

Discover more from Last Week in AI
Last Week in AI #143: Google has 100+ robots in its offices, an AI lie detector, multilingual beats bilingual translations
Everyday Robots deploys 100+ robots around Google offices to develop assistive robots, Utah company to sell AI lie detector, Meta open-sourced its multilingual AI translation system, and more!
Top News
Alphabet is putting its prototype robots to work cleaning up around Google’s offices
Alphabet has announced that its Everyday Robots Project team has been deploying its robots to carry out custodial tasks on Google’s Bay Area campuses. The team was started at Google X, a division within the company dedicated to ambitious moonshot projects, in 2016. Last week, the company’s chief robots officer put out a blog post describing their effort to put their robots to work outside of its labs:
We are now operating a fleet of more than 100 robot prototypes that are autonomously performing a range of useful tasks around our offices. The same robot that sorts trash can now be equipped with a squeegee to wipe tables and use the same gripper that grasps cups can learn to open doors.
The blog post also explains that these robots are trained to do such tasks via machine learning, rather than by being manually programmed. While the approaches they describe are not novel and are being worked on by many AI labs, deploying robots outside these labs among non-researchers has rarely been attempted. Everyday Robots is now doing just that, and it plans to keep expanding these efforts going forward.
Our take: Demonstrating successful robot learning in labs is far easier than in the real world. It’s exciting to see these reports of Everyday Robots seemingly accomplishing that, but it’s hard to say if the solution has been tailor-made for Google or if it is general enough for the robots to work in other settings. Regardless, these efforts are still prototypes, and I would not be surprised if it takes a decade or more for this technology to be commercialized.
More on this: Inside X’s Mission to Make Robots Boring
A Utah company says it’s revolutionized truth-telling technology. Experts are highly skeptical.
EyeDetect, by Utah company Converus, claims to be able to tell whether someone is lying from their eye movements. While the creators claim the system has an accuracy rate of between 83 and 87 percent, experts are skeptical. Critics claim that human truth-telling is too subtle for any dataset--relying on such a system could result in unjust outcomes.
Our take: I feel like some AI systems today are reminding me of the sort-of joke that you can find a correlation between literally any two things. There are genuine causal claims behind many of the things prediction systems do, but indulging in the idea that we can predict anything from anything fabricates causal claims out of nothing. You can train a machine learning system to accurately predict some dependent variable from an independent variable that is in fact quite unrelated because you have the data to do so. The developers even make the claim that "by entirely automating the test, it avoids the possibility of human bias" (and we all know how accurate of a statement that is). As the article points out, this system in particular feels like the sort with dystopian implications even if it does work well. I think there are a lot of potentially beneficial AI systems that might be troublesome if not implemented correctly. This seems to be one that isn't good news, whether it works or not.
Meta AI Open-Sourced It’s First-Ever Multilingual Model (Won The WMT Competition): A Step Towards Future Of Machine Translation
AI-powered translations have mostly required “bilingual” datasets - pairs of sentences from 2 languages for translating from one to the other. Such bilingual models are ineffective for translating to or from languages that do not have giant datasets. Meta’s recently open-sourced multilingual models sidestep these issues by clever data mining and training procedures, and training on large models using 52 billion parameters allowed the multilingual models to outperform bilingual ones for the first time.
Other News
Research
Researchers Develop A Unified Framework For Evaluating Natural Language Generation (NLG) - "Researchers from Carnegie Mellon University, Petuum Inc., MBZUAI and UC San Diego recently took a step in this direction by developing a single framework for NLG evaluation that makes it easier to create metrics for various language generation tasks and characteristics."
Research Professor Adji Bousso Dieng Is Out to Change the World With AI - "This fall, Assistant Professor Adji Bousso Dieng became the first Black female faculty member in the engineering school’s 100-year history, and the first Black faculty member in the computer science department."
Big data privacy for machine learning just got 100 times cheaper - "Rice University computer scientist Ashumali Shrivastava (left) and graduate student Ben Coleman discovered an inexpensive way to implement rigorous personal data privacy when using or sharing large databases for machine learning."
Applications
Solving entertainment’s globalization problem with AI and ML - "The recent controversy surrounding the mistranslations found in the Netflix hit “Squid Game” and other films highlights technology’s challenges when releasing content that bridges languages and cultures internationally."
AI can quickly identify structure of drugs designed for legal highs - "An artificial intelligence can identify designer drugs that have similar effects to substances such as cocaine and heroin, but which can’t be detected by current tests"
Machine Learning Shushes Stressed Dogs - "If there’s one demographic that has benefited from people being stuck at home during Covid lockdowns, it would be dogs. Having their humans around 24/7 meant more belly rubs, more table scraps, and more attention."
Google launches Bot-in-a-Box to nudge along conversational AI - "The new feature from Google Cloud makes it easier for businesses to launch chatbots using existing customer FAQs."
