Profile picture
Arvind Narayanan @random_walker
, 7 tweets, 4 min read Read on Twitter
AI analog of asking women to smile: editing faces to make them look more/less smiling also made them look more feminine/masculine because of the correlations in the training data.
arxiv.org/abs/1609.04468
AI labels images of men cooking as women. homes.cs.washington.edu/~my89/publicat…
Commercial gender classifiers did 8%–21% better on male faces than female faces,12%–19% better on lighter faces than darker faces, and worst on darker female faces.
proceedings.mlr.press/v81/buolamwini…
English to Turkish and back to English, using Google Translate (and every other translation engine).
opus.bath.ac.uk/55288/4/Calisk…
Error rate of YouTube's automatic voice transcription varies based on speaker's dialect region and gender.
rachaeltatman.com/sites/default/…
AI-generated word analogies reflect societal stereotypes.
papers.nips.cc/paper/6228-man…
(There are tons of other examples of bias in computer vision and natural-language processing systems, but in this thread I've highlighted ones that are backed by research papers, as a starting point for a deeper exploration of the work happening in this space.)
Missing some Tweet in this thread?
You can try to force a refresh.

Like this thread? Get email updates or save it to PDF!

Subscribe to Arvind Narayanan
Profile picture

Get real-time email alerts when new unrolls are available from this author!

This content may be removed anytime!

Twitter may remove this content at anytime, convert it as a PDF, save and print for later use!

Try unrolling a thread yourself!

how to unroll video

1) Follow Thread Reader App on Twitter so you can easily mention us!

2) Go to a Twitter thread (series of Tweets by the same owner) and mention us with a keyword "unroll" @threadreaderapp unroll

You can practice here first or read more on our help page!

More from @random_walker see all

Profile picture
In 2018 the blockchain/decentralization story fell apart. For example, a study of 43 use cases found a 0% success rate. theregister.co.uk/2018/11/30/blo…
Let's talk about some mistaken assumptions about decentralization that led to the blockchain hype, and what we can learn from them.
(For context, I think the technology is sound, interesting and important from a CS perspective. That’s why I’ve been researching and teaching it since 2013. bitcoinbook.cs.princeton.edu But I’ve also been speaking/writing about the pitfalls of decentralization for ~10 years.)
Blockchain proponents have a vision of _society_ in which centralized entities are weakened/eliminated. But blockchain tech is a way to build _software_ without centralized servers. Why would the latter enable the former? It’s a leap of logic that’s left unexplained.
Read 14 tweets
Profile picture
Less than 5% of cryptographers work on creating/breaking encryption algorithms. Encryption isn't the most important or most fundamental cryptographic construction. It's also mostly a solved problem. Attackers rarely even try to break encryption; they just exploit software bugs.
Here are a few of the cool things that modern cryptographers work on.
1. Secure Multiparty Computation: a powerful & broad set of techniques that enables (e.g.) 2 hospitals to do a genetic study on their combined patient records—without either sending its records to the other.
2. Theory of cryptography: how do questions about computational complexity (such as P vs NP) relate to which cryptographic algorithms are or aren’t secure? This is best understood as a quest to reveal the computational laws of nature and isn't directly motivated by applications.
Read 6 tweets
Profile picture
The online ad tech industry has created an extraordinarily invasive surveillance infrastructure that threatens our democracy. But the most tragic thing is how little has been gained—how poorly ad tech works despite all the data.
Even if rampant ad fraud is curbed, there's a deeper reason why ad tech doesn't work: measuring the returns to advertising is a statistical problem that's near impossible even with perfect tracking. This neat paper shows why and deserves to be better known justinmrao.com/lewis_rao_near…
One reason why print/TV ads work—by conspicuously spending money, the firm signals to consumers that it is big & will stick around, is invested in its reputation, and has products worth advertising. Online ads don't achieve this. The more targeted the ad, the *weaker* the signal.
Read 3 tweets

Related threads

Profile picture
✨🧠 The ecosystem that has grown up around @TensorFlow in the last few years blows my mind. There's just so much functionality, compared to some of the other, newer frameworks.

