PubHawk
Science April 2017 PMID 28408601

Semantics derived automatically from language corpora contain human-like biases.

Aylin Caliskan, Joanna J Bryson, Arvind Narayanan

Full text PDF Download social card

Mentions

218
Peak: 4 in Mar 2022
across 6 sources since Oct 2021

Extended reach

460,915
potential audience

Total likes

837

Video views

8,390

Mentions per month

Oct 2021 – Nov 2024

Where it's discussed

  • News125 articles
  • Blogs45 articles
  • X / Twitter30 posts
  • Facebook12 posts
  • Reddit4 posts
  • YouTube2 videos

AI synopsis

What people are saying

The discussions reflect a recognition of the article’s significance by referencing related studies, such as designings of bias research or the reflection of societal prejudices in AI models. Words like 'classic,' 'demonstrated,' and mentions of methodology emphasize the analytical tone, while engaging examples like autonomous cars and word embeddings highlight real-world relevance.

AgreementModerate agreement

Most discussions acknowledge the validity of the publication's findings, emphasizing the existence of human-like biases in language models.

InterestModerate interest

Participants show moderate curiosity, engaging with topics like implicit bias and implications for AI systems without deep emotional investment.

EngagementModerate engagement

The discourse includes references to studies, methodologies, and personal experiences, indicating a reasonable level of thoughtful engagement.

ImpactModerate impact

Multiple comments suggest the article's importance in shaping understanding of AI biases and influencing future research or application, signaling moderate impact.

Abstract in brief

  • Machine learning models that analyze language can automatically uncover humanlike biases present in large text datasets from the internet.
  • These biases include neutral ones, like those related to insects or flowers, as well as problematic prejudices related to race, gender, or societal roles.
  • The findings suggest that language data reflects historical and cultural biases, offering a way to identify and possibly address these biases in technology and society.]
Read the full abstract

Top posts

Most liked on X and Bluesky
Deb Raji
@rajiinio · X

@DrDesmondPatton Yeah, I think this is a classic text on this (by @aylin_cim, @j2bryson & @random_walker): https://t.co/rwoMVjhelj

Apr 21, 2022 · 13 likes · 30,112 followers View post →
Joanna J Bryson
@j2bryson · X

@GaryMarcus I guess this is yet another thing that my paper with @aylin_cim & @random_walker showed – mining language doesn't give you just any experience, it gives you the lived experience of those who produced it, expressed as knowledge including implicit biases. https://t.co/xShhDfDMcI

Apr 3, 2023 · 4 likes · 2 shares · 28,701 followers View post →
Joanna J Bryson
@j2bryson · X

@SteveStuWill @jonathanstray I designed our 2017 #aibias study knowing #implicitBias was one of the largest effects discovered in psychology. We now demonstrated its reality with an entirely different methodology, yet it’s STILL under attack. https://t.co/xShhDfDMcI

Dec 30, 2022 · 4 likes · 2 shares · 28,701 followers View post →
@timnitGebru (@dair-community.social/bsky.social)
@timnitGebru · X

@DrDesmondPatton @teemu_roos @emilymbender @moinnadeem https://t.co/CegvzedVPZ from 2017.

Apr 21, 2022 · 3 likes · 163,124 followers View post →
Rineke Verbrugge
@RinekeV · X

@j2bryson Yes, I was thinking of that paper too. For those who haven't read it, see https://t.co/wXgprOF8px Looking forward to @j2bryson's talk at #HHAI2022 tomorrow!

