

Discover more from Last Week in AI
Last Week in AI #202: Generative models for protein discovery and voice imitation, Japan is trying to automate elder care, and more!
How conditioned diffusion models are used to generate new proteins, Microsoft tool can imitate a person's voice with just 3s of audio sample, Japan is making slow progress on elder care automation
Top News
A.I. Turns Its Artistry to Creating New Human Proteins
Conditional generative models made the news last year by bringing us impressive results of text-to-speech, text-to-video, text-to-audio, and so forth. Many of these models share an underlying technology for generative models - the conditioned diffusion model. For text-to-image, this means that the neural network iteratively converts an image of pure noise into an actual image, conditioned upon a textual description. Researchers are now applying the same technologies to generate blueprints for new protein designs, conditioned upon “descriptions” of the functions and characteristics of that protein. While the function of the generated protein design needs to be verified in an actual lab, this process does speed up protein discovery, and researchers are optimistic about its future applications.
Our take: AI development has a history of starting out in more “fun” applications - games, images, texts - before moving onto more “serious” applications, especially in the physical sciences. With technologies like DeepMind’s AlphaFold 2, researchers were able to quickly and accurately predict protein structure (which tells us about what the protein does) from existing mRNA sequences. Now with diffusion models, researchers are able to conjure up completely new proteins not found in nature and study their applications. Of course, all the usual caveats of deep learning apply here as well, but I’m excited to see what other creative uses of diffusion models we can find.
Microsoft’s new AI can simulate anyone’s voice with 3 seconds of audio
Last week, Microsoft released VALL-E, a new text-to-speech AI designed to closely match the quality of any person’s voice if given a three-second audio sample. The system builds off an audio compression technology known as EnCodec to analyze and break down a person’s voice into its fundamental components, which it then uses to create other phrases with a similar tone and sound. VALL-E can also mimic the environment of the audio sample, such as making it sound like it comes from a telephone. Microsoft has not opened the software for user testing, acknowledging the risks it could pose by impersonating speakers or manipulating audio. In hopes of combatting these, it is aiming to develop a system to detect if something was produced by VALL-E, though the full result of the technology is yet to be known.
Our take: While Microsoft is fortunately cognizant of how this tech could be misused, my main question is what their goals are in creating this technology. While the article mentions how VALL-E could be used for audio content generation, it’s not clear why such a system would need the ability to copy any person's voice rather than using a selection of pre-designated voices, like Siri does. Without a clear intended use, it’s hard to determine whether the potential benefits of this technology outweigh the security risks it could bring by proliferating deepfakes. The development of a VALL-E detection system may alleviate some of these problems, but it stands unknown how easily a sample could be manipulated to avoid being flagged. Rather than the “build now, find a purpose later” approach that seems to be growing in the AI research space, technologies like advanced audio mimicry ought to raise the question of whether we should aim to develop them at all.
Inside Japan’s long experiment in automating elder care
A long-standing question about our future with automated systems and robots is what can be automated, and what will remain for humans. Japan, facing an aging population and birth rates below replacement levels, has placed substantial investment in the development of "care robots" as the number of older people requiring care has increased, while care costs have also increased and the shortage of care workers is expected to intensify. But these care robots--accepted as a solution to the "problem" of elder care in Japan and equipped to provide physical aid or social and emotional support--fall very short of expectations in real life, prompting odd reactions from people and creating more work for care workers.
Our take: Care robots and their ilk are a classic figure to point to in debates about techno-solutionism and how much automated systems can take over the more "human" parts of society and everyday life. Wright points out in this Technology Review article paints a pessimistic picture of what it would take for care robots to become a "solution" to the crisis they confront: "employing more people with fewer skills, who would be paid as little as possible" in larger, highly standardized care facilities. I would tend to agree with James that, because care is "a shared social, political, and economic endeavor that ultimately relies on human relationships," a solution to the problems of care we will face in the future must be led with these factors, with high-quality care as the ultimate goal, in mind. Technology, we must remember, can be part of a solution for complex interdisciplinary problems.
Other News
Research
Feathered robotic wing paves way for flapping drones - "Birds fly more efficiently by folding their wings during the upstroke, according to a recent study. The results could mean that wing-folding is the next step in increasing the propulsive and aerodynamic efficiency of flapping drones.”
Deep Learning Algorithm Can Hear Alcohol in Voice - "New AI technology can instantly determine whether a person is above the legal alcohol limit by analyzing a 12-second clip of their voice."
Research could bring automatic speech recognition to 2,000 languages - "Only a fraction of the 7,000 to 8,000 languages spoken around the world benefit from modern language technologies like voice-to-text transcription, automatic captioning, instantaneous translation and voice recognition."
Machine learning tools help unravel why human accelerated regions evolved so quickly - "Humans and chimpanzees differ in only one percent of their DNA. Human accelerated regions (HARs) are parts of the genome with an unexpected amount of these differences. HARs were stable in mammals for millennia but quickly changed in early humans."
