Last Week in AI #176: Drones beat human pilots in first fair race, better call quality with AI, how artists view AI-generated art, and more!
Autonomous drones beat human pilots in first fair drone race, UW developed AI-powered earbuds to enhance speech quality, interview with artists on their experience with DALL-E 2
A year ago researchers from the University of Zurich showcased their autonomous drones that were able to beat the fastest human pilots. However, that race wasn’t “fair” in the sense that the AI algorithm commanding the drones had extra information that human pilots didn’t have. In particular, the algorithm had access to near-perfect location and velocity estimation of the drones using motion capture systems, high-quality maps of the race course beforehand, and stereo cameras that can give depth information. This year, the team’s autonomous drones raced on even playing fields without these handicaps, and its AI was able to beat the best human-controlled time by 0.5s in a three-lap race, a significant lead in the world of drone racing.
Our take: This development is representative of AI progress ins many fields, where the researchers first make a working system with additional assumptions and then slowly chip away at these assumptions for a more robust and adaptable AI system. This win shows the maturity of both the software of AI as well as the hardware since the algorithm was run on the drone’s onboard computers.
Researchers at the University of Washington invented ClearBuds, wireless earbuds that use deep learning for high-quality background noise suppression. This is particularly useful in a world where people are constantly on the move and have to occasionally take calls in crowded scenarios. ClearBuds makes two key technical contributions: 1) implementing a synchronized, binaural microphone array and 2) a neural network capable of running on a lightweight mobile device for enhancing speech. A user study conducted on thirty-seven participants showed ClearBuds outperforming the Apple AirPods Pro.
Our Take: The rising popularity of hybrid work also means workers have to take meetings and calls in less controlled backgrounds with crying infants, blenders whirring in the kitchen, someone placing a pizza order, etc. Improved background noise cancellation is a boon in these scenarios, and it’s impressive that the learning-based approach outperforms more classical noisy cancellation algorithms. We look forward to further progress from the UW researchers along these lines.
Since OpenAI's DALL-E opened the floodgates for generating images that resembled text inputs, artists and "artists" have tried their hands at incorporating AI into their work. Recent upgrades--DALL-E 2 and Imagen--generate even more realistic images. While DALL-E 2 is limited-access, open-source versions like dalle-mini (now called Craiyon) are available for anyone to use--OpenAI acknowledges, "however vaguely," that image generators may reinforce existing social biases. The Guardian interviewed a few artists who have been experimenting with DALL-E 2. Their takes reflect different responses to these visual-language models: that of course, they are reflections of us, that they are interesting but limited. One isn't sure she would even use DALL-E.
Our Take: DALL-E and its ilk have been very interesting advances to see over the past few years, and I won't deny that the progress we've made in multimodality is important. But I also feel skeptical about whether there is an actual use for these sorts of models (while the uses for a plain GPT-3, however, limited it might be, are slightly more clear). They have provided plenty of entertainment for ML Twitter (see Weird Dall-E Mini Generations), but this doesn't constitute much in the way of an actual use case. It's possible that artists will find these tools interesting and incorporate them into their work, but it's hard to say much at this stage.
Meta's 'Make-A-Scene' AI blends human and computer imagination into algorithmic art - "Text-to-image generation is the hot algorithmic process right now, with OpenAI’s Craiyon (formerly DALL-E mini) and Google’s Imagen AIs unleashing tidal waves of wonderfully weird procedurally generated art synthesized from human and computer imaginations."
DeepMind AI learns simple physics like a baby - "Neural network could be a step towards programs for studying how human infants learn."
Doctors using AI catch breast cancer more often than either does alone - "Radiologists assisted by an AI screen for breast cancer more successfully than they do when they work alone, according to new research. That same AI also produces more accurate results in the hands of a radiologist than it does when operating solo."
Inside a radical new project to democratize AI - "A group of over 1,000 AI researchers has created a multilingual large language model bigger than GPT-3—and they’re giving it out for free."
New ‘cyborg’ method combines human and machine intelligence to find 40,000 ring galaxies - "Machine learning is unlocking the cosmos. Scientists from the National Astronomy Meeting announced that humans worked together with machine intelligence to find 40,000 ring galaxies throughout the universe, a press statement reveals."
Deformable pump gives soft robots a heart - "The Tin Man didn't have one. The Grinch's was three sizes too small. And for soft robots, the electronically powered pumps that function as their "hearts" are so bulky and rigid, they must be decoupled from the robot's body -- a separation that can leak energy and render the bots less efficient."
