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 23,000 subscribers

Last Week in AI #125: AI art, AI to predict protein folding, sanctions against facial recognition

AI Generated Art Scene Explodes as Hackers Create Groundbreaking New Tools , US sanctions a Chinese facial recognition company, DeepMind's AI for protein structure is coming, and more!

Last Week in AI

Jul 19, 2021

Welcome to the latest Last Week in AI! This week, we'll highlight and summarize articles about the new VQGAN+CLIP art scene, US sanctions of a Chinese facial recognition company, and the open-sourcing of DeepMind’s AlphaFold. Plus, we’ll have our usual round up of a bunch more news about AI advances, business, concerns, and more! If you were referred by a friend, go to lastweekin.ai to subscribe and/or follow us on twitter @LastWeekinAI. If you enjoyed this issue, feel free to share:

AI Generated Art Scene Explodes as Hackers Create Groundbreaking New Tools

A new method for creating art with AI has become popular: just tell an AI model what you want it to draw, and watch it happen. The basic technical details are simple: a neural network that can generate images (such as a Generative Adversarial Network, or GAN) is hooked up to a second neural network that can evaluate how well a given image matches with some arbitrary text, and the text you provide is used find an image the first network can generate that is also a good match for your test "prompt". While AI models that are able to generate images from text have existed for years, the existence of much larger neural networks for both image generation and image-text matching has led to this so called VQ-GAN+CLIP approach (named after the two AI models that are typically used) to be able to generate impressively varied and consistently surprising images. AI art enthusiasts figured this out over the course of months following OpenAI releasing CLIP earlier this year.

US sanctions a Chinese facial recognition company with Silicon Valley funding

Hitachi Kokusai Electric Surveillance System — Image Credits: The Verge

Among the information we have learned about the persecution of the Uighur population in the Chinese province of Xinjiang is the extensive use of surveillance in the region. In particular, the Chinese government has teamed up with companies that provide facial recognition software in the development of a real-life panopticon. In keeping with US criticism of human rights abuses in Xinjiang, the US Department of Commerce moved to sanction 14 Chinese companies linked to the abuses, including DeepGlint, a facial recognition company with ties to Chinese police surveillance and funding from US-based Sequoia Capital (the investment had occurred before the Xinjiang news became public). The Commerce Department's Entity List, whose members US companies cannot engage in business without special permission, also includes Sequoia-backed Yitu Technology. The sanctions are not focused on China alone: they apply to companies in Iran, Russia, and Canada, among others.

DeepMind's AI for protein structure is coming to the masses

3D view of human interleukin-12 bound to its receptor — Image Credits: Ian Haydon, UW Medicine Institute for Protein Design via Nature

Last year DeepMind demonstrated AlphaFold 2, a deep learning powered algorithm that can predict protein structures from amino acid sequences much more accurately than what was previously capable. This software, along with a new comparable algorithm called RoseTTaFold from an academic group, have both been recently released to the public. The team behind RoseTTaFold has also hosted a free online service to predict protein structures from submitted protein sequences. Accurate protein structure prediction can enable faster research and development in biotechnologies and potentially drug discovery. The release of such tools can accelerate research in this area, and in the future similar techniques may be used to solve more complicated problems, like "predicting the structure of complexes of multiple interacting proteins and applying the software to the design of new proteins." For more on this topic, see our brief from last year covering AlphaFold 2.

If you are not subscribed yet and like our weekly AI newsletters, feel free to subscribe below!

News

Advances

Does Your Video Need a Better Soundtrack? Ask an AI to Write One

"In a recent demo conducted over Zoom, I watched as the music composition app Dynascore transformed the entire emotional tenor of a short video multiple times in under a minute, all without altering a single frame of the visuals."

MIT robot could help people with limited mobility dress themselves

"Robots have plenty of potential to help people with limited mobility, including models that could help the infirm put on clothes. That's a particularly challenging task, however, that requires dexterity, safety and speed."

MIT Proposes Novel End-to-End Procedure for Corrupted Data Cleaning, Estimation, and Inference

"Economic data is imperfect, and even the most carefully curated datasets can have noisy, missing, discretized, or privatized variables."

Improving Playtesting Coverage via Curiosity Driven Reinforcement Learning Agents

"This research paper has been accepted for publication at IEEE Conference on Games 2021. As modern games continue growing both in size and complexity, it has become more challenging to ensure that all the relevant content is tested and that any potential issue is properly identified and fixed."

Stanford's AI Researchers Introduce QA-GNN Model That Jointly Reasons With Language Models And Knowledge Graphs

"Question-answering systems are the backbone of our digital lives."

Reducing the Computational Cost of Deep Reinforcement Learning Research

"It is widely accepted that the enormous growth of deep reinforcement learning research, which combines traditional reinforcement learning with deep neural networks, began with the publication of the seminal DQN algorithm."

EleutherAI Researchers Open-Source GPT-J, A Six-Billion Parameter Natural Language Processing (NLP) AI Model Based On GPT-3

"GPT-J, a six-billion-parameter natural language processing (NLP) AI model based on GPT-3, has been open-sourced by a team of EleutherAI researchers. The model was trained on an open-source text dataset of 800GB and was comparable with a GPT-3 model of similar size."

