Discover more from Last Week in AI

Weekly text and audio summaries of the most interesting AI news, as well as editorials commenting on recent events

Over 20,000 subscribers

AI for AI chips, AI's failure at moderating hate speech, and more!

Last Week in AI #120

Last Week in AI

Jun 14, 2021

AI still sucks at moderating hate speech

Incendiary speech is one of the most persistent issues plaguing social media platforms. Given how easily information can spread through social networks and the fact that controversial speech often invites strong reactions, hate speech finds itself at the center of discussions on moderating social media content. Platforms like Facebook have tapped into language AI technology to rise to the challenge, but that technology still struggles: a recent study found that four of the best AI-based hate speech detectors struggled in different ways to distinguish toxic and innocuous sentences. While the study's results cast doubt on the efficacy of hate speech detection systems, the researchers were able to create a taxonomy of different types of hate speech and evaluate how the AI systems fell short at a more granular level. This alone represents useful progress and should help researchers better diagnose and improve hate speech detection systems.

Google is using AI to design its next generation of AI chips more quickly than humans can

As machine learning models become larger and, as a result, compute costs for those models increase, the industry has turned its attention to developing more efficient AI models. This involves creating models that themselves require less compute to train and developing underlying hardware that can more quickly perform the computations that power deep learning models. In a meta-twist, Google has decided to use machine learning to design its own machine learning chips, and its engineers claim that the algorithm's designs are "comparable or superior" to those created by humans and that the algorithm completes its designs far more quickly than humans can. While Google has been working on using machine learning to create chips for years, this effort, described in a paper in Nature, is the first to be applied to a commercial product: Google's TPU chips.

Podcast

Check out our weekly podcast covering these stories!
Website | RSS | iTunes | Spotify | YouTube

News

Advances & Business

‘Care bots’ are on the rise and replacing human caregivers - "If you Google “care bots”, you’ll see an army of robot butlers and nurses, taking vital signs in hospitals, handing red roses to patients, serving juice to the elderly. For the most part these are just sci-fi fantasies. The care bots that already exist come in a different guise."
Microsoft’s ZeRO-Infinity Library Claims to Train 32 trillion Parameters Deep Learning Models On A Cluster of Just 32 GPUs - "In a recent announcement by Microsoft, it has been unveiled that an addition is being made to their open-source DeepSpeed, an AI training library that optimizes memory usage and trains very large deep learning models."
U of T AI research star Raquel Urtasun raises US$83.5-million for self-driving startup four months after leaving Uber - "One of the world’s leading autonomous vehicle researchers, University of Toronto computer science professor Raquel Urtasun, has raised US$83.5-million for her three-month-old self-driving technology startup, one of the largest inaugural venture financings of a woman-led Canadian startup."
What Really Happened When Google Ousted Timnit Gebru - "She was a star engineer who warned that messy AI can spread racism. Google brought her in. Then it forced her out. Can Big Tech take criticism from within?"
What Matters in Adversarial Imitation Learning? Google Brain Study Reveals Valuable Insights - "AI’s mastery of complex games like Go and StarCraft has boosted research interest in reinforcement learning (RL), where agents provided only with the game rules engage in self-play to elevate their performance to human level and beyond."
Mythic launches analog AI processor that consumes 10 times less power - "Elevate your enterprise data technology and strategy at Transform 2021. Analog AI processor company Mythic launched its M1076 Analog Matrix Processor today to provide low-power AI processing."
AI Model May Lead to Moodier, More Cinematic Drones - "Researchers from CMU are working to enable drone flights that automatically conduct cinematography in response to a desired mood. "
MIT Researchers Leverage Machine Learning for Better Lidar - "The researchers are hopeful that their work will help pave the way to future self-driving systems that require less manual intervention – by both programmers and drivers – to deliver satisfactory results."
What would you pay for autonomous driving? Volkswagen hopes $8.50 per hour - "The future of driving may cost you $8.50 per hour if Volkswagen follows through on its boardroom musings. The German automaker is considering charging an hourly fee for access to autonomous driving features once those features are ready."
Google Hopes AI Can Turn Search Into a Conversation - "Google often uses its annual developer conference, I/O, to showcase artificial intelligence with a wow factor. In 2016, it introduced the Google Home smart speaker with Google Assistant. In 2018, Duplex debuted to answer calls and schedule appointments for businesses."
Facebook’s AI can copy the style of text in photos from a single word - "Facebook today introduced TextStyleBrush, an AI research project that can copy the style of text in a photo from just a single word."
EleutherAI claims new NLP model approaches GPT-3-level performance - "A grassroots collection of AI researchers, EleutherAI this week released GPT-J-6B (GPT-J), a model the group claims performs nearly on par with an equivalent-sized GPT-3 model on various tasks."