An AI-Driven Bias Checker for News Articles, Available in Python - "Researchers in Canada, India, China and Australia have collaborated to produce a freely-available Python package that can effectively be used to spot and replace ‘unfair language’ in news copy."
Meta develops AI system for reviewing Wikipedia citations - "Meta Platforms Inc. has developed an artificial intelligence system that can scan a Wikipedia article, analyze the sources cited by the article and identify if some of them may need to be changed."
Using AI to diagnose birth defect in fetal ultrasound images - "In a new proof-of-concept study led by Dr. Mark Walker at the University of Ottawa's Faculty of Medicine, researchers are pioneering the use of a unique Artificial Intelligence-based deep learning model as an assistive tool for the rapid and accurate reading of ultrasound images."
This D&D inspired game's AI image generator is now free for all - "I love me a good AI Dungeon (opens in new tab) session, and I'm happy to announce the devs have just made their AI image generator free to use with the base AI. At the same time though, it looks like the free version now includes ads so, swings and roundabouts."
An experimental horror ARG is testing the boundaries of AI art - "According to ancient Nine Inch Nails (NIN) lore, we’re living in Year Zero, which began on February 10th, 2022."
Meet the startups using AI to help doctors fight burnout - "A new wave of healthcare startups is using artificial intelligence to reduce the load on medical professionals."
Argo AI lays off 150, closes Washington D.C. operation - "Argo AI announced late last week it is laying off 150 employees and shutting down its autonomous vehicle operation in Washington, DC. A number of now former employees posted about the layoffs on LinkedIn."
Amazon Picks 7 Startups for First Black Founders Build with Alexa Cohort - "Amazon has chosen seven startups for the first Black Founders Build with Alexa cohort."
XPeng’s robotics unit raises US$100 million with the aim of commercialising its business in two years - "XPeng Robotics, the robotics unit of Chinese electric vehicle (EV) maker XPeng Motors, has raised US$100 million in a funding round led by IDG Capital and expects to launch its first robots for household use in two years."
Amnon Shashua’s AI21 Labs raises $64 million for natural language processing platform - "AI21 Labs, an Israeli startup aiming to change the way people read and write, announced on Tuesday the completion of its $64 million Series B funding round, bringing the company’s valuation to $664 million."
Tesla AI leader Andrej Karpathy announces he's leaving the company - "Tesla AI and Autopilot leader Andrej Karpathy announced Wednesday that he's no longer working for the electric vehicle maker. CEO Elon Musk thanked Karpathy for his work in response."
AI drug miner XtalPi strikes gold with $400M infusion, its second VC megaround in a year - "Nine-figure funding rounds are rare in the world of medtech—barely a dozen companies achieved the feat in 2020—but XtalPi, which has developed software that uses artificial intelligence to identify and model the most promising new drug compounds, just did it twice in one year."
AI cancer pathology developer Paige nets $100M in venture capital funding - "After scoring two European approvals last month for its artificial-intelligence-powered cancer pathology platforms, Paige has secured $100 million in new funding to boost its development of biomarkers as clinical applications."
Why investors see the potential in AI-powered, copy-generating adtech companies - "Sophisticated AI systems like OpenAI’s GPT-3 can write prose that’s impressively human-like, or at least good enough to fool the average person. They’ve been used to generate essays, poetry, stories, news reports and more to impressive effect."
The Carbon Footprint of Machine Learning - "Machine learning (ML) is great for augmenting human intelligence but leaves a carbon footprint."
The Kitten Effect - "One thing I've noticed with image-generating algorithms is that the more of something they have to put in an image, the worse it is. A few years later, Dall-E 2 is a LOT more coherent despite having a larger job to do. But it's still susceptible to the kitten effect."
We asked AI image app DALL-E 2 to show what the San Francisco Bay Area could be - ""I keep waiting for [Sutro Tower] to stalk down the hill and attack the Golden Gate Bridge," longtime San Francisco Chronicle columnist Herb Caen famously quipped."
This AI image generator lets you type in words and get weird pictures back - "It only took Matt Laming, a 19-year-old from the United Kingdom, about a month to hit a million followers on Twitter. And all it required was sharing a steady stream of the most outlandish computer-generated images that he and a bunch of internet strangers could think up."
Copyright © 2022 Skynet Today, All rights reserved.