Top News

The AI Defending Ukraine

Clearview AI's latest journey takes it to Ukraine, where the besieged country has decided to use the facial recognition tool in its defense. Clearview augmented its existing dataset of 10 billion images with another 2 billion from Russian social network VKontakte, giving its tool the ability to identify Russian soldiers and find their addresses and families. Ukraine initially refused Clearview's service, but later accepted it as Russian war crimes and propaganda ramped up. While even Clearview CEO Hoan Ton-That doesn't know exactly what Ukraine is using his company's tech for, journalist Will Lockett has a few theories: these include undermining Russian propaganda and identifying Russian soldiers who commit war crimes.

Our Take: As with any use of Clearview, Ukraine's is bound to be controversial. That being said, I think it's certainly interesting that Clearview offered its services to Ukraine, and potential use cases like these make it difficult for me to assume that Clearview and its product are inherently irredeemable. Indeed, some very good outcomes could result from using technology for the purposes Lockett noted, even if Clearview's system is not perfect. Furthermore, as Lockett observes, Ukraine seems to be using Clearview with caution. The true danger of tools like Clearview's is in putting too much trust in them and using their predictions to make high-stakes decisions. But at the same time, Ukraine's choice could open a "Pandora's Box" of using similar AI systems in warfare. Only time will tell what those uses look like.

Can Machine Learning Translate Ancient Egyptian Texts?

Four large ancient Egyptian hieroglyphs are engraved on the wall of a temple with columns in the background.

Ubisoft has developed the Hieroglyphics Initiative, which aims to help researchers decode Egyptian hieroglyphs using machine learning. Initially, translators used paper and pencil to sketch or trace inscriptions on tomb walls. More recently, large-format photography revolutionized the field. Under the Hieroglyphics Initiative attempts, Google Arts and Culture released an open-source program, where researchers can upload a photo of text taken from inscriptions on tomb walls and create a facsimile layer of the image, and zoom in on a specific section. The AI-powered software then sequences and groups the hieroglyphs into categories such as birds, reptiles, or human body parts, and helps identify each hieroglyphic sign using Gardiner’s Sign List. These outputs are then used by researchers to better analyze the texts.

Our Take: Using AI for preserving and interpreting historical texts is a useful application. However, the open-source system is less than 30% accurate in detecting and translating the hieroglyphics. Moreover, even SOTA software like Google Translate can not translate the hieroglyphics as Egyptian Grammar is not that easy to understand, and hieroglyphic signs can have multiple meanings. Hence, this suggests that a lot more training data is required for machine learning algorithms to make sense and translate accurately. Moreover, incorporating knowledge and intuition from Egyptologists in how they translate hieroglyphics can also be helpful for more accurate learning algorithms.

Papercup raises $20M for AI that automatically dubs videos

Dubbing videos is an expensive and laborious process, costing as much as $75 per minute even for simple videos. Ideally, AI-powered solutions can help to automatically dub content in different languages while preserving the voice and emotions of the original actors. Papercup aims to do just that, and it recently translated all 30 seasons of Bob Ross’s “The Joy of Painting.” The company’s latest fundraising will allow it to research more advanced AI dubbing technology for more expressive voices and support more languages. Concerns remain regarding the ethics of recreating the voices of actors both living and deceased, and whether or not AI-dubbed content can retain subtle emotions and meanings expressed in the original language.

Our take To be fair, AI-dubbing doesn’t have to be perfect - even human-produced dubs or subs have inaccuracies and often succumb to challenges in dialogues with tricky local contexts. In addition, there are clear economic incentives for automatic translations, even if the output needs to be reviewed or revised by native speakers. Expect to see more companies in this space and in synthetic content in general.

Other News

Research

AI Learns Coral Reef ‘Song’ - "A new artificial intelligence algorithm trained using the soundscapes of both healthy and degraded reefs can determine reef health 92% of the time."

E-skin that can feel pain could create new generation of touch-sensitive robots - "An electronic skin which can learn from feeling ‘pain’ could help create a new generation of smart robots with human-like sensitivity."

Creating artificial intelligence that acts more human by 'knowing that it knows' - "A research group from the Graduate School of Informatics, Nagoya University, has taken a big step towards creating a neural network with metamemory through a computer-based evolution experiment."

A new optical brain-like chip can process 2 billion images per second - "University of Pennsylvania researchers have invented a new optical chip that can process over 2 billion photos per second. The gadget consists of a neural network that processes data as light without the use of components that slow down standard computer chips, such as memory."