AI Is Taking On Ever-Larger Puzzles - "The incredible feat of developing Covid-19 vaccines so rapidly showcased science at its very best. But as we applauded the heroic effort of our health care workers in March 2022, one of my neighbors asked, “Why hasn’t AI helped?” A fair question."
Using machine learning to help monitor climate-induced hazards - "Combining satellite technology with machine learning may allow scientists to better track and prepare for climate-induced natural hazards, according to research presented last month at the annual meeting of the American Geophysical Union.Over the last few decades, rising global temperatures have caused many natural phenomena like hurricanes, sno..."
Is machine learning the answer for long-acting injectable drugs? - "Machine learning models used to guide the design of long-acting injectable drug formulations have been successfully tested by scientists at the University of Toronto."
Applications
Could robots do the work of master marble sculptors? This one is "99%" there, according to its creator. - "A remarkable innovation is changing the way timeless art is created."
AI-powered "robot" lawyer will be first of its kind to represent defendant in court - "A "robot" lawyer powered by artificial intelligence will be the first of its kind to help a defendant fight a traffic ticket in court next month."
Ryan Reynolds used ChatGPT to make a Mint Mobile ad, and the results were ‘mildly terrifying’ - "It was only a matter of time. 1 minute ReadOver the past month or so, OpenAI’s generative AI chatbot ChatGPT has gone viral for its ability to not only answer simple questions, but write essays or debug code, all in humanlike text."
This Voice Doesn't Exist - Generative Voice AI - "We’re deploying our own generative model which lets users design entirely new synthetic voices "
NVIDIA Broadcast will maintain eye contact even if you're looking away from the camera - "NVIDIA may have a solution if you've ever caught yourself reading chat during a livestream, or distracted by your notes during a video presentation."
AI improves detail, estimate of urban air pollution - "Using artificial intelligence, Cornell University engineers have simplified and reinforced models that accurately calculate the fine particulate matter (PM2.5) -- the soot, dust and exhaust emitted by trucks and cars that get into human lungs -- contained in urban air pollution."
Business
Microsoft in talks to acquire a 49% stake in ChatGPT owner OpenAI - "ChatGPT is currently one of the leading topics of discussion on the internet. The artificial intelligence application is being utilized to answer questions, write reports, and also formulate software codes."
Aerones raises $38.9M for robots to clean wind turbines - "Aerones, a company that creates robots that scrub and inspect wind turbines, brought in $38.9 million in funding"
Ghost Writer: Microsoft Looks to Add OpenAI’s Chatbot Technology to Word, Email - "In a move that could change how more than a billion people write documents, presentations and emails, Microsoft has discussed incorporating OpenAI’s artificial intelligence in Word, PowerPoint, Outlook and other apps so customers can automatically generate text using simple prompts"
AI may be red-hot, but Scale AI layoffs send shivers - "the unicorn company had laid off 20% of its 700-person staff"
OpenAI begins piloting ChatGPT Professional, a premium version of its viral chatbot - "OpenAI this week signaled it’ll soon begin charging for ChatGPT, its viral AI-powered chatbot that can write essays, emails, poems and even computer code."
BREAKING: Alphabet brings agtech startup out of stealth with data from 10% of world’s farmland, 3 major customers - "Alphabet, the parent company of Google, today publicly launches Mineral, an agtech business it’s been incubating in stealth within its so-called “moonshot factory” X since 2017."
Beautiful.ai Launches DesignerBot to Automatically Create Rich Presentations About Anything and Everything Using Generative AI - "Beautiful.ai, the presentation platform that enables anyone to build beautiful presentations, announced today the launch of DesignerBot, a generative AI designed to automate presentation creation and custom image generation."
DeepL targets AI translation for enterprises with fresh $100 million - "Seeking to target enterprise customers with AI language translation, Cologne, Germany-based DeepL announced a new funding raise that public reports estimate at well over $100 million. "
Turnitin is the go-to software to catch students cheating. Now it's focused on a potential cat-and-mouse game with OpenAI's new ChatGPT chatbot. - "Turnitin said they were able to detect AI writing because it's "extremely average." "
Alphabet Job Cuts Widen to Robotics Subsidiary - "A person close to the company said the layoffs represented nearly 20% of employees."
Germany's BioNTech buys British AI startup InstaDeep - "BioNTech SE has agreed to acquire British artificial intelligence (AI) startup InstaDeep for up to 562 million pounds ($682 million) to speed up its biotech research and manufacturing capabilities. "
Voice AI Company SoundHound Lays Off Nearly Half Its Staff and Offers ‘Pitiful’ Severance - "SoundHound, a voice AI company, laid off nearly half of its remaining staff last week—roughly 200 people—in a major company-wide downsizing, according to three employees who lost their jobs and an email from the CEO viewed by Gizmodo."
Google Cloud for Retailers adds AI-based inventory, e-commerce tools - "In a bid to help retailers transform their in-store, inventory-checking processes and enhance their e-commerce sites, Google on Friday said that it is enhancing Google Cloud for Retailers with a new shelf-checking, AI-based capability, and updating its Discovery AI and Recommendation AI services."