Need to Fit Billions of Transistors on a Chip? Let AI Do It

"Artificial intelligence is now helping to design computer chips - including the very ones needed to run the most powerful AI code. Sketching out a computer chip is both complex and intricate, requiring designers to arrange billions of components on a surface smaller than a fingernail."

AI Teaches a Robot How to Learn to Walk

"The system allows legged robots to adapt to their environment by exploring and interacting with the world."

Facebook's BlenderBot 2.0 bot surfs the web for knowledge

"Facebook today took the wraps off of BlenderBot 2.0, which it says is the first chatbot that can build long-term memory while searching the internet for up-to-date information."

DeepMind Introduces It's Supermodel AI 'Perceiver': A Neural Network Model That Could Process All Types Of Input

"DeepMind recently released a state-of-the-art deep learning model called Perceiver via a recent paper."

Business

Google CEO Still Insists AI Revolution Bigger Than Invention of Fire

"The artificial intelligence revolution is poised to be more 'profound' than the invention of electricity, the internet, and even fire, according to Google CEO Sundar Pichai, who made the comments to BBC media editor Amol Rajan in a podcast interview that first went live on Sunday."

Discord buys AI anti-harassment company

"Discord has purchased a company working on AI tools to help detect harassment online. The company, Sentropy, monitors online networks for abuse and harassment, then offers users a way to block problematic users and filter out messages they don't want to see."

Retail stores are packed with unchecked facial recognition, civil rights organizations say

"More than 35 organizations are demanding top US retailers cease using facial recognition to identify shoppers and employees in their stores, which companies have used to deter theft and identify shoplifters."

OpenAI disbands its robotics research team

"OpenAI has disbanded its robotics team after years of research into machines that can learn to perform tasks like solving a Rubik's Cube"

Announcing the winners of the Women in AI Awards at Transform 2021

"VentureBeat leadership made the final selections out of the over 100 women who were nominated during the open nomination period."

Concerns & Hype

GitHub's Commercial AI Tool Was Built From Open Source Code

"Earlier this month, Armin Ronacher, a prominent open-source developer, was experimenting with a new code-generating tool from GitHub called Copilot when it began to produce a curiously familiar stretch of code."

The future of self-driving? Maybe less like Elon Musk and more like Domino's pizza robots

"As companies like Tesla and its CEO Elon Musk come to Austin, Texas, the booming city and new tech hub has grown so much it has struggled to make good on its "keep it weird" motto."

Man Wrongfully Arrested By Facial Recognition Tells Congress His Story

"Williams' testimony was part of a hearing held by the House of Representatives' subcommittee on crime, terrorism, and homeland security, which dealth with how law enforcement uses the highly controversial surveillance technology."

Human Evaluations No Longer the Gold Standard for NLG, Says Washington U & Allen AI Study

"For years, natural language generation (NLG) researchers have been recruiting humans to evaluate their models' text outputs."

OpenAI Codex shows the limits of large language models

"This article is part of our reviews of AI research papers, a series of posts that explore the latest findings in artificial intelligence. In a new paper, researchers at OpenAI have revealed details about Codex, a deep learning model that generates software source code."

Feeding the machine: We give an AI some headlines and see what it does

"There's a moment in any foray into new technological territory when you realize you may have embarked on a Sisyphean task."

Fight for the Future launches campaign to keep facial recognition off U.S. college campuses

"Fight for the Future, the Boston-based nonprofit promoting causes related to copyright legislation, online privacy, and internet censorship, today announced that it will team up with advocacy group Students for Sensible Drug Policy to launch an effort to ban facial recognition from university campus"

Federal watchdog says A.I. vendors need more scrutiny

"Companies that sell artificial intelligence technologies to the federal government should prepare for tougher scrutiny of their products, said the Government Accountability Office."

Holly Herndon Releases AI Deepfake Tool That Lets Others Make Music With Her Voice

"Holly Herndon has released a new artificial intelligence tool - which the composer is also referring to as her 'digital twin' - called 'Holly+' that allows users to upload any polyphonic audio and receive a new version of that music sung in Herndon's own voice."

Maker of machine learning cheat says "my intent was never to do anything illegal" after Activision clampdown

"The maker of a machine learning cheat that hit the headlines last week has insisted "my intent was never to do anything illegal" after Activision stepped in to shut development down."

Black teen kicked out of skating rink after facial recognition camera misidentified her

"FOX 2 - A local roller skating rink is coming under fire for its use of facial recognition software after a teenager was banned for allegedly getting into a brawl there."

Voice clone of Anthony Bourdain prompts synthetic media ethics questions

"A New Yorker review of 'Roadrunner,' a documentary about the deceased celebrity chef Anthony Bourdain by the Oscar-winning filmmaker Morgan Neville, reveals that a peculiar method was used to create a voice over of an email written by Bourdain."

What Ever Happened to IBM's Watson?

"A decade ago, IBM's public confidence was unmistakable. Its Watson supercomputer had just trounced Ken Jennings, the best human 'Jeopardy!' player ever, showcasing the power of artificial intelligence."

Expert Opinions & Discussion within the field

Training AI: Reward is not enough

"Contrary to Silver et al.'s claims, reward is not enough."

Explainers

What are Diffusion Models?

"Diffusion models are a new type of generative models that are flexible enough to learn any arbitrarily complex data distribution while tractable to analytically evaluate the distribution."