Bugs Bite | 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 June 7–11, 2021
____
Midway through Ingenuity’s sixth test flight on Mars, the autonomous helicopter began to pitch and roll unpredictably. Fortunately, as the device initiated its descent, it managed to level off and set down safely. But it was a harrowing few moments to watch. Crashing your drone in your neighbor’s back yard is one thing, but flipping your $80 million helicopter on Mars — a helicopter carrying a piece of fabric from the wing of the 1903 Wright Flyer, no less? That’s game over.
NASA recently published their findings on the in-flight anomaly. “Approximately 54 seconds into the flight, a glitch occurred in the pipeline of images being delivered by the navigation camera. This glitch caused a single image to be lost, but more importantly, it resulted in all later navigation images being delivered with inaccurate timestamps. From this point on, each time the navigation algorithm performed a correction based on a navigation image, it was operating on the basis of incorrect information about when the image was taken.”
It was a glitch. The stability margin NASA built in, and the decision not to rely on nav cam data during descent, helped the bot survive to fly another day. But a bug very nearly took down the off-world copter. Back on Earth, the other type of bug is giving terrestrial AVs hell. And with Brood X emerging, it seems the bugs may come for us all — autonomous or not.
____
#FacialRecognition
In the United States, there is no federal legislation governing the development or use of facial recognition technology. If the government has been waiting for the market’s invisible hand to work its magic, well, they might be in luck. A group of investors managing $4.5 trillion in assets has called on 34 market leaders in facial recognition to pursue the technology ethically. Their complete statement, with specific asks, can be found here.
Investors call for ethical approach to facial recognition technology, via Reuters.
Meanwhile, the Canadian Privacy Commissioner reported to Parliament this morning that the RCMP broke the law using Clearview AI and that the investigation’s findings are “extremely troubling.” Specifically, he said that the national police force violated a section of the Privacy Act in using the software. NDP MP Charlie Angus agreed, saying the report is “startling and incredibly disturbing,” adding, “Privacy rights are fundamental. Clear rules over the consent and taking of private information must be a priority for our government.”
RCMP’s use of facial recognition tech violated privacy laws, investigation finds, via CBC.
If you’re not interested in waiting for fund managers to convince Clearview AI to at the very least “disclose the accuracy” of their model and build “grievance mechanisms,” what can you do to protect your face? According to The Verge, you could subtly alter the images you upload online to create adversarial attacks on Clearview’s model. Otherwise, “the only way to stop Clearview from gathering your data is by not allowing it on the public internet in the first place.”
Is there any way out of Clearview’s facial recognition database?, via The Verge.
#MaaS
Mosquitos are annoying. But they’re also, according to scientists, “the most dangerous animal on earth.” A new book, The Mosquito: A Human History of our Deadliest Predator, estimates that mosquitoes have killed 52 billion people in total — almost half of all humans who have ever lived. The most effective way to reduce mosquito populations is the “sterile insect technique,” where sterile males are released into the population to mate with females (who only mate once). The challenge has historically been sexing the mosquitos, but one team has developed technology that uses computer vision to sex mosquitos in their larval state. That’s early enough in their lifespan that sterile males are now a viable, shippable subscription product.
How AI and mosquito sex parties can save the world, via VentureBeat.
#HotTopic
We shared a piece recently that explained how Ring became the largest civilian surveillance network in the US. Over the past year, they’ve expanded their partnership efforts beyond the police to include over 350 fire departments, allowing agencies to request footage from Ring users without a warrant. Citizens can deny the request, but the agencies can then contact Ring directly with reports of “uncooperative or unavailable” camera owners. Per the Buzzfeed News article, participating fire departments have issued 214 requests for footage in the past year alone.
Ring Has Now Partnered With 350 Fire Departments, via Buzzfeed News.
#AIID
We all make mistakes. But only some of those mistakes give rise to a racist chatbot. Otherwise known as Incident 6, Microsoft Tay lives on in infamy — as a warning to others — in the AI Incident Database. The database is managed by Partnership on AI, with the goal of folks working in AI both documenting and learning from ‘good faith errors.’ Wired has less charitably dubbed it the, “AI Hall of Shame.” Either way, it’s good reading.
Don’t End Up on This Artificial Intelligence Hall of Shame, via Wired.
____
📄 Paper of the Week
Humans use both of our eyes to perceive the relative distance of objects in our field of view. However, even with an eye patch, we can be quite good at depth estimation. Re-creating this capability in computer vision is known as monocular depth estimation, and this paper has some impressive results. The researchers use a double estimation method that predicts depth at a whole-image level and then zooms in to detail smaller patches of the image. Traditionally, monocular depth estimation papers deal with very visually consistent images, such as the view from an autonomous vehicle. To see something working on large crowd scenes is significant. An extra tidbit: depth estimation is a great auxiliary loss if you happen to have depth data for your image dataset.
____
Think The Juice was worth the squeeze? Sign up here to receive it weekly.
