hallmark of a critic is the ability to point out a problem but not propose a good solution. After publishing The Filter Bubble, I was invited all over to give talks about how personalized media techniques generate hollow models of individuals, generating the risk that people only see banal content as opposed to the more important things they aspire to read to play their part as citizens. I didn’t want to be one of those annoying people who can only point out a problem. I wanted to actually work on a solution. So, I got together with Peter Koechley, at the time an editor at the Onion, and we decided to collaborate on a project to repurpose the algorithms used to curate and spread highly shareable content on topics people should read to make good decisions and be good citizens. Upworthy was the result.

What’s the new strategy?

We have our own proprietary content management system and collect enormous amounts of data each month on user behavior to track and understand how people engage with content. With storage so cheap and computing so fast, it’s easy for us to collect an enormous amount of data. The hard part is making that data actionable, especially for our creative staff, who are storytellers rather than quants. So our focus is on harvesting behavioral data as insights our writers can use to tell better stories. Think of making writing for the web like standup comedy: you put out a riff, see whether people laugh, and refine it to make more people laugh in the future.

What distinguishes Upworthy from comparable sites?

Our focus on promoting engagement and sharing enables us to write a fraction of stories comparable sites promote. We reach about 100 million people per month with only about 200 stories. Other sites of a similar size and user base generate thousands of pieces of content per month to reach the same audience.

What is the concern about personalization algorithms you addressed in The Filter Bubble?

There is way more information on the internet than any of us could ever read and way more products than we could ever buy. As such, personalization algorithms play an important role, but the problem is that you can end up in an uncanny valley of the self, where the model guiding personalization represents not the full you with all your interests, but only a bad 2-D version of you. Take Netflix as an example. Say you sign up and watch two romantic comedies. You may also love Italian neorealism, but the Netflix recommender system will base its representation of what you like on the data it has. This limited representation then gets amplified through future recommendations. I called this a “you loop” in the book.

What are the implications beyond repetitive recommendations?

For one, we can generate a false impression that we have the ability to make informed choices. The core issue with personalization algorithms is that they shape not only what you see, but also what you don’t see. When we do something as minor as watch a movie or buy a product, or as major as vote for president, we must be aware of the fact that, by simply trusting algorithms, we can’t measure the degree to which we base choices on the full set of available options.

Facebook published an article in Science in May 2015 that drew a large critical response from the community, including critiques from Zeynep Tufecki and Christian Sandvig. The Facebook article argued that, “compared to algorithmic ranking, individuals’ choices about what to consume had a stronger effect limiting exposure” to content with opinions against a user’s stated ideological affiliations. Where do you stand on that?

I agree with Tufecki that personal choice compounds rather than exculpates algorithmic filtering. Sure, beliefs influence what people are interested in, and, in turn, what they click on; but you have to be a very determined reader to dig beyond the limited range of options presented by the tools. The result is that most of us are then exposed to a doubly small range of content. It’s our right as consumers to ask ourselves how much agency we want to make choices about what we see and don’t see.

image
A chart from the appendix of the Facebook report, showing that the higher an item is in the newsfeed, the more likely it is clicked on.

If personalization risks exacerbating political polarization, where can constructive cross-partisan exchange occur online?

One of my favorite discoveries while researching The Filter Bubble was that constructive cross-partisan conversations occurred in two unlikely places: message boards for sports teams and fan space for the television series Lost. The fact that people shared something in common, e.g., they were all Giants fans, created a trusted space for open disagreement about race, class, and other delicate social issues. When united by a common reference point, people are more apt to engage with ideas than in an adversarial, debate-focused context.  

What can technologists do to address issues created by algorithmic filtering?

There are two main things: 1) accepting that behavioral signals are insufficient to explain a person, and working to increase our understanding of human psychology, and; 2) building recommendation systems that try to account for people’s aspirational desires as well as their current behavior. That people click more frequently on celebrity stories than political essays does not mean they don’t want to learn about what’s going on in world. One thing that’s interesting about Facebook’s models of the self, versus Google’s model of the self, is that our Facebook identities are somewhat aspirational. Some dismiss our socially-doctored self as inauthentic. I like to celebrate it as aspirational.

What can consumers do to escape algorithmic filter bubbles?

Marshall McLuhan was prescient (side note: we love this classic Annie Hall scene). The first step is for consumers is to become more conscious about the interplay between the medium and message, to accept that our actions on popular websites shape what future stories we’ll see and won’t see. Only then can we learn the basic algorithmic rules and use them to our advantage. In the 70s and 80s, a generation of people figured out how news editors think and why certain stories made first-page headlines to the exclusion of others. Our generation’s task is to understand how people writing algorithms think and what metrics they’re using to promote one story over another. And then, to make the effort to read things of less interest to our present self than to the future self we aspire to become.

(For those interested in art history and early modern optics, Tim’s Vermeer is a fantastic study of the techniques Vermeer may have used to create his works.)

