Video: My keynote presentation on "Human Values and AI" at this year's AMLD:
with a brief history on how data and human values have come together to build AI
how AI can help expand/enrich data
how AI can help social scientists understand human values
Evidence supports this policy. When state supreme courts introduced mandatory retirement (age-based term limits) in recent decades, performance of the court (quantity and quality of rulings) went up.
Paper link:
and coming soon to
@AEAjournals
Policy
Recently posted at
@cepr_org
: "Text Algorithms in Economics" with
@StephenEKHansen
(forthcoming in ARE
@AnnualReviews
).
Ungated version:
Companion code notebooks:
& now a 🧵 on the highlights...
In its meeting on December 6/7, the ETH Board
@ETH_Rat
formally granted tenure to Elliott Ash
@ellliottt
@ETH_en
. We are proud to have Elliott and his entire team
@eth_cle
. Please join us in congratulating Elliott on this well-deserved achievement! 🎆🎆
An intriguing paper by Elliott Ash, Sharun Mukand and
@danilaserra_eco
uses data on political ads to show that “idea-based” politics (e.g., identity politics) becomes more important following negative economic shocks, such as surgeon imports from China:
Counterpoint: complex laws may actually be better laws.
More elaborate state laws boost economic growth. Why? They mean more careful consideration & rules for key edge cases, rather than writing shorter, but broader, laws that lump everything together.
Now hiring: Predoc positions in AI+Economics at
@ETH_en
Zurich -- work on exciting projects at this new intersection of fields.
Submit an application here by Nov 22nd:
Also reviewing applications for PhD students and postdocs.
@econ_ra
AI+Econ = 🤩
The 2nd AI+Econ workshop was a smashing success. Thanks to all the presenters, discussants, and others who made it happen!
See you next year for round 3
“In politics, when reason and emotion collide, emotion invariably wins.” -- Drew Westen, The Political Brain.
Interesting idea! What do the data say about that?
New working paper: "Emotion and Reason in Political Language", with
@gloriagennaro
.
PDF:
With
@sergallet
, Dominik Hangartner, Yotam Margalit, and
@matteopins
, our new working paper presents more compelling evidence on how
@FoxNews
sowed skepticism and slowed responses to the COVID-19 pandemic:
(thread)
AI+Econ = 🤩
The 2nd AI+Econ workshop was a smashing success. Thanks to all the presenters, discussants, and others who made it happen!
See you next year for round 3
Published Paper Summary Thread!
“Cross-Domain Topic Classification for Political Texts”
with
@osnabruegge
and Massimo Morelli
now Open Access at
@polanalysis
paper:
code:
This stellar work by
@BenjaminArold
is now his JMP for the upcoming season!
New version:
(revised for resubmission at QJE 🤩)
More at Ben's website:
including new projects and teaching on NLP/ML for econ.
📢New WP🚨
“Evolution vs. Creationism in the Classroom:
The Lasting Effects of Science Education”
Key result: Evolution teaching affects students` related knowledge, adulthood beliefs, and life choices.
More here:🧵
👉
1/8
Schedule Now Online!
4th Monash-Warwick-Zurich Text-as-Data Workshop:
Spanning time zones globally:
Session 1, Sept 1st Morning Europe, Evening Asia
Session 2, Sept 1st Morning USA, Evening Europe
Session 3, Sept 2nd, Morning Europe, Evening Asia
⭐The 8th
@MonashWarwick
@ETH
CEPR Text-as-Data Workshop will take place on 16-17 Sept on Zoom.
There are sessions both in the morning European time (evening in Australia) and in the afternoon European time (morning in the US).
Details and register here:
Turns out
#gpt3
is not so bad at writing abstracts for economics papers.
See the human-generated paper (“Text Algorithms in Economics”) and machine-generated abstract here:
Example code and other supporting material to be posted soon!
Check out the newly posted JMP by Germain Gauthier (
@PinchOfData
)
"Measuring crime prevalence and reporting: Method and application to
#MeToo
"
An important entry for social science and for crime policy. (1/3)
As part of the new version of “In-group Bias in the Indian judiciary”, we open-sourced our name classifier, which automatically assigns gender based on Indian names.
Try the web app here:
This will be very cool! We are training an AI to read in a paper and convert it to slides. To help us do that, we are crowd-sourcing the source files for papers and slides. See here:
A conference is coming up, and you need new slides? What if you had a digital assistant to convert your paper into a (draft) slide set?
A research team at ETH Zurich has been working on a tool to convert articles to slides.