👉Consider this an ever-expanding thread for me to take notes + wrap my brain around products. Ready?
1) @TensorFlow Extended (TFX)

It's no secret that I 💕 #TFX and all of its tooling for deploying machine learning models into production. If you care about keeping your models up-to-date and monitoring them, you should check out the product + its paper.

tensorflow.org/tfx/?hl=zh-cn
2) @TensorFlow Hub

If you want to train your model on a small data set, or improve generalization, you'll need to use something called transfer learning. #TFHub modules make it easy—and are available in an #OSS marketplace: tfhub.dev.

site: tensorflow.org/hub/
Read 39 tweets
Profile picture
1/ Who wants to hear some scientific intrigue?

A few weeks ago, a group of physical chemists posted a paper online announcing the observation of superconductivity at room temperature.

Today I posted a comment pointing out something funny in their data.
arxiv.org/abs/1808.02929
2/ Room-temperature superconductivity has been a holy grail in physics for literally over 100 years. If we could find a material that was superconducting at room temperature, it would allow us to transport electrical power for free, and would revolutionize a bunch of industries.
3/ There is no fundamental reason (that we know) why some material couldn't be superconducting at room temperature. But after a century of trying to find such material the best superconductor still needs to be cooled to 90 Kelvin (-183 Celsius).
Read 46 tweets
Profile picture
1/ What are #deepfakes + #AI generated synthetic media; what threats do they pose to #humanrights, #journalism and #publicsphere? What should we fear? What can we do? Thread w/ key tech + usages from recent expert mtg summarizing blog.witness.org/2018/07/deepfa…
2/We've had fake images and CGI for years, so what's different now (beyond enabling us to transplant Nicolas Cage's face to other people). The ease of making #deepfakes is captured by the vlogger Tom Scott
3/ = Barriers to entry to manipulate audio and video are falling - costs less, needs less tech expertise, uses open-source code + cloud computing power. Plus, sophistication of manipulation of social media spaces by bad actors makes it easier to weaponize and to micro-target.
Read 25 tweets
Profile picture
Deep reinforcement learning overfits. Often, a trained network can only play the particular level(s) you trained it on! In our new paper, we show how to train more general networks with procedural level generation, generating progressively harder levels.
arxiv.org/abs/1806.10729
The paper, written by @nojustesen @ruben_torrado @FilipoGiovanni @Amidos2006, me and @risi1979, builds on the General Video Game AI framework, which includes more than a hundred different games and lets you easily modify games and levels (and generate new ones).
We also build on ours and others' research on procedural content generation for games, a research field studying algorithms that can create new game content such as levels. Useful not only for game development but also for AI testing.
More on PCG:
pcgbook.com
Read 11 tweets
Profile picture
If you have an interest in #AI, #HealthTech or #PatientSafety - then please read this evidence based thread which tells the story of @DrMurphy11 & an #eHealth #AI #Chatbot.

Read on, or see single tweet summary here👇0/44
is not a 'nerd'; he's a pretty typical NHS consultant with an interest in #PatientSafety.

On 6th Jan 2017 a tweet about an #eHealth #AI #Chatbot in #NHS trials caught his attention, so he thought he'd take a look. 1/44

Evidence here 👇
Dr Murphy downloaded the @babylonhealth App & tried the #Chatbot with a few simple clinical presentations. It quickly became apparent that the #Chatbot had flaws, raising the question if the App had been validated as a triage tool?

As evidenced by👇 2/44
Read 45 tweets
Profile picture
Thread on interesting machine learning papers on Arxiv (1/x)
"Stochastic Video Generation with a Learned Prior": arxiv.org/abs/1802.07687… project: sites.google.com/view/svglp/ code: github.com/edenton/svg
"Can we steal your vocal identity from the Internet?: Initial investigation of
cloning Obama’s voice using GAN, WaveNet and low-quality found data": arxiv.org/abs/1803.00860…
Read 9 tweets

Trending hashtags

#deepfakes #Headache #BigBang #TFadvent #Creatability #statsmisconception #Ukraine #manufacturing #LAWS #openstandards #aii #dataprotection #NeurIPS2019 #HeartAttacks #autonomousAI #MachineLearning #fraud #atheists #IEEE #Babylon #TFHub #statsmisconceptions #smartciities #PatientSafety #PolicingInCollapse

Did Thread Reader help you today?

Support us! We are indie developers!

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

Become a Premium Member and get exclusive features!

Premium member ($3.00/month or $30.00/year)

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!