#2019APPAM Pre-Conference Workshop: Deploying Machine Learning Tools for Public Policy Impact

Missed the #2019APPAM Pre-Conference Workshop: Deploying Machine Learning Tools for Public Policy Impact? Here is a twitter recap of the session.

 

 

 

 

 

 

 

Agenda for Pre-Conference Workshop: Deploying Machine Learning Tools for Public Policy Impact #2019APPAM pic.twitter.com/ionkbgEgq9

— Rafael M Batista (@RafMBatista) November 6, 2019

.@profjensludwig opens the Pre-Conference Workshop on Deploying Machine Learning Tools for Public Policy Impact at #2019APPAM pic.twitter.com/HldMtzozrt

— APPAM (@APPAM_DC) November 6, 2019

If you're coming into public policy today you might not recognize the revolutions that have taken place the last 30 yrs.

Causal inference / RCTs and Behavioral economics were unheard of then and now commonplace.

ML / AI as the latest revolution taking place #2019APPAM pic.twitter.com/4h6OwJ6Ojc

— Rafael M Batista (@RafMBatista) November 6, 2019

Goals of "prediction" and "estimation" are different #2019APPAM pic.twitter.com/JWmKtQzlmo

— Rafael M Batista (@RafMBatista) November 6, 2019

Important to recognize *prediction* policy problems#2019APPAM pic.twitter.com/5rE62wWnuX

— Rafael M Batista (@RafMBatista) November 6, 2019

Alex Chouldechova (@HeinzCollege) on choosing between different models w/ near-optimal accuracy.

What is the trade-off to accuracy, for example, if you also value "fairness"?

Less than you'd imagine. #2019APPAM pic.twitter.com/O0dqCoOx04

— Rafael M Batista (@RafMBatista) November 6, 2019

For example, improving educational outcomes often focus on interventions. But surely *which* teachers to hire is also an important question. This is a prediction problem.

Read more here: https://t.co/jsor9OxA96#2019APPAM

— Rafael M Batista (@RafMBatista) November 6, 2019

In the #ML community, the domains that the people in this room are good at (ie public policy; econ; etc) is a huge blind spot.

Similarly, many in Public Policy are not (yet) using the tools that those in ML are expert at. @achould #2019APPAM

— Rafael M Batista (@RafMBatista) November 6, 2019

Question was asked at #2019APPAM regarding biased algorithms and regulation--

For those interested see piece by @CassSunstein & @profjensludwig https://t.co/cvYWPAdzCM

— Rafael M Batista (@RafMBatista) November 6, 2019

Next up is @jlanastas speaking about intersectionality of causal inference, ML, and public policy#2019APPAM pic.twitter.com/ndJN6vAYs9

— Rafael M Batista (@RafMBatista) November 6, 2019

Could image recognition algorithms and new computer vision techniques issue in a new era of fair and just government?#2019APPAM pic.twitter.com/dl5DcteZn8

— Rafael M Batista (@RafMBatista) November 6, 2019

Type I and Type II errors in algorithms have direct policy implications tor racial bias and fairness. pic.twitter.com/qkdxDxPUcg

— Rafael M Batista (@RafMBatista) November 6, 2019

@jlanastas provides example where they use image recognition algorithms to better understand political party and ideology

See preprint here: https://t.co/EzCghStRdO#2019APPAM pic.twitter.com/mYYexmqhaH

— Rafael M Batista (@RafMBatista) November 6, 2019

Next up at #2019APPAM Pre-conference, James Evans (@UChicago) pic.twitter.com/iRS8ivaxFo

— Rafael M Batista (@RafMBatista) November 6, 2019

When does #BigData matter? When you have rare (but consequential) events that you want to understand#2019APPAM pic.twitter.com/I8sIBl96iQ

— Rafael M Batista (@RafMBatista) November 6, 2019

Pay attention to "datafication" of social world. With #ML, everything is data!#2019APPAM pic.twitter.com/ohcr1eySDC

— Rafael M Batista (@RafMBatista) November 6, 2019

.@ChenhaoTan is up next#2019APPAM pic.twitter.com/giGDiAnFNr

— Rafael M Batista (@RafMBatista) November 6, 2019

@ChenhaoTan starts with Twitter prediction

Check out for yourself how this works here https://t.co/AbUUgWzPeM#2019APPAM pic.twitter.com/wO2la1C0yC

— Rafael M Batista (@RafMBatista) November 6, 2019

Policy decisions are challenging for humans. Machines can help. But there's insight in understanding how and when. #2019APPAM pic.twitter.com/HXcoFZxFoo

— Rafael M Batista (@RafMBatista) November 6, 2019

An example, decision aids can highlight important features that algorithm is using.

But new form of data is needed. #2019APPAM pic.twitter.com/u6uGCoQMZg

— Rafael M Batista (@RafMBatista) November 6, 2019

Program formulation is key if we want to use Machine Learning in policy and decision making.#2019APPAM

— Madeleine F. Wallace (@MadeleineWR) November 6, 2019

.@Aaroth speaking on ethical algorithms#2019APPAM pic.twitter.com/Jp4KaQXUB8

— Rafael M Batista (@RafMBatista) November 6, 2019

Regarding interpretability of algorithms, George Chen reminds us that there are many medications that we don't know how to explain how they work, but we know they work. Difference may be that for medicine we trust they've gone through rigorous clinical trials#2019APPAM

— Rafael M Batista (@RafMBatista) November 7, 2019

.@jenniferdoleac presenting a counterintuitive notion whereby providing information which *seems* like it will negatively affect certain groups can actually be beneficial.

Ie Adding information can benefit disadvantaged group.#2019APPAM pic.twitter.com/fS297IvQ1P

— Rafael M Batista (@RafMBatista) November 7, 2019

​