🍩 AI for pastry puffs takes on cancer cells and deepfakes have entered the cheerleading discourse
🚗 A timeless joke: When will Tesla's AutoPilot work? Next year.
The Good
In Japan, bakeries sell more if they have tons of varieties to pick from. But memorizing the prices of hundreds of items was a pain, so a bakery asked a tech company to build a tool that could automatically recognize different baked goods. They did, a few years before the deep learning breakthrough would make it easier, and the approach is turning out to be helpful with all sorts of things, INCLUDING recognizing CANCER CELLS. The tech, called AI-Scan, can do even more:
AI-Scan algorithms have since been used to distinguish pills in hospitals, to count the number of people in an eighteenth-century ukiyo-e woodblock print, and to label the charms and amulets for sale in shrines. One company has used it to automatically detect incorrectly wired bolts in jet-engine parts.
The system’s secret weapon is a backlight, which helps clearly define edges and standardizes the light condition. There seems to be other novel approaches that makes this system still special amid a sea of similar tools, but the reporter doesn’t explain them.
This story has some other strange reporting issues: the author was stunned by the computer vision system in 2019 but is a programmer and has applied for a job in the field a few years beforehand, strangely describes the recent history of deep learning, and (this is probably the New Yorker’s fault) spends way too much time explaining how AI works in some sections and then drops terms like “parameters” without any explanation. But the nugget of the story, that a computer vision system built to figure out if you got a strudel or a croissant can help with cancer research is freaking neat.
+Bonus content—A link to CHAI - Donuts mind if I do, a song by a Japanese all-girl band I love and shown in the gif above.
The Bad
Welp, deepfakes have entered the cutthroat world of suburban cheerleading. Police arrested a Pennsylvanian cheer mom for sending manipulated images and videos of her daughter’s rivals to coaches. In a nod to old fashion harassment, the mom also sent texts to the girls telling them to kill themselves!
The cops called the images deepfakes, but I’m not sure if we should take the word of local police on the type of tech used in attacks like this. Did I try to do some digging to see what the cheer mom did for a living to see how likely it was she was able to do it on her own, or might be on the right forums to get someone to make her one? Maybe. But apps like Reface could let anyone, no matter how tech adverse, put faces on images and videos that would be convincing to many people. Get ready, this is the start of a story that happens in every community across the country, and eventually the world.
More News
There is a new video of Tesla’s AutoPilot out in the real world, and well, let’s just say the cross country drive this year isn’t going to be happening.
GPT-3, the massive natural language tool, occasionally spits out real phone numbers from the data it was trained on. So the company behind it is creating a filter.
The same New York Times reporter that revealed the existence of Clearview AI has an update on the company’s many lawsuits, as well as their actual origin story.
AI can make accessibility better for folks with disabilities—but not always, like this tool that is suppose to fix pages so that people who use screen readers and other assistive technologies can access them.
Israel got some help from AI-powered surveillance to figure out the ship that was responsible for an off-shore oil spill near their shores.
Not AI, but from me: I wrote about livestream shopping, the latest way Amazon is trying to get your dollars.
***
Until we are all jamming to this song,
Jackie