EU AI Regulations | The Juice
Zumo Labs presents The Juice, a weekly newsletter focused on computer vision problems (and sometimes just regular problems). Get it while it’s fresh.
Week of May 10–14, 2021
____
What’s the difference between a high risk use of artificial intelligence and an unacceptable risk? What restrictions should be placed on high risk systems that wouldn’t be necessary for AI systems with minimal risk? These are hard questions, and, unfortunately, not the sort of questions you can necessarily trust for-profit enterprises to answer for themselves.
Last month, the European Commission published the Artificial Intelligence Act, a comprehensive proposal for regulations surrounding AI. It’s over 100 pages, and aims to answer with certainty some of the toughest questions surrounding the continued use of AI. It’s a lot of legalese to work through though, so our own Elena has summarized the document in this handy TL;DR.
Of particular interest? High risk AI systems in the EU must be trained on datasets that are “relevant, free of errors and complete,” per the new framework. That reads like a prescription for made-to-order synthetic datasets, if you ask us.
____
#Football
Did you know the founder of DeepMind, Demis Hassabis, is a lifelong Liverpool fan? That may be why DeepMind has teamed up with Liverpool to publish a paper regarding applications for AI in soccer. The goal of the research is not to replace managers with soccer robots, but to use computer vision to better analyze the game. With enough data, a model may even be able to predict how a player may react in a given situation.
DeepMind Wants to Use AI to Transform Soccer, via Wired.
#Footfall
A company called Tec-Innovation currently manufactures a smart shoe called the InnoMake, which uses ultrasonic sensors to notify the wearer of obstacles in their path. Now Tec-Innovation has teamed up with Graz University of Technology in Austria to add computer vision to the shoes, which should allow the shoes — and through haptic feedback and/or a phone app, the wearer — to better differentiate objects and impediments ahead.
Seeing-Eye Shoes Pair Computer Vision With Haptic Feedback, via Hackaday.
#BakeryScan
Before deep learning proved itself and became the preferred approach to computer vision in 2012, vision algorithms were tediously hand-coded and combined to accomplish specific tasks. This fantastic article from The New Yorker chronicles the development of one such algorithm, which was assigned the difficult task of differentiating between “a carbonara sandwich, a ham corn, and a minced potato” (among other things) in Japanese bakeries.
The Pastry A.I. That Learned to Fight Cancer, via The New Yorker.
#SelfDriving
While not as dramatic as Lyft selling off its entire self-driving division, there’s been a leadership shakeup over at autonomous driving company, Waymo. Their head of auto partnerships and their CFO have both announced their plans to leave, joining several other executives — head of manufacturing, treasurer, chief safety officer, and CEO — who have left the company over the past five months.
Waymo to lose its CFO and head of automotive partnerships, via TechCrunch.
#BlackInAI
“Three groups focused on increasing diversity in artificial intelligence say they will no longer take funding from Google. In a joint statement released Monday, Black in AI, Queer in AI, and Widening NLP said they acted to protest Google’s treatment of its former ethical AI team leaders Timnit Gebru and Margaret Mitchell, as well as former recruiter April Christina Curley, a Black queer woman.”
Black and Queer AI Groups Say They’ll Spurn Google Funding, via Wired.
#Edge
The USPS has partnered with Hewlett Packard Enterprise and Nvidia on a system called the Edge Computing Infrastructure Program, which is helping them reduce the amount of time it takes to scan and locate mail by bringing image processing in-house. There are bigger plans for the operation — 30 different applications are currently planned for the tech — but if this first step keeps my mail from getting lost, that’s progress.
USPS Embraces AI to Sort Packages, via EE Times.
____
📄 Paper of the Week
This week we take you a little outside our usual computer vision focus. This group out of Harvard provides a great summary of the current lay of the land for machine learning in medicine “from protein interactions to disease networks, all the way to healthcare systems and scientific knowledge”. The ML world today tends to focus on tasks such as computer vision and natural language processing, but once steam picks up on the medical field, the potential for massive changes in the way we live will skyrocket.
____
Think The Juice was worth the squeeze? Sign up here to receive it weekly.