Jun 14, 2022 · 2 likes · 1,146 followers View post →
Joanna J Bryson
@j2bryson · X

@kokkonutter @jpskeete I don't know if this makes it less uncomfortable, but after doing my work on the implicit biases and seeing their scale, I think this is not necessarily all explicit, intentional beliefs. So double blinding helps us be our best selves. https://t.co/xShhDfDMcI

Apr 10, 2023 · 1 like · 28,701 followers View post →

All mentions

Masataka Nakayama
@p_grin_rin · X

https://t.co/itzEhnga5O

Nov 11, 2023 · 268 followers View post →
salumc
@salumc · X

https://t.co/l1e4DKMhIn

Sep 6, 2023 · 4,643 followers View post →
kaveinology
@kaveinthran · X

RT @j2bryson: @SteveStuWill @jonathanstray I designed our 2017 #aibias study knowing #implicitBias was one of the largest effects discovere…

Sep 1, 2023 · 2 shares · 1,113 followers View post →
Joanna J Bryson
@j2bryson · X

@kokkonutter @jpskeete I don't know if this makes it less uncomfortable, but after doing my work on the implicit biases and seeing their scale, I think this is not necessarily all explicit, intentional beliefs. So double blinding helps us be our best selves. https://t.co/xShhDfDMcI

Apr 10, 2023 · 1 like · 28,701 followers View post →
Rito
@AllesistKode · X

RT @j2bryson: @GaryMarcus I guess this is yet another thing that my paper with @aylin_cim & @random_walker showed – mining language doesn't…

Apr 3, 2023 · 2 shares · 1,202 followers View post →
Rebekah Wegener
@rebekahwegener · X

RT @j2bryson: @GaryMarcus I guess this is yet another thing that my paper with @aylin_cim & @random_walker showed – mining language doesn't…

Apr 3, 2023 · 2 shares · 933 followers View post →
Joanna J Bryson
@j2bryson · X

@GaryMarcus I guess this is yet another thing that my paper with @aylin_cim & @random_walker showed – mining language doesn't give you just any experience, it gives you the lived experience of those who produced it, expressed as knowledge including implicit biases. https://t.co/xShhDfDMcI

Apr 3, 2023 · 4 likes · 2 shares · 28,701 followers View post →
Emiel van Miltenburg
@evanmiltenburg · X

@HadasKotek I was also thinking you could try using the same approach as with word embeddings but then using sentence embeddings. But that sounds so simple someone must have done it already. (So similar to https://t.co/K3OiLgn3bh)

Mar 30, 2023 · 2,194 followers View post →
Boots Whitlock
@BootsWH · X

@Tesla @elonmusk This is 1st principles woke. It's a good perspective. If you care about safety you'll pay attention to this. https://t.co/8b98v47F4R

Mar 24, 2023 · 1 like · 1,748 followers View post →
Amit Sharma
@amitsharmalie · X

Are autonomous cars putting people's lives at risk? https://t.co/oomSSibyvM via @YouTube

Mar 24, 2023 · 42 followers View post →
Joanna J Bryson
@j2bryson · X

@marksongs22 @lizweil More history: my most cited article, allegedly on #AIBias, actually came from that semantics, cognitive science, & evolution-of-cognition research stream, NOT from my #AIEthics research stream. I was glad @ScienceMagazine spotted that & called it cog sci! https://t.co/xShhDfDMcI

Feb 7, 2023 · 28,701 followers View post →
4annegs
@4annegs · X

RT @j2bryson: @SteveStuWill @jonathanstray I designed our 2017 #aibias study knowing #implicitBias was one of the largest effects discovere…

Dec 30, 2022 · 2 shares · 23,980 followers View post →
Joanna J Bryson
@j2bryson · X

@SteveStuWill @jonathanstray I designed our 2017 #aibias study knowing #implicitBias was one of the largest effects discovered in psychology. We now demonstrated its reality with an entirely different methodology, yet it’s STILL under attack. https://t.co/xShhDfDMcI

Dec 30, 2022 · 4 likes · 2 shares · 28,701 followers View post →
Joanna J Bryson
@j2bryson · X

@jadelgador https://t.co/xShhDfDMcI

Aug 5, 2022 · 1 like · 28,701 followers View post →
Joanna J Bryson
@j2bryson · X

@mlamons1 @1Br0wn No, it's just from our society. AI trained on the open Internet matches human implicit biases. https://t.co/xShhDfDMcI or for the blog version https://t.co/MOOkByetcA

