Last Week in AI #147: Failure to regulate 'killer robots', how TikTok reads your mind, the Transactions on Machine Learning Research
U.N. talks adjourn without deal to regulate ‘killer robots’, TikTok document offers a new level of detail about their algorithm, the new Transactions on Machine Learning Research journal, and more!
This week, 125 nations that are part of the Convention on Certain Conventional Weapons (CCW) convened to discuss a treaty to regulate the use of autonomous lethal weapon systems (LAWS), colloquially known as ‘killer robots.’ LAWS have no human-operated "kill switch" that must be used to approve lethal use of the weapon, unlike existing semi-autonomous ‘human in-the-loop’ weapons such as drones. A majority of the 125 nations supported creating such regulations, but were opposed by countries such as Russia, India and the United States. The discussions did not lead to any significant results, with the only outcome being a vague agreement to continue discussions in the future. The Campaign to Stop Killer Robots, a non-profit that supports regulations for LAWS, said the outcome fell “drastically short.”
Our take: This is not a surprising result, but disappointing nonetheless. Reports earlier this year suggested that an A.I drone used in Lybia may have been operating autonomously, and if true that would make it the first use of a ‘killer robot’ in combat. With AI technology already being very powerful and semi-autonomous robotics technology such as drones already being used in war, it is only a matter of time (possibly, not very much time) until ‘killer robots’ start being used on the battlefield, and if no regulations are passed before that the consequences could clearly be bad.
It's no secret that TikTok is addicting. But how does this vehicle for online culture keep you hooked? The New York Times acquired an internal document developed to explain to non-technical hires how TikTok's algorithm works. With a goal of adding daily active users, the TikTok algorithm focuses on two main things: "retention" and "time spent"--in short, keeping users on the app and making them stay there for as long as possible. This sounds simple, but a study of any recommendation algorithm uncovers a lot under the surface.
In developing its video recommendation logic, TikTok's engineers have thought through important aspects of the user experience like how repetitive recommendations may induce boredom. While the usual commentators such as Guillarme Chaslot have pointed out the issues, UC San Diego professor Julian McAuley says the algorithm itself is nothing new, but the setting and vast quantities of data at TikTok's disposal make it powerful. While concerns about TikTok's relation to the Chinese government and its danger to US national security exist, the NYT article concludes that this "menace" seems entirely hypothetical and subject to one's analysis of the US-China relationship and the future of technology and culture.
Our take: Recommendation engines are among the most prominent uses of AI, used in countless websites — just about every social media site, search engine, and entertainment platform such as YouTube. With that, we’ve seen a multitude of problems
AI researchers have created a new journal on machine learning research (TMLR) to accelerate AI research and avoid some of the problems with existing research publication processes. Like other journals, TMLR will accept year-round submissions with no specific deadlines. Unlike other journals, TMLR will use OpenReview to make the review process public and transparent, will do double blind reviewing (the identity of authors and reviewers are unknown to each other), and will aim for a making fast decisions within 2 months. The latter is really important, because the 6 months+ reviewing periods for traditional journals is simply too slow for the rate of AI research advances. TMLR will also evaluate a paper based on the validity and relevance of its claims, not its potential impact, which is very hard to accurately gauge. Lastly, the journal has built-in incentives that encourage more reproducibility studies and reward quality reviewers.
Our take: The current conference and journal publication models all seem ill-suited for the fast-paced developments in AI, and many AI researchers are unsatisfied with the status quo. While the success of TMLR will take years to tell, it is encouraging to see such experimentations.
Spiking Neural Networks - “In August 2014, a significant advance in computing made the cover of the journal Science. It was IBM’s 5.4 billion-transistor chip that had a million hardware neurons and 256 million synapses.”
Sentence Transformer Fine-Tuning (SetFit): Outperforming GPT-3 on few-shot Text-Classification while being 1600 times smaller - “The GPT-n series show very promising results for few-shot NLP classification tasks and keep improving as their model size increases (GPT3–175B). However, those models require massive computational resources and they are sensitive to the choice of prompts for training. In this work, we demonstrate Sentence Transformer Fine-tuning (SetFit), a simple and efficient alternative for few-shot text classification.”
Facebook AI’s FLAVA Foundational Model Tackles Vision, Language, and Vision & Language Tasks All at Once - “A Facebook AI Research team presents FLAVA, a foundational language and vision alignment model that explicitly targets language, vision, and their multimodal combination all at once, achieving impressive performance on 35 tasks across the vision, language, and multimodal domains.”
MIT Researchers Investigate Deep Learning’s Computational Burden - “A team of researchers from MIT, Yonsei University, and University of Brasilia have launched a new website, Computer Progress, which analyzes the computational burden from over 1,000 deep learning research papers.”
Machine learning can now forecast lightning sooner than other methods - “In a new study led by the University of Washington, researchers have demonstrated artificial intelligence’s ability to improve lightning forecasts. Lightning strikes led to the devastating California Lightning Complex fires of 2020, but the strikes are still relatively hard to predict.”
Nonsense can make sense to machine-learning models - “For all that neural networks can accomplish, we still don’t really understand how they operate.”
Human brain cells in a dish learn to play Pong faster than an AI - “Living brain cells in a dish can learn to play Pong when they are placed in what researchers describe as a “virtual game world”.“We think it’s fair to call them cyborg brains,” says Brett Kagan, chief scientific officer of Cortical Labs, who leads the research.”