Student-powered machine learning - "From their early days at MIT, and even before, Emma Liu '22, MNG '22, Yo-whan “John” Kim '22, MNG '22, and Clemente Ocejo '21, MNG '22 knew they wanted to perform computational research and explore artificial intelligence and machine learning."

Artificial intelligence reveals a never-before described 3D structure in rotavirus spike protein - "Of the three groups of rotavirus that cause gastroenteritis in people, called groups A, B and C, groups A and C affect mostly children and are the best characterized."

DALL·E mini, an open source alternative to DALL·E - "Surely since DALL·E 2 came out you’ve been wanting to try it out, but since it’s not available to everyone you have been left wanting. In this post we offer you a way to try it: DALL·E mini, an open source alternative to DALL·E. But first things first."

Applications

Farm Robots Will Solve Many of Our Food Worries - "A robot army is beginning its march across rural America, promising to transform the future of food. Twenty-five intelligent machines were dispatched last month to the Midwest and the Mississippi Delta, where they will advance over newly planted fields at 12 miles an hour, annihilating baby weeds."

Many Parents OK With Use of AI to Help Treat Child With Respiratory Illness in ED - "Parents are generally receptive to computer-assisted management of children with respiratory illnesses in the emergency department though some express concerns, according to a study published online May 13 in Academic Pediatrics. Sriram Ramgopal, M.D."

How AI could be used to detect guns before school shootings - "With little progress on gun control measures in Congress, some envision next-generation weapons detection technologies as a potential deterrent to mass shootings."

Machine learning helps determine health of soybean fields - "Using a combination of drones and machine learning techniques, researchers from The Ohio State University have recently developed a novel method for determining crop health and used it to create a new tool that may aid future farmers."

Detroit police are using AI surveillance checkpoints to target gun possession this summer - "The Detroit Police Department is introducing new artificial intelligence technology to identify people with firearms, with a plan to place security checkpoints in various outdoor locations this summer."

How artificial intelligence brings new superpowers to supercomputers - "The world’s fastest and most powerful supercomputers are capable of doing many things, but increasingly the world of high performance computing (HPC) is leaning in on artificial intelligence (AI)."

Google says it used machine learning to massively improve Chrome's phishing detection - "Google has long been employing machine learning to improve its products, including Chrome. The company says it has been able to make the browser safer to use. In March, Google rolled out a new on-device machine learning model that's able to detect 2."

Stunning Insights from James Webb Space Telescope Are Coming, Thanks to GPU-Powered Deep Learning - "NVIDIA GPUs will play a key role interpreting data streaming in from the James Webb Space Telescope, with NASA preparing to release next month the first full-color images from the $10 billion scientific instrument."

AI Approach Offers Higher Rate of Diabetes Remission Than Standard Care - "Researchers found that an artificial intelligence-based intervention offered significant rates of remission compared to standard care alone for type 2 diabetes patients."

Business

AI-driven robot boat Mayflower crosses Atlantic Ocean - "A crewless ship designed to recreate the Mayflower's historic journey across the Atlantic 400 years ago has crossed the ocean, project bosses have said. The Mayflower Autonomous Ship (MAS) completed a 2,700-mile (4,400km) trip from Plymouth in the UK to Halifax in Nova Scotia, Canada, on Sunday."

Agtech robotics firm FarmWise just raised another $45 million - "The rest of the startup universe may be struggling to bring in funds, but it’s still a good time to get a robotic raise. Agtech is high on that list. The median age of farmers is 55 years old in the United States, and finding human help is increasingly more difficult of late."

Picsart Adds AI Enhancement Tool to Upscale and Improve Photos - "Picsart has announced a new artificial intelligence (AI) tool that it says removes noise, upscales images to higher resolution, and improves overall image quality."

How Machine Learning Is Changing the Store Paradigm - "Changing demands in the retail environment have created all-new operational challenges, as retailers strive to deliver exceptional omnichannel customer experiences."

Insilico Medicine plans to launch a fully automated AI-driven robotics lab for drug discovery - "Insilico Medicine's end-to-end AI-driven drug discovery platform, Pharma.AI, draws its strength from the quality and quantity of its data. "

This AI attorney says companies need a chief AI officer — pronto - "When Bradford Newman began advocating for more artificial intelligence expertise in the C-suite in 2015, “people were laughing at me,” he said."

Shield AI raises $165M at a $2.3B valuation to fuel development of its military autonomous flying systems - "Technology built with defense in mind is getting some significant and serious traction at the moment, spurred by world events, advances in technology and a growing appetite from end users to invest in more innovative ways to protect themselves."

Walmart's bid to challenge Amazon on delivery expands as it quietly tests robots from Uber-backed Serve Robotics - "Walmart continues to pump resources into its evolving delivery strategy as it takes on e-commerce giant Amazon."

