Last Week in AI #157: Russia's AI-powered weapons, AI to prevent robocalls, concerns about EU's narrow definition of AI, and more!
AI-based voice screening tools can help identify illegitimate calls, EU's technology-specific definitions of AI risk making its regulations overly rigid
Russia’s war against Ukraine continues. Part of what’s tragic about the situation is the overwhelming superiority of Russia’s military (the 3rd largest) over Ukraine’s in terms of manpower, equipment, and overall funding — AI-enabled weaponry included. Russia is using multiple types of drones which are used to destroy remote ground targets, deliver payloads onto a target’s coordinates that are set manually or in the image from the drone’s guidance system, and self-destruct when hitting its target. On top of that, Russia is also using or working on unmanned ground vehicles, crewless naval and undersea vehicles, and AI-guided missiles that could decide to switch targets mid-flight.
Our take: AI has not revolutionized warfare, but it has certainly impacted it. Unmanned and autonomous weapons will no doubt make it possible for wars to be even more relentless and destructive. Hopefully, part of the outcome of these events will be additional pressure to create regulations and conventions for this sort of technology.
The number of robocalls in the U.S. rose steadily through the last decade. To combat this problem, the FCC mandated all phone providers to verify caller IDs. However, this system is not very effective. Some companies are now working on AI-based call screening tools that act on the recipients’ phones to prevent robocalls. These are essentially voice assistants that ask unfamiliar numbers questions like “who do you want to speak to?”. Based on the answer from the caller and the way the answer is delivered, a machine learning algorithm can guess the legitimacy of a call and whether or not to forward it to the user. Researchers show such approaches can achieve 90%+ accuracy in classifying spam calls, and some are working to deploy these as apps.
Our take: It might be unreasonable to expect the world to be free of robocalls, and these prevention methods will definitely have weaknesses that can be exploited. For example, a malicious actor could deploy its own AI voice chatbot to convincingly answer voice prompts. However, the hope is that doing so at scale across millions of calls will incur significant costs that will simply make it impossible to profit from robocalls.
The EU's Draft Artificial Intelligence Act (AIA) represented one of the first serious attempts at comprehensive legislation dealing with AI systems. While it had a number of faults, many also lauded it for its forward thinking. Naturally, the question of what exactly is being regulated--what "Artificial Intelligence" is--will be an important one to the legislation. While the AIA currently operates with a broad definition, member states are pushing back and think that systems being legislated should be capable of “learning, reasoning, or modeling implemented with the techniques and approaches” listed in an annex, and they are also “generative systems.” But it is known that generative models, for instance, form just a fraction of all AI models and certainly of all potentially concerning AI systems.
Our take: Bryson, the author of this WIRED article, has it right. A broad definition of AI systems will be far more effective and robust to special interests than the narrow definition being proposed by EU member stats. Bryson notes that a good definition of intelligence for the AIA is one that allows regulators to focus on consequences without getting bogged down in technical details--for instance, “the computing of appropriate action from context." While the criteria are fair, this is a definition that likely could be applied to many computer programs that don't use AI techniques. Regulators wishing to turn their attention to AI systems should be precise in defining those systems and the consequences they wish to consider, and making definitions too broad is another potential pitfall.
Stony Brook University Researchers Introduce 'StandardSim' : A Large-Scale Photorealistic Synthetic Dataset Featuring Annotations For Semantic Segmentation, Instance Segmentation, Depth Estimation, And Object Detection - "The dataset has more photos and annotations for more tasks than existing datasets for retail and change detection."
Stanford University use AI computing to cut DNA sequencing down to five hours - "A Stanford University-led research team has set a new Guinness World Record for the fastest DNA sequencing technique using AI computing to accelerate workflow speed."
AI outdoes radiologists when it comes to identifying hip fractures, study shows - "When researchers pitted machine learning against human radiologists, the computer won, classifying hip fractures 19 percent more accurately than human experts."
Can machine-learning models overcome biased datasets? - "A model’s ability to generalize is influenced by both the diversity of the data and the way the model is trained, researchers report."
Machine learning and advanced imaging improve prediction of heart attacks – Physics World - "A machine learning model that combines data from two advanced imaging techniques can predict a patient’s risk of a future heart attack better than conventional clinical assessment."