To fine-tune the prototype, we need your help!
(1/2)
👉This year's Workship in AI+Economics, coming to you December 6th-7th, 2024, in Zurich🇨🇭👈
co-organized with
@sergallet
&
@YanagizawaD
CFP:
(submit by 𝗝𝘂𝗹𝘆 𝟭𝟵𝘁𝗵)
Keynote:
@MelissaLDell
Pics from 2023:
📢Call For Papers: 3rd Zurich Workshop in AI+Economics📢
Conf: Dec 6th-7th, 2024 in Zurich
Papers due: July 19th, 2024
Keynote by:
@MelissaLDell
Submit here:
More details:
co-organized with
@ellliottt
@YanagizawaD
Are you a social scientist working with text-as-data and/or other unstructured data? Present your work at the next Monash-Warwick-Zurich Text-As-Data Workshop on April 15-16, 2024! We are already at the 7th edition of the MWZ Workshop 🥳 Submit here:
Requesting JMPs on Text Data in Economics / Social Science
Reply / email with a link for a share here, an invite to an upcoming workshop, and maybe a spot on a future syllabus!
#EconTwitter
Check out this new prof position at
@ETH_en
-- both "data science" and "international relations" broadly construed!
Applications welcome from poli sci / economics / related fields.
A highlight of this week's MWZ text-as-data workshop, the discussion by
@akorinek
and
@WayneGeerling
on what AI assistants like GPT mean for economics -- research and teaching:
Thanks to
@essobecker
for moderating!
#EconTwitter
"What is this about?"
... is the starting question in reading a text, whether the reader is human or machine.
To help machines do this, we introduce DocSCAN, an NLP tool that can assign topics to unlabeled documents!
paper:
code:
Monash-Warwick-Zurich Text-As-Data Conference, Second Edition
Schedule Now Online
August 30th-31st, 2021 (on Zoom)
Register:
(co-organized with
@essobecker
)
Now online! Our paper (with Matia Vannoni and Massimo Morelli) in political analysis using text analysis to extract legal provisions and measure delegation in U.S. state legislation.
#FirstView
from
@polanalysis
-
Measuring Discretion and Delegation in Legislative Texts: Methods and Application to US States -
- Matia Vannoni,
@ellliottt
& Massimo Morelli
Using data on
#US
states from 1965 to 2012, Massimo Morelli
@Unibocconi
& Matia Vannoni
@Kingspol_econ
argue that
#regulation
may be good or bad for the
#economy
depending on its type and the information and incentives of the regulators
Happy to announce a new version of working paper with
@sergallet
and
@T_Giommoni
-- using ML to analyze and support anti-corruption policy -- with new policy simulations on politically neutral audit targeting.
Link to paper:
Excited to share that I will join
@Cambridge_Uni
, Econ Dep, this Sept as Assistant Professor!
Looking forward to working with the⭐️economists there!
I am forever grateful to my amazing supervisors
@ellliottt
Ash, Ludger
@Woessmann
and Davide Cantoni. You are my role models! 🙏
📢Call For Papers: 3rd Zurich Workshop in AI+Economics📢
Conf: Dec 6th-7th, 2024 in Zurich
Papers due: July 19th, 2024
Keynote by:
@MelissaLDell
Submit here:
More details:
co-organized with
@ellliottt
@YanagizawaD
Very cool to hear a
@planetmoney
episode about our paper “Ideas have Consequences: The Impact of Law and Economics on American Justice” with
@snaidunl
and
@Daniel_L_Chen
.
Read the original here:
It seemed like a crazy idea at the time. Convince a bunch of judges to go on a swanky retreat ... to learn economics?
But by the 1990s, some 40 percent of federal judges had done the program — and it may have forever changed the law. Our latest episode:
Request: Links to job market papers (in economics, political science, etc) using text data or machine learning, for a Zurich reading group series -- authors will receive constructive feedback!
Forthcoming in AEJ: Applied Economics: "Labor Supply and Directed Technical Change: Evidence from the Termination of the Bracero Program in 1964" by Shmuel San.
Now accepting PhD and postdoc applications for transdisciplinary project making scientific findings more useful.
See this video for an example of our data-centric approach:
Hi all! We have open PhD and Postdoc positions on using AI to improve scientific communication and combat misinformation. Positions are part of an interdisciplinary collaboration between researchers at ETH Zurich (
@mrinmayasachan
@ellliottt
) and UZH (Markus Leippold).