Alexandr Wang is the new youngest self-made billionaire - "The son of Chinese immigrants to the US, Wang dropped out of college after his freshman year to found Scale Al, a company that helps companies like Uber, Toyota, and Airbnb optimize their artificial intelligence systems."

Concerns

The Long, Hype-Strewn Road to General Artificial Intelligence - "Last month, DeepMind, a subsidiary of technology giant Alphabet, set Silicon Valley abuzz when it announced Gato, perhaps the most versatile artificial intelligence model in existence."

‘AI’ Shopping Startup Exaggerated Tech Capabilities to Potential Investors - "Some startups are bold and original. And some, like Nate, had more modest goals: automatically filling out shoppers’ contact and payment information on retailers’ websites. In exchange for sparing them a minute or two of data entry on their phones, Nate charged shoppers $1 per transaction."

Phishing attacks will use powerful text generation, say machine-learning engineers - "Armorblox engineers will discuss how machine learning will be used in phishing attacks at the RSA Conference on June 7."

How Safe Are Systems Like Tesla’s Autopilot? No One Knows. - "Automakers and technology companies say they are making driving safer, but verifying these claims is difficult."

AI Trained on 4Chan Becomes ‘Hate Speech Machine’ - "AI researcher and YouTuber Yannic Kilcher trained an AI using 3.3 million threads from 4chan’s infamously toxic Politically Incorrect /pol/ board."

Axon Pauses Plans for Taser Drone as Ethics Board Members Resign - "After Axon announced plans for a Taser-equipped drone that it said could prevent mass shootings, nine members of the company’s ethics board stepped down."

Analysis

We’re told AI neural networks ‘learn’ the way humans do. A neuroscientist explains why that’s not the case. - "Recently developed artificial intelligence (AI) models are capable of many impressive feats, including recognising images and producing human-like language. But just because AI can perform human-like behaviours doesn’t mean it can think or understand like humans."

Artificial General Intelligence Is Not as Imminent as You Might Think - "To the average person, it must seem as if the field of artificial intelligence is making immense progress."

Huge “foundation models” are turbo-charging AI progress - "They can have abilities their creators did not foresee"

How DALL-E could power a creative revolution - "Thoughts on my first week with OpenAI’s amazing text-to-image AI tool"

Policy

Pentagon’s New AI Chief Vows to Crack ‘Bureaucratic Inertia’ on Tech Advances - "The Pentagon’s new head of artificial intelligence wants to speed up technological modernization after an onslaught of what he calls “valid” criticism from recently departed senior leaders who expressed frustration at slow progress."

Europe’s Artificial Intelligence Debate Heats Up - "Europeans agree that they want to regulate AI. But they are divided on issues ranging from facial recognition and social scoring to the definition of AI. Each political group of the European Parliament has submitted several hundred amendments, bringing the total to several thousand."

Real-time facial recognition surveillance planned in Ireland, moratorium demanded - "New legislation is expected to open the door to the use of facial recognition within a range of surveillance technologies in Ireland, including CCTV cameras and police body cams, automatic number plate recognition (ANPR or LPR in the U.S.), according to reports by the Irish Times."

Explainers

What exactly is ‘AI-generated art’? How does it work? Will it replace human visual artists? - "Josh, I’ve been hearing a lot about ‘AI-generated art’ and seeing a whole lot of truly insane-looking memes. What’s going on, are the machines picking up paintbrushes now? Not paintbrushes, no."

Fun

What would Mona Lisa look like with a body? DALL-E 2 has an answer - "An experiment shows that DALL-E 2 can credibly adopt and continue the style of historical artists – even within their works."

This AI-powered Jack Nicklaus ‘twin’ will allow fans to interact with the golf legend - "Digital Jack, that is — an AI-powered interactive version of Nicklaus which, when it launches later this year, will be available to give fans, through their web browsers, a chance to quiz the Golden Bear about his career and seek out his advice about their own games."

Let's Force an AI to Design Us Some Cars - "Designing a car is hard. Our own Andy Kalmowitz learned this firsthand, after attempting to hand-render the upcoming Apple Car before getting yelled at in the office for not knowing who Richard Scarry is."

AI program DALL-E mini prompts some truly cursed images - "If you’ve seen some slightly distorted images on Twitter recently, it’s not just reality continuing to collapse on itself. An open AI program called DALL-E mini has overtaken Twitter in the last week, churning out a surreal stream of warped art."

Copyright © 2022 Skynet Today, All rights reserved.

Last Week in AI

Last Week in AI #171: Clearview AI in…