Last Week in AI #170: Tricking autonomous vehicles with phantom obstacles, AI to track and prevent deforestation, Google limits deepfakes in Colab, and more!

Jun 6, 2022

UCI Researchers: Autonomous Vehicles Can Be Tricked Into Dangerous Driving Behavior

Researchers at the University of California, Irvine have designed a testing tool, called PlanFuzz, which can automatically detect security vulnerabilities in widely used automated driving systems. These security vulnerabilities are part of the planning module of autonomous driving systems which decides when to change lanes and slow down and stop, among other functions. They used their system to evaluate autonomous driving systems Apollo and Autoware. The researchers found that ordinary objects, like cardboard boxes and bicycles placed on the side of the road, caused vehicles to permanently stop on otherwise empty roads and intersections. They also found that by perceiving nonexistent threats from such ordinary objects, autonomous cars did not change lanes as planned. While acknowledging the need to be cautious for autonomous vehicle companies like Uber and Tesla, they stress that even overly cautious behavior can also be a cause of concern for road safety.

Our Take: Recognizing objects on roads and roadsides is fundamental to safety in autonomous driving. We applaud the motivation behind the development of PlanFuzz. In our opinion, the sim-to-real gap is of particular interest. Particularly, simulators can be tuned using a diverse set of parameters to exhibit a wide range of behaviors, some of which are shown in the videos by the authors to be conservative. On the other hand, we know that vehicles already on the road, like those from Cruise and Waymo, can easily drive past trash cans and traffic cones, the same objects tested in the study. That being said, we do believe there is potential for PlanFuzz to make AVs safer by identifying causes of phantom braking and triaging the risk level of unknown objects on the roads.

Satellites and AI Can Help Solve Big Problems—If Given the Chance

Using AI models to analyze large-scale satellite imagery data has many applications that can improve the world. Examples include tracking and identifying deforestation in the Amazon rainforest, spotting illegal fishing ships that turn off GPS tracking. monitoring new trails created by off-road vehicles, predicting floods, and recognizing irrigation patterns. However, making these AI observations is just the first step. Organizations that aim to leverage these predictions for public good often run into bureaucratic red tapes, lax law enforcement, and little political support.

Our take: This article gives yet another example of how AI and technology can’t magically fix real-world problems on their own. At the end of the day, individuals and organizations need to take the predictions produced by these models and turn them into action, something the AI models cannot help with. However, it is nonetheless exciting to see how recent AI advances have been widely adopted by many organizations across the world, and there is definitely a lot of potentials for AI to do good in the future.

Google bans deepfake-generating AI from Colab

Over the years, Google Colab has become a de-facto demo platform for the AI community. Many courses use Colab for assignments, and image generation tools like CLIP+VQGAN are easily usable with Colab. Recently, a few organizations found that Google will no longer allow deepfake-generating AI on its Colab. This is not yet a wholesale ban but appears to be based on a blacklist for certain code. While the popular DeepFaceLab can no longer be used, other popular tools like FaceSwap are still fully operational.

Our take: Concerns about deepfakes have persisted for years by this point, and many have called for restrictions on their use. Google's move represents an attempt to limit open use of deepfakes, but, as UW adjunct professor Os Keys notes, does not back that action with policy. While it might seem beneficial to impose a ban on deepfakes from widely-used platforms like Colab, such a ban doesn't seem likely to substantially diminish their spread, given how easy it is to make code available--furthermore, far-reaching bans may just encourage more creativity in doing so. It seems likely that the battle between deepfakes and detectors will continue to be a game of cat and mouse. Pragmatically, it seems the best hope lies in establishing use norms and allowing researchers to study deepfake systems in order to counteract them.

Other News

Research

Caltech unit creates AI helping drones to withstand violent winds - "Caltech researchers are developing a drone with rapidly reacting artificial intelligence (AI) capacities that allow it to adapt in flight to extreme wind similar to tornado or hurricane conditions."

UC Berkeley’s Automated Crossword Solver Achieves 99.9% Letter Accuracy, Wins Top Tournament - "For more than a century, the crossword puzzle has been one of the world’s most popular games for testing vocabulary, trivia, common sense and wordplay — all while requiring players to reason out multiple intersecting answers."

The Man With 100,000 Brains: AI’s Big Donation to Science - "Jorge Cardoso wears many hats, and that’s appropriate given he has so many brains. A hundred thousand of them to be exact. Cardoso is a teacher, a CTO, an entrepreneur, a founding member of the MONAI open source consortium and a researcher in AI for medical imaging."

Nvidia’s AI-powered supercomputers advance nuclear fusion research - "The most powerful supercomputers on the planet are used to perform all manner of complex operations. Increasingly, they are used to enable artificial intelligence for research that could one day impact billions of people."