#NLP
Link to paper — “Heroes, villains, and victims, and GPT-3: Automated extraction of character roles without training data”!
With
@maria_antoniak
and
@dominik61256191
Requesting JMPs on Text Data in Economics / Social Science
Reply / email with a link for a share here, an invite to an upcoming workshop, and maybe a spot on a future syllabus!
#EconTwitter
Thanks to
@DinaPomeranz
and co organizers for inviting
@sergallet
to present “A machine learning approach to analyze and support anti-corruption policy” (also with
@T_Giommoni
)!
Link to paper:
This ACL Best Paper, "Tracking the Trails of Political Biases Leading to Unfair NLP Models", is exemplary of a new group of papers pointing to "political bias" in language models like ChatGPT.
Another from yesterday:
Is it really political bias?
🧵:
Best Papers (3/3)
📢S1: Ethics & NLP (Poster)
📌 From Pretraining Data to LMs to Downstream Tasks: Tracking the Trails of Political Biases Leading to Unfair NLP Models
🔍Impact of political biases on LMs and downstream NLP tasks🤝
🔗
🧵(5/5)
Five days left to submit a paper to the 3rd Monash-Warwick-Zurich "Text as Data" Workshop, co-organized with
@essobecker
.
Submit here by Jan 16th:
Workshop is public and held on Zoom, Thu/Fri Feb 17-18, 2022.
The next online seminar on the economics of crime is Thursday at 1pm ET.
#EconTwitter
@ellliottt
will present his paper “Measuring Gender and Religious Bias in the Indian Judiciary”
More info:
Excited to share this new research by two PhD students and a predoc at
@eth_cle
, expanding work on cable news to vaccine hesitancy.
Media skepticism of COVID-19 vaccines reduces vaccine takeup!
In a new working paper,
@gochristoph
, Léo Picard, and I investigate the role of cable television news on COVID-19 vaccination compliance in the United States.
(thread)
⏰ Deadline Approaching!
🚀Apply by Nov 22 to become an ETH AI Center
#PhD
or
#Postdoc
!
We're hosting an EXTRA Online Q&A Session
📅Monday, Nov 20, 16:15 - 17:00 CET.
Zoom link on our website 🔗
Don't miss this chance to get the info you need!
Just traveling back from three amazing days
@ETH
Zurich, where I will start my postdoc this fall in the group of
@ellliottt
@eth_cle
. So exited! Thanks a ton to everyone who supported me on this journey, especially
@Woessmann
🥳
The standing room only ASREC session at the 2023 ASSA was amazing!
Thanks to presenters Daron Acemoglu, Paola Giuliano & Elliott Ash, and discussants Raquel Fernandez, Leah Boustan & Erzo Luttmer for fantastic presentations and discussions. And thanks to all those who attended😍
🔜ETH Zurich AI Policy Summit 2022
Online: Oct 10-12th
On-Site/Hybrid: Oct 13th
AI+Econ Workshop: Oct 14-15th
Info and registration:
proudly co-organized with
@PiottiAyisha
and
@sergallet
List of speakers/panelists:
#AIPolicySummit2022
The ASREC session at 2023 ASSA is going to be amazing!
We will have presentations from Daron Acemoglu, Paola Giuliano &
@rodrikdani
Discussions by Raquel Fernandez,
@leah_boustan
&
@Erzo_FP
#NLLP
2021: In "Machine Extraction of Tax Laws from Legislative Text",
@ellliottt
&
@MalkaGuillot
introduce a text classification system to detect if a piece of text is from tax law & automatically distinguish the tax source. Useful for lawyers& empirical legal studies!
#EMNLP2021
Schedule Now Online!
4th Monash-Warwick-Zurich Text-as-Data Workshop:
Spanning time zones globally:
Session 1, Sept 1st Morning Europe, Evening Asia
Session 2, Sept 1st Morning USA, Evening Europe
Session 3, Sept 2nd, Morning Europe, Evening Asia
We then apply the measure to the transcripts of 156 years (!) of speeches in U.S. Congress (1858-2014).
Emotional expression spikes during times of war, and has been mostly increasing -- especially since ~1979.
This paper is one of many to be presented at next week's Monash-Warwick-Zurich Text-as-Data Workshop.
Schedule Now Online
August 30th-31st, 2021 (on Zoom)
Register:
More evidence is needed on how ChatGPT and related technologies will influence politics. While these technologies can help industry lobbyists, they can also help consumer rights groups -- and everybody else.