Concerns & Hype

The current state of affairs and a roadmap for effective carbon-accounting tooling in AI - "EcoQoS is a new Quality of Service (QoS) level introduced to Windows that developers can now opt-in to run their work efficiently, leading to better energy efficiency/increased battery life, reduced fan noise and power/thermal throttling."
Is there any way out of Clearview’s facial recognition database? - "In March 2020, two months after The New York Times exposed that Clearview AI had scraped billions of images from the internet to create a facial recognition database, Thomas Smith received a dossier encompassing most of his digital life."
When AI Becomes Childsplay - "Despite their popularity with kids, tablets and other connected devices are built on top of systems that weren’t designed for them to easily understand or navigate. But adapting algorithms to interact with a child isn’t without its complications—as no one child is exactly like another."
It’s 2021. Do You Know What Your AI Is Doing? - "Responsible AI has been one of my big topics for a few years now, the subject of many articles, blogs and talks I’ve given to audiences around the world."
India’s ‘Ugliest’ Language? Google Had an Answer (and Drew a Backlash). - "A Google fact box singled out Kannada, a language spoken in the country’s south. The faux pas highlights the algorithm’s fallibility. It was an odd, unanswerable question. Still, it was on the mind of at least one Google user in India."
The Double Exploitation of Deepfake Porn - "Over the past three years, celebrities have been appearing across social media in improbable scenarios. You may have recently caught a grinning Tom Cruise doing magic tricks with a coin or Nicolas Cage appearing as Lois Lane in Man of Steel."
Is there any way out of Clearview’s facial recognition database? - "In March 2020, two months after The New York Times exposed that Clearview AI had scraped billions of images from the internet to create a facial recognition database, Thomas Smith received a dossier encompassing most of his digital life."
If a killer robot were used, would we know? - "A recent UN report on Libya implies—but does not explicitly state—that a Turkish Kargu-2 drone was used to attack humans autonomously using the drone’s artificial intelligence capabilities. I wrote about the event in the Bulletin, and the story went viral."

Analysis & Policy

The Future of AI in Law: Changing the Legal Landscape - "Artificial intelligence (AI) is one of the fastest-growing technological industries today, but what effects will it have on legal practices?"
The Biden Administration Launches the National Artificial Intelligence Research Resource Task Force - "Task Force members will help develop a roadmap to democratize access to research, tools that will promote AI innovation and fuel economic prosperity Today, the White House Office of Science and Technology Policy (OSTP) and the National Science Foundation (NSF) announced the newly formed National Art"
Across China, AI ‘city brains’ are changing how the government runs - "At the Linshu county headquarters in eastern China, a huge screen covering an entire wall provides round-the-clock information on its 600,000 residents."

Expert Opinions & Discussion within the field

DeepMind scientists: Reinforcement learning is enough for general AI - "This article is part of our reviews of AI research papers, a series of posts that explore the latest findings in artificial intelligence."
Graphs at ICLR 2021 - "The field of geometric deep learning is booming, there’s no way around it. And graph neural networks are the rockstars, sitting in the driver seat. Graphs were clearly at the center of a lot of attention at ICLR 2021."

That's all for this week! If you are not subscribed and liked this, feel free to subscribe below!