Machine Learning Can Help Predict Severe COVID-19 - "Data collected based on nearly 300 COVID-19 patients shows machine learning techniques to be 88.5 percent accurate with respect to predicting the severity of the disease."
UC Berkeley robot navigation could chart a new-course for self-driving systems - "An approach that uses hours of video and elements of reinforcement learning can guide a robot vehicle almost two miles without a full map of the surroundings."
Viz.ai software gains FDA clearance for automatically spotting brain aneurysms - "After receiving regulatory clearances for its artificial intelligence programs for quickly spotting a stroke, Viz.ai has secured a new FDA green light aimed at another life-threatening brain condition: cerebral aneurysms."
New Machine Learning Model Flags Abnormal Brain Scans in Real-time - "Researchers from King's College London have developed a deep learning framework based on convolutional neural networks to flag clinically relevant abnormalities at the time of imaging, in minimally processed, routine, hospital-grade axial T2-weighted head MRI scans."
First AI research center inaugurated in Republic of Congo - "The Prime Minister of the Republic of the Congo, Anatole Collinet Makosso, inaugurated Thursday the African Center for Research in Artificial Intelligence (CARIA), the first of its kind in the country."
The Pentagon is working on an algorithm to detect Covid early - "Preliminary results from an experiment using fitness trackers show promise."
PolyCoder is an open source AI code-generator that researchers claim trumps Codex - "While PolyCoder doesn’t match the performance of top code generators in every task, the researchers claim that PolyCoder is able to write in C with greater accuracy than all known models, including Codex."
How vacation photos of zebras and whales can help conservation - "Vacation photos of zebras and whales that tourists post on social media may have a benefit they never expected: helping researchers track and gather information on endangered species."
How GitHub Uses Machine Learning to Extend Vulnerability Code Scanning - "Applying machine learning techniques to its rule-based security code scanning capabilities, GitHub hopes to be able to extend them to less common vulnerability patterns by automatically inferring new rules from the existing ones."
AI Machines Have Beaten Moore's Law Over The Last Decade, Say Computer Scientists - "Computational performance has followed Moore's Law since the dawn of the computer age. Not any more."
How AI is shaping the cybersecurity arms race - "AI can help detect patterns within large quantities of data that human analysts can’t see. "
Companies Borrow Attack Technique to Watermark Machine Learning Models - "Researchers continue to improve on a technique for embedded crafted outputs into machine-learning models, an anti-copying technique originally thought up by adversarial researchers."
This machine-learning model can pinpoint failing or hacked power grid components - "Machine learning could one day help energy providers better pinpoint failing or compromised components in power grids, or better identify traffic congestion for local authorities, according to a study."
First Wholly AI-Developed Drug Enters Phase 1 Trials - "For several years we have been hearing about the potential of Artificial Intelligence (AI) to improve traditional drug discovery and development. In the last two years, clinical trials have begun."
Mercedes-Benz to Offer Hands-Free Traffic Jam Assist in US This Year - "The German automaker's Drive Pilot offers Level 3 conditional autonomy that works at low speeds in heavy congestion."
Companies to Increase Spends on AI Solutions this Year, Says IDC - "The global artificial intelligence (AI) market is projected to grow by almost a fifth to reach $432.8 billion in 2022, according to a new report from IDC. The analyst firm expects spending on AI to increase by 19."
Facial recognition company Clearview AI seeks first big deals, discloses research chief - "Clearview AI, whose search engine for faces has become an unrivaled police tool, this year is aiming to win its first big U.S. government contracts and expand its team by a third even as the startup fights challenges in the courts and Congress, its chief executive told Reuters."
Neosapience gets $21.5M to use AI-powered synthetic avatars for creators - "Artificial intelligence-powered voice and video technologies have been gaining steady popularity in recent years. Korean startup called Neosapience has developed a synthetic voice and video platform, Typecast, that lets users turn text into a video without recording and editing in a studio. "
Report: 70% of orgs are spending $1M or more on AI - "Enterprises are using AI to innovate, scale up and drive competitive advantage as well as gain internal efficiencies."
Kraft-Heinz and NotCo Form Joint Venture for AI-Powered Food Products - "Today Kraft-Heinz and NotCo, the food tech company behind the NotCo brand of plant-based foods, announced they are forming a joint venture to develop a lineup of plant-based food products."