(1/N) Happy to announce the 4th
@MonashWarwick
-
@ETH
Zurich Text-As-Data Workshop, to be held online September 1st-2nd, 2022.
Co-organized with the great
@ellliottt
Ash.
Submit papers by midnight July 1st, 2022.
Paper Submission Form:
More info 👇
The next ViCE seminar is Thursday at 1pm ET.
#EconTwitter
@ellliottt
will present: "A Machine Learning Approach to Analyze and Support Anti-Corruption Policy"
More info:
May 13th: Online Workshop in Computational Analysis of Law hosts their first conference!
Interesting time-zone-agnostic format, with presentations at 9am ET, noon ET, and 9pm ET.
Register here:
Thanks to
@m_livermore
for organizing!
Higher Fox News viewership in an area (exogenously induced by lower channel position) leads local newspaper content to more closely resemble Fox News transcripts than CNN or MSNBC transcripts
#polisciresearch
Fourth:
@ellliottt
@davidjiyangcai
Draca Liu analyze effect of large scientific recruitment initiative in China on returning (junior) scientists + peers; finds neutral productivity effects for participants (drop, followed by rebound)
@m_clem
You could embed the strings in both datasets, for example with fasttext, and find the closest vector in the other dataset. That’s computationally intensive, but there are fast algorithms like faiss, you could do PCA to reduce dimensionality, and there could be other tricks.
This is amazing. The next step is to unbundle recommendation engines form the platforms, to allow anybody to write their own recommendation system based on the same data points.
@fgilardi
Thanks to
@cage_warwick
for the shoutout!
“Gender Attitudes in the Judiciary: Evidence from U.S. Circuit Courts” (with
@ariannaornaghi
and
@Daniel_L_Chen
)
Paper here:
A good opportunity to share the highlights (thread):
A: Extractive summarization on legal documents!
Just posted: "Legal Extractive Summarization of U.S. Court Opinions":
(with E. Bauer,
@dominsta_nlp
& N. Gu, all of
@ETH_en
)
Repo:
Web demo:
Thanks to
@VISconvention
for the chance to present at this event and to the audience for great questions.
slide deck here ("Reading Minds with Natural Language Processing"):
Wow - I am thrilled to receive the
🥳Fürther Ludwig-Erhard-Preis 2023 🥳
for best PhD, recognized for innovation, practical relevance, and societal impact
Special thanks to my fantastic PhD supervisor Ludger
@Woessmann
and all amazing co-authors.
Current hypothesis: LLMs are a lot like surveys.
When they see a context ("The cat and the") they basically conduct a *survey* over every datapoint in a training dataset.
It's like asking every datapoint "what do YOU think the next word might be"?
And then...
Now online! Our paper with
@AlliPatter
and
@JFagan46
on how budget pressures reduce racially discriminatory targeting by traffic police — they target (wealthier) white drivers instead.
Suggests police discrimination responds to incentives, perhaps could be reduced by policies.
Simple summary of Loper Bright, the big “Chevron” decision from yesterday:
1) The case involves a group of commercial fishermen who sued the National Marine Fisheries Service (NMFS) over a rule requiring them to fund at-sea monitoring programs. The fishermen argued that the
The MWZ Text-as-Data Workshop is about to start. It's online-only today Mon & tmw Tue. Join us for the conference or parts of it! 🙂 To get the Zoom link, click on the registration link in the program. We will send you the link ASAP. See you there 😊
Happy to share the presentation from the first Econ & Data Science reading group!
@MNDjourel
presented "Media Persuasion through Slanted Language: Evidence from the Coverage of Immigration".
Link to video:
Link to paper:
Moore v. U.S. -- Thursday's SCOTUS decision on taxing unrealized gains -- summarized in plain English:
1) The case involves Charles and Kathleen Moore, who invested in an Indian company called KisanKraft. The company is a Controlled Foreign Corporation (CFC), meaning it is
Are you working in the social sciences with text-as-data (and/or other unstructured data)? Present your work at the 6th edition of the Monash-Warwick-Zurich Text-As-Data Workshop on 18-19 September 2023! Submit your paper here:
This episode brings up some thorny issues with ChatGPT and related technologies, which are no longer simply autoregressive language models but now have ideological constraints imposed on the outputs.
Starting LIVE in 5 minutes...
What does ChatGPT tell us about the promises and dangers of AI? What explains its evident biases? What are the largest lurking potential abuses?
Plus:
@RepKenBuck
on Big Tech censorship and antitrust laws as a solution.