Discover more from Last Week in AI
Last Week in AI #155: New way for AI to see in 3D, AI helps translating ancient languages, lawmakers pressure Feds to stop using Clearview AI, and more!
How neural radience fields took the AI 3D vision field by storm, ways that AI algorithms can assist decoding lost languages, and the latest on the government's use of Clearview AI
Almost two years ago, the paper NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis came out and introduced the idea of optimizing a neural network to be able to render a 3D scene given a set of images of it from different angles. Since then, refinements and variations on NeRF have been a huge area of research in AI, with the original paper already having almost 750 citations. The basic idea of NeRF is covered in this article, as is much of the work related to it that has come out since:
“Researchers at Nvidia, which makes computer chips for both AI and computer games, have published papers that use NeRF to generate 3D images from photo collections, to produce more realistic textures in animation, and point to advances for video games. Facebook (now Meta) has developed an approach similar to NeRF that could be used to flesh out scenes in Mark Zuckerberg’s much-vaunted Metaverse. Yann LeCun, chief AI scientist at Meta and a pioneer of the approach that shook things up in 2012, calls the new work “fascinating” and the results “quite impressive.”
Our take: This article is right in portraying NeRF as a big deal. It has indeed been a hot area of research, with a ton of exciting ideas related to it coming out often. Just last week, Block-NeRF Scalable Large Scene Neural View Synthesis came out and impressed many. In general, there’s no doubt much more research this decade will focus on problems having to do with 3D perception and reasoning, having conquered such problems as object detection and recognition in 2D images over the past decade.
In recent years, many researchers have been trying to apply advances in AI to better understand and decipher ancient, lost languages. One group developed an algorithm that tries to align words from an unknown language with words from a known one and achieved decent results when compared to translations made by experts. However, these algorithms are severely limited by the small amount of data that we usually have on ancient languages. Instead of doing translation, which is quite difficult, researchers have used statistical analyses to prove if a set of symbols indeed do form a language, and to look for connections among unknown languages and with known ones. It’s unlikely that AI can fully automate language decipherment, but it will be able to aid and speed up human efforts. From a researcher in the field:
“My thinking is: Let the system propose a list of candidates and let the experts see, Okay, maybe this theory is more correct than the other”
Our take: This article illustrates very well the benefits and limitations of existing, data-driven approaches to AI. Given sufficient data, AI algorithms can find interesting statistical relationships among unknown language symbols in hours as opposed to the years it took some experts. But, interpreting these relationships and grounding them into something meaningful still needs a lot of human work, which requires combining clues from more “abstract” data points like cultural and historical contexts.
Democratic lawmakers are stepping up their efforts to discourage federal agencies from using Clearview AI's facial recognition technology. Four Democrats, Sens. Ed Markey and Jeff Merkley and Reps. Pramila Jayapal and Ayanna Pressley, addressed their letters to the Departments of Justice, Defense, Homeland Security, and the Interior. Arguing that the use of Clearview's technology posed a threat to Americans' expectation of free movement and speech without surveillance, the lawmakers asked the agencies to end their use of Clearview's tech. Clearview CEO Hoan Ton-That has pointed to a NIST test and data from the Innocence Project to claim that his technology "shows no detectable racial bias" and is comparatively better than eyewitness lineups.
Our take: This is just the latest chapter in the long tale of Clearview AI. After all the controversy Clearview has endured, it's interesting to see federal agencies still using the technology. Indeed, government use of surveillance technology does not align well with American values, but that does not at all mean that US government agencies would stop themselves from taking advantage of promising technology. The Democratic lawmakers raise valid points, pointing out that facial recognition systems raise concerns beyond their accuracy. The discussion will probably break into multiple levels. Questions that need to be answered include (a) should US government agencies be using facial recognition technology at all? and (b) if yes, what standards need to be met?
Sony's AI Drives a Race Car Like a Champ - "The company built GT Sophy to master the game Gran Turismo, but it may help the development of real self-driving cars."
MIT Researchers Introduce a Machine Learning Technique that can Automatically Describe the Roles of Individual Neurons in a Neural Network with Natural Language - "Researchers at MIT have devised a mechanism for deciphering the inner workings of black-box neural networks."
Researcher pair drones, machine-learning against marine litter - "A project in support of the National Oceanic and Atmospheric Administration (NOAA) is pairing sensor-equipped drones with machine-learning applications in an effort to automate the identification and mapping of marine waste and create effective methods for its collection and disposal."
Machine Learning Uncovers Violence During Apartheid, Earns Top Honor from Harry Frank Guggenheim Foundation for University of Kentucky Researchers - "It takes community and collaboration to define violence, understand the root cause and prevent it from happening in the future."
Researchers Tackle COVID-19 with AI - "A pair of papers coauthored by Anima Anandkumar, Caltech's Bren Professor of Computing and Mathematical Sciences, were selected as finalists for the 2021 Association for Computing Machinery (ACM) Gordon Bell Special Prize for High Performance Computing-Based COVID-19 Research."
This AI is set to help stop illegal fishing - "New computer vision model marks the first time research and conservation have come together at AI2."
Machine learning predicts marital discord - "The life satisfaction of both partners and the woman's percentage of housework turned out to be the most important predictors of union dissolution"
The brain’s secret to lifelong learning can now come as hardware for artificial intelligence - "An electronic chip that can be reprogrammed on demand may enable artificial intelligence to learn more continuously like the human brain does, researchers have discovered."