Meta's Zuckerberg unveils AI projects aimed at building metaverse future - "Meta is working on artificial intelligence research to generate worlds through speech, improve how people chat to voice assistants and translate between languages, CEO Mark Zuckerberg said on Wednesday, as he sketched out key steps to building the metaverse."
RightHand cashes in on the white-hot warehouse robotics space with a $66M raise - "Funding will go toward the usual things: hiring, getting more office space and further scaling globally."
QuantrolOx uses machine learning to control qubits - "QuantrolOx, a new startup that was spun out of Oxford University last year, wants to use machine learning to control qubits inside of quantum computers."
Baidu beats quarterly revenue estimates on AI, cloud services - "Chinese search engine giant Baidu Inc beat quarterly revenue Wall Street estimates on Tuesday powered by growth in its artificial intelligence (AI) cloud business. Revenue grew to 33.09 billion yuan ($5.24 billion) for the fourth quarter to Dec. 31, beating the 32."
Shell sees AI as fuel for its sustainability goals - "Energy giants are under significant pressure by governments and consumers to reduce carbon emissions. For multinational oil and gas company Shell, artificial intelligence may be a key catalyst for fulfilling that long-term goal."
Baidu plans 100-city robot taxi rollout by 2030 - "Chinese web giant Baidu has revealed plans to introduce its autonomous taxi service to 65 cities by the year 2025, then add another 35 cities by 2030."
Walmart’s Choose My Model Helps Shoppers Try on Clothes Virtually - "Walmart launched Choose My Model, a virtual tool powered by computer vision and artificial intelligence on the company’s website and app that allows shoppers to pick a person who resembles their height, shape and skin tone to show how clothes would fit, according to a (March 2) press release."
AI computer maker Cerebras nabs TotalEnergies SE as first energy sector customer - "The energy giant, a devotee of giant computers, is enthusiastic about the early results of running very large, data-intensive simulation tasks such as oil and gas exploration."
Starship Technologies Raises $100M in 30 Days, as Autonomous Delivery Demand Tripled in 2021 - "Starship Technologies today announced that it has raised $100 million in the past 30 days, including a $42 million Series B round."
Law Firms Turn to AI to Vet Recruits, Despite Bias Concerns - "New York’s Cadwalader passed over a law student vying for a summer job until an artificial intelligence algorithm flagged her as a good match."
MLOps Is a Mess But That's to be Expected - "Does this sound familiar? You read an article that said doing machine learning was the job to get in 2022, being not only crazy in-demand but commanding among the highest industry salaries around. That sounds nice: job security and money. What’s not to like?"
This is the reason Demis Hassabis started DeepMind - "A year after it took biologists by surprise, AlphaFold has changed how researchers work and set DeepMind on a new course."
A Decade of Deep Learning: How the AI Startup Experience Has Evolved - "In this interview, Socher discusses a number of topics, including: how things have changed for AI startups in the last decade; the differences between doing AI in startups, enterprises, and academia; and how new machine learning techniques, such as transformer models, empower companies to build advanced products with a fraction of the resources they would have needed in the past."
The US Copyright Office says an AI can’t copyright its art - "The US Copyright Office has rejected a request to let an AI copyright a work of art. Last week, a three-person board reviewed a 2019 ruling against Steven Thaler, who tried to copyright a picture on behalf of an algorithm he dubbed Creativity Machine."
Tesla's Autopilot feature is reportedly being investigated by German regulators - "Tesla's automated lane changing system, known as Autopilot, is being probed by German regulators, newspaper Bild am Sonntag reported Sunday, citing a spokesperson for the agency."
China Is About to Regulate AI—and the World Is Watching - "Sweeping rules will cover algorithms that set prices, control search results, recommend videos, and filter content."
Court rules that Waymo can keep its robotaxi emergency protocols a secret - "The California Superior Court in Sacramento has ruled in favor of Waymo, allowing the company to keep specific details about its autonomous vehicle technology a secret."
My mostly boring views about AI consciousness - "People have been asking if current ML systems might be conscious. I think overly strong answers to this in both directions include "certainly not" and "sure, but so might atoms" as well as almost any variant of "yes"."
Copyright © 2022 Skynet Today, All rights reserved.