AI predicts which mammals are most likely to spread covid-19 - "An AI tool has predicted 540 mammalian species that are most likely to spread covid-19 using information about where they live and aspects of their biology."
AWS makes AI and machine learning tangible with first major art debut at Smithsonian - "Amazon Web Services Inc. has commissioned its first-ever major art piece, a site-specific sculpture powered by artificial intelligence and designed by artist and architect Suchi Reddy that will be the centerpiece of the Smithsonian’s “Futures” exhibit."
Cohere partners with Google Cloud to train large language models using dedicated hardware - "Google Cloud, Google’s cloud computing services platform, today announced a multi-year collaboration with startup Cohere to “accelerate natural language processing (NLP) to businesses by making it more cost effective."
Delivery Robots are Ready to Take Your Order - "Autonomous, self-driving rovers have arrived at the University of Arizona, ready to deliver food and beverages across campus."
Business
AI-powered writing assistant Writer nabs $21M - "Writer, which bills itself as an AI writing assistant for marketing teams, today announced that it raised $21 million."
OpenAI makes GPT-3 generally available through its API - "Starting today, any developer in a supported country can sign up to begin integrating the model with their app or service."
Hungary’s OTP Bank Invests in AI Supercomputer - "The system, developed by Silicon Valley’s SambaNova, will leverage AI language models to automate more banking services for Hungarian speakers"
Baidu beats quarterly revenue estimates on robust ad sales, AI - "China's Baidu Inc BIDU.O on Wednesday reported quarterly revenue above Wall Street estimates, helped by stronger advertising sales and demand for its artificial intelligence and cloud products."
Chipmaker Nvidia forecasts robust revenue on AI, metaverse buzz - "The world’s biggest maker of graphics and AI chips expects fourth-quarter revenue to grow above analysts’ expectations."
Baidu's Apollo providing far more self-driving mobility services than Waymo - "Baidu Apollo Go provided 115,000 rides in the third quarter. This compares to Waymo's quarterly order volume of about 20,000 to 40,300."
Driverless car start-up WeRide partners with major Chinese automaker GAC on robotaxis as competition rises - "Chinese autonomous driving start-up WeRide has partnered with major automaker Guangzhou Automobile Group (GAC Group) to launch driverless taxis on the latter's ride-hailing service Ontime."
Grammarly raises $200M at a $13B valuation to make you an even better writer using AI - "Grammarly, the popular auto-editing tool for writing, has raised $200 million in funding at a $13 billion valuation from new investors including Baillie Gifford and funds and accounts managed by BlackRock, among others."
Apple reportedly wants to launch a self-driving EV in 2025 with a custom chip - "Apple has completed “much of the core work” on a new processor meant to power its secretive autonomous electric car project known as Titan, Bloomberg’s Mark Gurman reports."
New Starbucks Pickup store in NYC uses Amazon Go cashierless technology - "Starbucks has opened a new “pickup cafe” in New York City that uses Amazon Go’s “just walk out” cashierless tech, the companies said Thursday."
Concerns
South Korea is Selling Millions of Photos To Facial Recognition Researchers - "Travelers to the country are having their biometric data sold to private companies under a program recently uncovered by civil rights groups."
Americans are confused about AI - "A new poll about Americans' attitudes toward AI shows deep ambivalence about what the technology will do to their lives."
Contract lawyers face a growing invasion of surveillance programs that monitor their work - "Attorneys say the constant workday face scans, mandated by their bosses, are fueling fears of over-surveillance: “I will not subject myself to this indignity and the invasion of my privacy in my own home.""
We tried Tesla's 'full self-driving.' Here's what happened - "When a dozen small children crossed in front of our Tesla with "full self-driving," I had good reason to be nervous."
AI Will Create 97 Million Jobs, But Workers Don’t Have the Skills Required (Yet) - "Artificial intelligence technologies have reduced repetitive work and enhanced work efficiency, and as a result, almost every industry in the world is planning to leverage AI or has already implemented it in their business."
Policy
Businesses could face fines in city crackdown on bias for AI hiring tools - "Employers could soon face stricter rules for when they can use resume-scanning software and other automated tools"
The Department of Defense is issuing AI ethics guidelines for tech contractors - "The controversy over Project Maven shows the department has a serious trust problem. This is an attempt to fix that."
Expert Opinions
Radical Reads: From AI researcher to founder – an exclusive conversation with Pieter Abbeel - "The focus of the conversation was on the first principles of starting a business with unique consideration given to what it means to make the leap from AI researcher to AI founder."
Explainers
Easy for you, tough for a robot - "Engineers are trying to reduce robots’ clumsiness — and boost their common sense"
Podcast
Check out our audio discussions of last week’s AI news stories!
Subscribe to the podcast here: Website | RSS | iTunes | Spotify | YouTube
Copyright © 2021 Skynet Today, All rights reserved.