Stanford researchers propose ‘jury learning’ as a way to mitigate bias in AI - "The idea is to model “individual voices” in training datasets toward designing a system that makes it possible for developers to explore — and ideally shift — the behavior of AI systems."
Mapping the world's fungal networks with machine learning - "A monumental effort is underway to map the world’s fungal networks"
Artist uses AI to perfectly fake 70s science fiction pulp covers – artwork and titles - "It’s algorithmic, generative 70s sci-fi pulp, courtesy OpenAI. Artist Lewis tells us about how it was done."
How robots helped a collapsed mine get back in operation - "The robots placed three network infrastructure points within the mine, the deepest being 1.7 km, and set up eight robot hubs."
MuZero’s first step from research into the real world - "Collaborating with YouTube to optimise video compression in the open source VP9 codec."
DeepETA: How Uber Predicts Arrival Times Using Deep Learning - "Uber AI partnered with Uber’s Maps team on a project called DeepETA to develop a low-latency deep neural network architecture for global ETA prediction."
AI Foundation Models for the Rest of Us - "Pre-trained networks give smaller teams a leg up"
Mad Scientist Forces AI to Make Horrific Pictures Out of BuzzFeed Headlines - "An AI Was Force-Fed BuzzFeed's Meme-Worthy Headlines and Here's What Happened"
Now Physical Jobs Are Going Remote Too - "Advances in artificial intelligence and other technology allow machines to be operated from far away. The trend could spell trouble for workers."
Apple Buys Startup That Makes Music With Artificial Intelligence - "AI Music will be used to bulk up Apple’s audio technology"
Autonomous vehicles logged 4 million miles on CA's public roads last year - "The record number of autonomous test miles with a safety driver took place even with 28 of the 50 California permit holders reporting they did not test AVs during the reporting period, from Dec. 1, 2020, to Nov. 30, 2021."
Waabi’s new simulator could scale autonomous vehicle tech faster - "Testing autonomous vehicles on public roads is an expensive and time-consuming endeavor, and one that Raquel Urtasun, former chief scientist at Uber ATG, doesn’t think is the most expedient route to market."
Deepdub closes fresh financing round for AI that dubs movies, shows, and games - "AI technologies promise to streamline the process by automating various aspects of dubbing."
Scopio secures $50M in series C funding to develop telehematology imaging and AI analysis platform - "Scopio Labs secured $50M in series C funding, which the company will use to scale manufacturing of its FDA-cleared imaging and AI analysis platform to support the shift to fully telehematology."
Agot AI gives restaurants computer vision to see where food orders go wrong - "Its product confirms order accuracy in real-time for online ordering and notifies employees if an order needs a correction; for example, they forgot to add cheese or ketchup."
KoBold Metals Raises $192.5 Million to Use AI to Find Battery Minerals - "Investors participating in Series B funding round include Apollo Projects, BHP, and Canada’s largest pension fund"
Zillow’s Shuttered Home-Flipping Business Lost $881 Million in 2021 - "Real-estate company says in a letter to shareholders that it is targeting revenue of $5 billion by 2025"
Microsoft Keeps Its Finance Head Count Flat With AI, Bots and Other Tech - "The new tools help automate tasks. But ‘we love Excel,’ says Cory Hrncirik, who leads Microsoft’s Modern Finance project."
A robot bought my seven-year-old car for more than I paid brand-new - "Carvana gave me more than I ever dreamed it was worth — how?"
Turtles all the way down: Why AI’s cult of objectivity is dangerous, and how we can be better - "More than anything, AI is a mirror: something built in the image of humans, built to mimic humans, and thus inherit our flaws."
AI fails its job interview - "Our research suggests job seekers' experience with these AI-based interviews was poor."
Despite recent progress, AI-powered chatbots still have a long way to go - "Meta hasn’t given up on the idea of a human-like, AI chatbot experience — yet."
IRS To Ditch Biometric Requirement for Online Access - "The Internal Revenue Service (IRS) said today it will be transitioning away from requiring biometric data from taxpayers who wish to access their records at the agency’s website."
What Humans Lose When We Let AI Decide - "Our trust in AI leads us to confuse reckoning — decision-making based on the summing up of various kinds of data and technical images — with judgment."
MoviePass will track people’s eyes through their phone to make sure they see ads - "MoviePass head Stacy Spikes said that they wanted to avoid people putting down the phone and avoiding the ads"
Army Buys Artificial Intelligence-Infused Folding Quadcopters For Battlefield Use - "The drones will be able to scout ahead for ambushes, targets of opportunity, and greatly enhance troops' situational awareness on the front lines."
Pentagon’s AI center awards contracts to 79 companies in new test and evaluation agreement - "The Department of Defense’s artificial intelligence organization has awarded contracts 79 vendors to streamline the procurement and delivery of test and evaluation tools, with each contract worth up to $15 million."
Machines are getting better at writing their own code. But human-level is 'light years away' - "But computer scientist Dzmitry Bahdanau wrote on Twitter that human level coding is “still light years away.”"
Andrew Ng: Unbiggen AI - "The AI pioneer says it’s time for smart-sized, “data-centric” solutions to big issues"
Copyright © 2022 Skynet Today, All rights reserved.