- Kathryn 

Read more

Newer
Feb 3, 2016 · post
Older
Jan 12, 2016 · announcement

Latest posts

Nov 15, 2022 · newsletter

CFFL November Newsletter

November 2022 Perhaps November conjures thoughts of holiday feasts and festivities, but for us, it’s the perfect time to chew the fat about machine learning! Make room on your plate for a peek behind the scenes into our current research on harnessing synthetic image generation to improve classification tasks. And, as usual, we reflect on our favorite reads of the month. New Research! In the first half of this year, we focused on natural language processing with our Text Style Transfer blog series.
...read more
Nov 14, 2022 · post

Implementing CycleGAN

by Michael Gallaspy · Introduction This post documents the first part of a research effort to quantify the impact of synthetic data augmentation in training a deep learning model for detecting manufacturing defects on steel surfaces. We chose to generate synthetic data using CycleGAN,1 an architecture involving several networks that jointly learn a mapping between two image domains from unpaired examples (I’ll elaborate below). Research from recent years has demonstrated improvement on tasks like defect detection2 and image segmentation3 by augmenting real image data sets with synthetic data, since deep learning algorithms require massive amounts of data, and data collection can easily become a bottleneck.
...read more
Oct 20, 2022 · newsletter

CFFL October Newsletter

October 2022 We’ve got another action-packed newsletter for October! Highlights this month include the re-release of a classic CFFL research report, an example-heavy tutorial on Dask for distributed ML, and our picks for the best reads of the month. Open Data Science Conference Cloudera Fast Forward Labs will be at ODSC West near San Fransisco on November 1st-3rd, 2022! If you’ll be in the Bay Area, don’t miss Andrew and Melanie who will be presenting our recent research on Neutralizing Subjectivity Bias with HuggingFace Transformers.
...read more
Sep 21, 2022 · newsletter

CFFL September Newsletter

September 2022 Welcome to the September edition of the Cloudera Fast Forward Labs newsletter. This month we’re talking about ethics and we have all kinds of goodies to share including the final installment of our Text Style Transfer series and a couple of offerings from our newest research engineer. Throw in some choice must-reads and an ASR demo, and you’ve got yourself an action-packed newsletter! New Research! Ethical Considerations When Designing an NLG System In the final post of our blog series on Text Style Transfer, we discuss some ethical considerations when working with natural language generation systems, and describe the design of our prototype application: Exploring Intelligent Writing Assistance.
...read more
Sep 8, 2022 · post

Thought experiment: Human-centric machine learning for comic book creation

by Michael Gallaspy · This post has a companion piece: Ethics Sheet for AI-assisted Comic Book Art Generation I want to make a comic book. Actually, I want to make tools for making comic books. See, the problem is, I can’t draw too good. I mean, I’m working on it. Check out these self portraits drawn 6 months apart: Left: “Sad Face”. February 2022. Right: “Eyyyy”. August 2022. But I have a long way to go until my illustrations would be considered professional quality, notwithstanding the time it would take me to develop the many other skills needed for making comic books.
...read more
Aug 18, 2022 · newsletter

CFFL August Newsletter

August 2022 Welcome to the August edition of the Cloudera Fast Forward Labs newsletter. This month we’re thrilled to introduce a new member of the FFL team, share TWO new applied machine learning prototypes we’ve built, and, as always, offer up some intriguing reads. New Research Engineer! If you’re a regular reader of our newsletter, you likely noticed that we’ve been searching for new research engineers to join the Cloudera Fast Forward Labs team.
...read more

Popular posts

Oct 30, 2019 · newsletter
Exciting Applications of Graph Neural Networks
Nov 14, 2018 · post
Federated learning: distributed machine learning with data locality and privacy
Apr 10, 2018 · post
PyTorch for Recommenders 101
Oct 4, 2017 · post
First Look: Using Three.js for 2D Data Visualization
Aug 22, 2016 · whitepaper
Under the Hood of the Variational Autoencoder (in Prose and Code)
Feb 24, 2016 · post
"Hello world" in Keras (or, Scikit-learn versus Keras)

Reports

In-depth guides to specific machine learning capabilities

Prototypes

Machine learning prototypes and interactive notebooks
Notebook

ASR with Whisper

Explore the capabilities of OpenAI's Whisper for automatic speech recognition by creating your own voice recordings!
https://colab.research.google.com/github/fastforwardlabs/whisper-openai/blob/master/WhisperDemo.ipynb
Library

NeuralQA

A usable library for question answering on large datasets.
https://neuralqa.fastforwardlabs.com
Notebook

Explain BERT for Question Answering Models

Tensorflow 2.0 notebook to explain and visualize a HuggingFace BERT for Question Answering model.
https://colab.research.google.com/drive/1tTiOgJ7xvy3sjfiFC9OozbjAX1ho8WN9?usp=sharing
Notebooks

NLP for Question Answering

Ongoing posts and code documenting the process of building a question answering model.
https://qa.fastforwardlabs.com

Cloudera Fast Forward Labs

Making the recently possible useful.

Cloudera Fast Forward Labs is an applied machine learning research group. Our mission is to empower enterprise data science practitioners to apply emergent academic research to production machine learning use cases in practical and socially responsible ways, while also driving innovation through the Cloudera ecosystem. Our team brings thoughtful, creative, and diverse perspectives to deeply researched work. In this way, we strive to help organizations make the most of their ML investment as well as educate and inspire the broader machine learning and data science community.

Cloudera   Blog   Twitter

©2022 Cloudera, Inc. All rights reserved.