HOWTO AI FOR GOOD ARE ADS, GAMES, AND GANS REALLY THE BEST WE CAN - - PowerPoint PPT Presentation

▶

Apr 30, 2023 143 likes •411 views

Julien Cornebise, Ph.D. 2019/03/21 - NVIDIA GTC HOWTO AI FOR GOOD ARE ADS, GAMES, AND GANS REALLY THE BEST WE CAN DO? A PATH (TOO) WELL TRODDEN 30 years of ICT4D Many fiascos One Laptop Per Child Telecentres

SLIDE 1

HOWTO “AI FOR GOOD”

Julien Cornebise, Ph.D. 2019/03/21 - NVIDIA GTC

SLIDE 2

ARE ADS, GAMES, AND GANS REALLY   THE BEST WE CAN DO?

SLIDE 3

A PATH (TOO) WELL TRODDEN

▸ 30 years of ICT4D ▸ Many fiascos ▸ “One Laptop Per Child” ▸ Telecentres ▸ PlayPump ▸ Many pitfalls to learn from!

SLIDE 4

“TECHNOLOGY - NO MATTER HOW WELL DESIGNED - IS ONLY A MAGNIFIER OF HUMAN INTENT AND CAPACITY.”

Kentaro Toyama (2015)

SLIDE 5

PARTNER HAND IN HAND WITH DOMAIN EXPERTS: EMPOWER THEM

SLIDE 6

“- We TOLD you it was hard. 

Yeah, but now that I’VE tried, we KNOW it’s hard.”

https://xkcd.com/1831/

SLIDE 7

NAVIGATING PITFALLS

▸ Path to concrete impact? ▸ Get rid of hype ▸ No “Tech Savior”, no “White Savior” ▸ Identify right depth of tech ▸ Hard question ▸ Are you here to solve a problem ▸ or to illustrate your method? ▸ Plan for Sustainability - long development ▸ Hackathon syndrome: GitHub as a cemetery ▸ Where are the engineers and designers?

SLIDE 8

2017 DECODE DARFUR: FINDING BURNED VILLAGES

Daniel Worrall Milena Marin @milena_iul

SLIDE 9

SLIDE 10 Stuart Price/Albany Associates Brian Steidle

SLIDE 11

SLIDE 12

DEEP LEARNING TO DECODE DARFUR

SLIDE 13

HIGH-RESOLUTION VIDEO REMOVED FROM RECORDED TALK

▸ Apologies, we cannot share online the video of the results, because the

location of the villages found is sensitive.

▸ Instead, we have made available low-resolution screenshots in the next slide.

SLIDE 14

Green: volunteers labels Red: predictions Urbanization patterns in Darfur Villages inside empty circles then isolated huts Urbanization patterns in Uganda – out of country generalization Micro-groupings of huts along criss-crossing roads

SLIDE 15

SLIDE 16

2018  TROLL PATROL: STUDYING ABUSE AGAINST WOMEN ON TWITTER

SLIDE 17 Slide redacted for online publication  https://decoders.amnesty.org/projects/troll-patrol/findings

SLIDE 18

STUDY: ML TO HUMANS TO ML TO HUMANS

▸ Build a dataset with ML prefilter ▸ Importance sampling to ▸ reduce variance ▸ drive engagement ▸ Human Crowdsourcing ▸ Data Analysis with Bootstrap ▸ Train Deep model and benchmark ▸ Carefully and expertly report and go public! ▸ Very sensitive topic, human rights expertise!

SLIDE 19

SLIDE 20

CONCRETE IMPACT

▸ Massive coverage ▸ Twitter stock tumbles -13% in one day ▸ 2.5 Billion USD wiped off ▸ Incentive to open up, reach out, act ▸ Twitter CEO spoke with Amnesty SG ▸ Twitter ML team reached out ▸ To be continued…

SLIDE 21

NOW WHAT?

…in addition to www.hrw.org/donate ;-)

SLIDE 22

SLIDE 23

SLIDE 24

SLIDE 25

SLIDE 26

THANK YOU!

eai4good@elementai.com