$10.5M Army Contract to CMU Lab Will Expand Use of AI in Predictive Maintenance - "The School of Computer Science's Auton Lab will use $10.5 million from the U.S. Army to lead a multi-institution effort to extend the capabilities of artificial intelligence in predictive maintenance."

China's AI defense can reportedly predict the course of hypersonic missiles - "Researchers at the Air Force Early Warning Academy in China have claimed to have developed artificial intelligence-based technology that can predict the course of hypersonic missiles even as they approach their target at five times the speed of sound, South China Morning Post (SCMP) reported."

AI model’s insight helps astronomers propose new theory for observing far-off worlds - "Machine learning models are increasingly augmenting human processes, either performing repetitious tasks faster or providing some systematic insight that helps put human knowledge in perspective."

Applications

Ex-golf pro links with Seattle-area AI experts on app that uses 3D motion analysis to improve game - "Before you take yet another mulligan, consider that the tool to improving your golf game might not be found in your bag of clubs."

How artificial intelligence 'blew up' tennis - "Tennis is such a technical sport. These days, anyone I play or coach is into tech, be it video analysis or longest rally stats."

How AI brought Val Kilmer’s ‘Iceman’ back into Top Gun: Maverick - "Artificial intelligence landed a major part in the new Top Gun sequel: Val Kilmer’s voice. Scientists helped Kilmer reprise his role as Iceman by using A.I. to craft a computer-generated replica of the actor’s voice that could read his lines."

AI could help us spot viruses like monkeypox before they cross over - "When a new coronavirus emerged from nature in 2019, it changed the world. But COVID-19 won’t be the last disease to jump across from the shrinking wild."

Machine Learning Shows That More Reptile Species May Be at Risk of Extinction Than Previously Thought - "Machine learning tool estimates extinction risk for species previously unprioritized for conservation. Species at risk of extinction are identified in the iconic Red List of Threatened Species, published by the International Union for Conservation of Nature (IUCN)."

Dutch police create deepfake video of murdered boy, 13, in hope of new leads - "Dutch police have received dozens of leads after using deepfake technology to virtually bring to life a teenager almost two decades after his murder. Sedar Soares was shot dead in 2003 while throwing snowballs with friends in the parking lot of a Rotterdam metro station."

How Deep Squeak, an AI program with a weird name, is detecting whales - "Artificial Intelligence is booming. And now an A-I program is being used to search for whales."

Energy Grids Plug into AI for a Brighter, Cleaner Future - "Electric utilities are taking a course in machine learning to create smarter grids for tough challenges ahead. The winter 2021 megastorm in Texas left millions without power. Grid failures the past two summers sparked devastating wildfires amid California’s record drought."

Depressed? This algorithm can tell from the tone of your voice - "Mental health issues have come into a clearer focus amid the pandemic. Depression became endemic, but it still too often goes undetected. Even when it does, health care providers struggle to meet demand."

Can machine learning deliver one-minute brain MRI scans? - "With a few enhancements, including machine learning, a quantitative technique called MR fingerprinting could make a one-minute clinical brain MRI scan a reality, according to a talk presented at the recent International Society for Magnetic Resonance in Medicine (ISMRM) meeting in London."

The Future of the Web Is Marketing Copy Generated by Algorithms - "As we move further into the 21st century, more and more aspects of our lives are being controlled by algorithms."

AI Reveals How the Brain’s Anatomy Changes With Autism - "There’s still a lot to unravel about autism, but we now have a better sense of what parts of the brain are acting differently."

Business

Robots Pick Up More Work at Busy Factories - "Robots are turning up on more factory floors and assembly lines as companies struggle to hire enough workers to fill rising orders. Orders for workplace robots in the U.S."

US robot orders surge 40% as labor shortages, inflation persist - "Orders for workplace robots in the United States surged 40% year-over-year in the first quarter of 2022 as companies are leveraging automation to combat ongoing labor shortages and cut costs as inflation continues to hover near a 40-year-high."

Mintlify taps AI to automatically generate documentation from code - "Mintlify, a startup developing software to automate software documentation tasks, today announced that it raised $2.8 million in a seed round led by by Bain Capital Ventures with participation from TwentyTwo Ventures and Quinn Slack, Sourcegraph’s co-founder."

Chick-fil-A tests autonomous delivery in Austin, Texas - "Two Chick-fil-A restaurants in Austin, Texas, will test Refraction AI’s self-driving vehicles for delivery, Refraction AI announced on Tuesday."

IBM adds side order of NLP to McDonald's AI drive-thru chatbots - "IBM says it is rolling out its natural language processing software to a greater number of McDonalds' drive-thrus months after buying the automated order technology unit from the fast food chain, along with the team that developed it."