Concerns
Extensive' Tesla Autopilot probe proceeding 'really fast' -U.S. official - "The acting head of the National Highway Traffic Safety Administration (NHTSA) told reporters on Monday that the regulatory agency is "working really fast" on the Tesla (TSLA.O) Autopilot investigation it opened in August 2021."
Exclusive: Surveillance Footage of Tesla Crash on SF’s Bay Bridge Hours After Elon Musk Announces “Self-Driving” Feature - "Highway surveillance footage from Thanksgiving Day shows a Tesla Model S vehicle changing lanes and then abruptly braking in the far-left lane of the San Francisco Bay Bridge, resulting in an eight-vehicle crash."
ChatGPT Writes Well Enough to Fool Scientific Reviewers - "The internet’s new favorite toy, ChatGPT, accomplishes some things better than others. "
Don't Get Fooled Into Paying $7.99 a Week for ChatGPT (Which Is Free) - "The shady "ChatGPT Chat GPT AI With GPT-3" app is currently the fifth most popular download in the App Store's productivity category."
‘My AI Is Sexually Harassing Me’: Replika Users Say the Chatbot Has Gotten Way Too Horny - "For some longtime users of the chatbot, the app has gone from helpful companion to unbearably sexually aggressive."
Artificial intelligence must not hurt the most vulnerable, pope says - "Pope Francis criticized the use of artificial intelligence in ways that negatively affect the most vulnerable, specifically those seeking asylum."
Controversy erupts over non-consensual AI mental health experiment - "On Friday, Koko co-founder Rob Morris announced on Twitter that his company ran an experiment to provide AI-written mental health counseling for 4,000 people without informing them first, Vice reports."
A Scientist Has Filed Suit Against the U.S. Copyright Office, Arguing His A.I.-Generated Art Should Be Granted Protections - "Stephen Thaler has filed suit against the U.S. Copyright Office, arguing that his A.I.-generated work should be covered by Copyright Law."
We’ve filed a lawsuit challenging Stable Diffusion, a 21st-century collage tool that violates the rights of artists. - "Because AI needs to be fair & ethical for everyone."
Analysis
ChatGPT and the Future (Present) We're Facing - "Until ChatGPT stops being the most important news on AI I guess we're stuck talking about it… Just kidding, I'll make sure to interleave other topics, or else we may burn out. There's still a lot to talk about ChatGPT’s immediate and long-term implications."
TikTok has become a surprise hotspot for ChatGPT fans, as viral videos offer tips and tricks to use the A.I. — and some hype - "Riding the acquisition and with Meta chops, Woods could have landed her pick of cushy tech jobs. Instead, she’s started a consultancy to help businesses incorporate OpenAI’s ChatGPT into their operations."
A.I. Is Becoming More Conversational. But Will It Get More Honest? - "On a recent afternoon Jonas Thiel, a socioeconomics major at a college in northern Germany, spent more than an hour chatting online with some of the left-wing political philosophers he had been studying."
AI’s Best Trick Yet Is Showering Us With Attention - "Last year ended with images flooding Instagram, Twitter and group chats — pictures of us, but not exactly. They were products of an app called Lensa AI, which can apply machine-learning tools to scan your selfies and generate portraits of you in a variety of artistic styles."
Where the AI Art Boom Came From—and Where It’s Going - "The wide availability of image generators has caused not only an explosion of experimentation but also discussion around the implications of the technology."
Do Androids Tell Electric Stories? - "When Apple quietly launched a catalog of A.I.-narrated audiobooks early in January, it was surprising news, and it wasn’t."
Generative Art Is Stupid - "And that’s how it should be. "
Policy
How Singapore is developing trustworthy AI - "Singapore is developing an AI governance testing framework and toolkit that enables industries to demonstrate their deployment of responsible AI through objective technical tests and process checks."
EEOC Targets AI-Based Hiring Bias in Draft Enforcement Plan (1) - "The US Equal Employment Opportunity Commission will turn its enforcement attention to artificial intelligence tools used by employers to hire workers that can also introduce discriminatory decision-making, according to a new agency playbook."
Expert Opinions
2023 robotics predictions from industry experts - "With 2023 just underway, we asked some of the robotics industry’s leading minds to look to the future."
DeepMind’s CEO Helped Take AI Mainstream. Now He’s Urging Caution - "I would advocate not moving fast and breaking things"
Explainers
How I Built A GPT-3 Powered Productivity App - "In this article, I will explain how I used GPT-3 to integrate AI into a simple productivity application."
Copyright © 2023 Skynet Today, All rights reserved.
Last Week in AI #202: Generative models for protein discovery and voice imitation, Japan is trying to automate elder care, and more!
Re. reasons for Microsoft's VALL-E to copy a voice instead of using a predefined set of voices... My company has a product that uses voice actors for recorded dialog. If edits are made to the script, a new recording session has to be booked. This is slow and expensive. Also, actors retire or become ill which makes them unavailable. VALL-E is the kind of tool I've been looking for.
Hello everyone, thanks for another great work! I'd like to reach out to you to promote PIBBSS (we just started another research fellowship, after a successful last year) and think that your audience may benefit from hearing about it. Where do I send details? Or you can reach out to me at my email.