Jun 22, 2022 · 28,701 followers View post →
Joanna J Bryson
@j2bryson · X

@MCoeckelbergh @Google @cajundiscordian @Floridi @David_Gunkel sorry, this should read “speaking of ‘bias’” or “speaking of prejudice”, to be consistent with one of my just earlier tweets to it, and the definitions we learnt about from reviewers & then explained here: https://t.co/xShhDfDMcI https://t.co/irjj4kGj3f

Jun 16, 2022 · 28,701 followers View post →
Rineke Verbrugge
@RinekeV · X

@j2bryson Yes, I was thinking of that paper too. For those who haven't read it, see https://t.co/wXgprOF8px Looking forward to @j2bryson's talk at #HHAI2022 tomorrow!

Jun 14, 2022 · 2 likes · 1,146 followers View post →
Deb Raji
@rajiinio · X

@DrDesmondPatton Yeah, I think this is a classic text on this (by @aylin_cim, @j2bryson & @random_walker): https://t.co/rwoMVjhelj

Apr 21, 2022 · 13 likes · 30,112 followers View post →
@timnitGebru (@dair-community.social/bsky.social)
@timnitGebru · X

@DrDesmondPatton @teemu_roos @emilymbender @moinnadeem https://t.co/CegvzedVPZ from 2017.

Apr 21, 2022 · 3 likes · 163,124 followers View post →
Tobias Martens
@tbsmartens · X

https://t.co/L1Xf3OrmQ9

Mar 30, 2022 · 937 followers View post →
Paul Smaldino
@psmaldino · X

@acerbialberto OK, yeah, it looks more along the lines of this paper that found that algorithms reflect the biases in human-produced content. https://t.co/hioziA0YtE

Mar 14, 2022 · 1 like · 7,301 followers View post →
Paul Smaldino
@psmaldino · X

@vlasceanu_mada @david_m_amodio This by @j2bryson and colleagues seems relevant. https://t.co/hioziA0YtE

Mar 14, 2022 · 1 like · 7,301 followers View post →
Yuanye Ma
@yuanye0111 · X

Semantics derived automatically from language corpora contain human-like biases https://t.co/FzuAtTZsbT

Mar 8, 2022 · 57 followers View post →
JJ Bryson 2
@j2blather · X

@AutoArtMachine @tweetycami @johnchavens @StanfordHAI @StanfordSML @Stanford No, just perfectly replicating them is common https://t.co/stFG5l0KDU

Feb 20, 2022 · 1 like · 1,781 followers View post →
Jinhee Kim
@TheJinheeKim · X

@ShannonVallor Not from 2020, but it got me into AI Ethics: https://t.co/2CTZrsFIq2

Dec 6, 2021 · 28 followers View post →
IPman
@IP_ad_man · X

RT @Andrea_ilsergio: Semantics derived automatically from language corpora contain human-like biases https://t.co/4xoe1A8jgu

Nov 25, 2021 · 2 shares · 494 followers View post →
Sir James Delon
@SirJamesDelon · X

RT @Andrea_ilsergio: Semantics derived automatically from language corpora contain human-like biases https://t.co/4xoe1A8jgu

Nov 25, 2021 · 2 shares · 200 followers View post →
Andrea Sergiacomi
@Andrea_ilsergio · X

Semantics derived automatically from language corpora contain human-like biases https://t.co/4xoe1A8jgu

Nov 25, 2021 · 2 shares · 1,084 followers View post →
Otto Koppius
@OKoppius · X

RT @EvilAICartoons: Note that using representative data is not always a good idea, as shown by @aylin_cim @j2bryson @random_walker in the…

Oct 7, 2021 · 1 share · 55 followers View post →
Evil AI Cartoons
@EvilAICartoons · X

Note that using representative data is not always a good idea, as shown by @aylin_cim @j2bryson @random_walker in the context of language models—used in chatbots and language translation tools— https://t.co/1TAsNdUatC

Oct 7, 2021 · 1 share · 1,836 followers View post →