Inaugural Day of AI brings new digital literacy to classrooms worldwide - "The first annual Day of AI on Friday, May 13 introduced artificial intelligence literacy to classrooms all over the world."

Alimentation Couche-Tard to deploy over 10,000 AI-powered cash registers at stores - "Convenience store retailer Alimentation Couche-Tard Inc. says it will deploy more than 10,000 touchless self-serve cash registers within three years at over 7,000 Couche-Tard and Circle K stores."

Imagene nets $21.5M for cancer biopsy-scanning AI - "The AI startup aims to help direct cancer patients to the proper targeted therapies using only digital images of biopsied tissue."

Meta reorganization aims to decentralize Facebook's AI efforts - "Facebook parent Meta announced a broad reorganization of its AI efforts Thursday, a move that will embed more work within product teams and will also see a top AI executive leave the company."

Concerns

A Face Search Engine Anyone Can Use Is Alarmingly Accurate - "PimEyes is a paid service that finds photos of a person from across the internet, including some the person may not want exposed. “We’re just a tool provider,” its owner said. For $29."

Think you can spot content written on AI? The truth is you've probably already read a lot of it - "Two years ago this weekend, GPT-3 was introduced to the world and although you may not have heard of it there's a good chance you've read its work. Or you may have used a website that runs GPT-3 code, or even conversed with it through a chatbot or a character in a game."

AI-generated images could make it almost impossible to detect fake papers - "I think we have reached the point where we can no longer tell if the paper is real or fake"

Tension Inside Google Over a Fired AI Researcher’s Conduct - "Google employees claim a senior researcher fired earlier this year sought to undermine two more junior AI researchers by suggesting their results were wrong or even falsified."

We're Publishing the Facebook Papers. Here's What They Say About the Ranking Algorithms That Control Your News Feed. - "Facebook employees knew that a computer-curated feed increased the time users spent on the social network—and that it led to unhealthy behaviors."

Eticas Foundation external audits VioGén: Spain’s algorithm designed to protect victims of gender violence - "VioGén is an algorithm that determines the level of risk faced by a victim of gender-based violence and establishes her protection measures in Spain. It is the largest risk assessment system in the world, with more than 3 million registered cases."

Who’s liable for AI-generated lies? - "The dark side of large language models can't be ignored… Who will be liable for harmful speech generated by large language models? As advanced AIs such as OpenAI’s GPT-3 are being cheered for impressive breakthroughs in natural language processing and generation — and all sorts of (productive)"

Oregon dropping AI tool used in child abuse cases - "Child welfare officials in Oregon will stop using an algorithm to help decide which families are investigated by social workers, opting instead for a new process that officials say will make better, more racially equitable decisions "

Policy

NATO’s AI Push And Military Implications – Analysis - " The technological advancements in Artificial Intelligence (AI), machine learning, big data analytics, robotics, quantum computing and virtual reality have led to the rise in use of autonomous systems in military applications."

What role for regulators in the developing a creditable AI audit industry? - "AI audits are used to check and verify that algorithmic systems are meeting regulatory expectations and not producing harms (either unintended or intended). Globally the regulatory requirements for AI audits are rapidly increasing"

Federal AI task force outlines ‘equitable future’ for R&D resources - "The task force behind the National AI Research Resource (NAIRR) issued an interim report Wednesday outlining how the AI data-and-research hub can put federal AI resources in the hands of more U.S. researchers. A final report is expected this November."

With New Hires, Pentagon’s AI Office Reaches Full Operating Capability - "Stood up earlier this year, the Defense Department’s Office of the Chief Digital and Artificial Intelligence Officer, or CDAO, reached full operating capability Wednesday with the appointment of two new personnel."

Pentagon Launches ‘Third Wave’ AI Initiative to Help Warfighting Effort - "A key research and development agency within the Department of Defense is launching a new program specifically focused on advancing algorithmic processing within Defense’s artificial intelligence projects."

Fun

A Guide To Asking Robots To Design Stained Glass Windows - "I love stained glass. Not so much your usual suburban house stained glass with a picture of lilies. The good stuff. Cathedral windows, Art Nouveau, Art Deco. Why did we stop doing that? I blame the conspiracy. Recently I’ve been experimenting with small-scale alternatives."

AI in Fiction and the Future of War - "Interestingly, AI is also often the hero when AI is the villain in the story. In Tron, released in 1982, the Master Control Program (MCP) has most of the same characteristics as Skynet, including an aim to obtain control over nuclear weapons."

AI versus corporate logos - "I recently started playing with DALL-E 2, which will attempt to generate an image to go with whatever text prompt you give it. Like its predecessor DALL-E, it uses CLIP, which OpenAI trained on a huge collection of internet images and nearby text."

Last Week in AI