Improving the factual accuracy of language models through web browsing - “We’ve fine-tuned GPT-3 to more accurately answer open-ended questions using a text-based web browser. “
Customizing GPT-3 for Your Application - “Developers can now fine-tune GPT-3 on their own data, creating a custom version tailored to their application. “
Meet your new A.I. best friend - “We will soon have A.I. companions to help us plan our future, deal with loss, or feel less lonely.”
S.Korea to test AI-powered facial recognition to track COVID-19 cases - “South Korea will soon roll out a pilot project to use artificial intelligence, facial recognition and thousands of CCTV cameras to track the movement of people infected with the coronavirus, despite concerns about the invasion of privacy.”
This AI art app is a glimpse at the future of synthetic media - “If you’ve been hanging out on Twitter lately, then you’ve probably noticed a profusion of AI-generated images sprouting all over your timeline like weird, algorithmic visions.”
MLCommons Association Unveils Open Datasets and Tools to Drive Democratization of Machine Learning - “The MLCommons Association, an open engineering consortium dedicated to improving machine learning for everyone, today announced the general availability of the People’s Speech Dataset and the Multilingual Spoken Words Corpus (MSWC).”
ruDALL-E neural network by Sber for image generation based on Russian-language descriptions now available on ML Space platform - “The commercial version of the world’s first ruDALL-E neural network by SberDevices and Sber AI, which creates images based on text descriptions in Russian, is now available on the ML Space platform in SberCloud’s DataHub of SberCloud’s pre-trained models and datasets.”
How Machine Learning Improves Visibility into Nuclear Power Plant Waste and Contamination Levels - “Machine learning can help gain a better understanding of nuclear power plant waste and even help predict waste contamination to limit any negative effects.”
Artificial intelligence accurately predicts who will develop dementia in two years - “Artificial intelligence can predict which people who attend memory clinics will develop dementia within two years with 92 percent accuracy, a largescale new study has concluded.”
The Sims 4 could use AI to automatically turn photographs into characters - “A patent filed by EA back in April 2020 (and subsequently approved in October 2021) reveals AI technology that could be used in The Sims 4 game to turn pictures of people into in-game characters with just two clicks.”
Robotic Research raises $228M Series A to build out commercial autonomous offerings - “Robotic Research, a self-driving technology company that has spent the last two decades developing on and off-road autonomous vehicles for the Department of Defense, raised a $228 million Series A round.”
California halts Pony.ai’s driverless testing permit after accident - “A California regulator on Monday said it has suspended a driverless testing permit for startup technology firm Pony.ai following an accident - the first time it has issued such a suspension.”
Golden era of robotics adoption’ kicks off in 2022, strategist says - “The robotics industry could reach a “key inflection point” in 2022, said Global X’s Jay Jacobs.”
North America is seeing a hiring boom in medical industry machine learning roles - “North America extended its dominance for machine learning hiring among medical industry companies in the three months ending October. The number of roles in North America made up 63.5 per cent of total machine learning jobs – up from 61.3 per cent in the same quarter last year.”
AI-powered VFX startup Wonder Dynamics raises $10M A round - “Wonder Dynamics aims to make “blockbuster-level” visual effects achievable by living-room-level creators using AI and cloud services, though it has kept its product closely under wraps.”
AI artist Botto made over $1 million selling NFT - “An artificial intelligence algorithm called Botto made about $ 1.3 million in its first six jobs as NFT. “
Motional partners with Uber to provide autonomous food deliveries in California - “Motional said it would start delivering meals from select restaurants on Uber’s food delivery app Eats in Santa Monica in early 2022.”
Levi’s AI Chief Says Algorithms Have Helped Boost Revenue - “The use of AI has helped the apparel company make better decisions in areas such as pricing and shipping, Chief Strategy and AI Officer Katia Walsh says”
The ‘Invisible’, Often Unhappy Workforce That’s Deciding the Future of AI - “Two new reports, including a paper led by Google Research, express concern that the current trend to rely on a cheap and often disempowered pool of random global gig workers to create ground truth for machine learning systems could have major downstream implications for AI.”
We invited an AI to debate its own ethics in the Oxford Union – what it said was startling - “Not a day passes without a fascinating snippet on the ethical challenges created by “black box” artificial intelligence systems. These use machine learning to figure out patterns within data and make decisions – often without a human giving them any moral basis for how to do it.”
How TikTok Reads Your Mind - “It’s the most successful video app in the world. Our columnist has obtained an internal company document that offers a new level of detail about how the algorithm works.”
Algorithms that detect cancer can be fooled by hacked images - “Artificial intelligence programs that check medical images for evidence of cancer can be duped by hacks and cyberattacks, according to a new study.”
France latest to slap Clearview AI with order to delete data - “Controversial facial recognition company, Clearview AI, which has amassed a database of some 10 billion images by scraping selfies off the Internet so it can sell an identity-matching service to law enforcement, has been hit with another order to delete people’s data.”
Job Applicant Resumes Are Effectively Impossible to De-Gender, AI Researchers Find - “Researchers from New York University have found that even very simple Natural Language Processing (NLP) models are quite capable of determining the gender of a job applicant from a ‘gender-stripped’ résumé – even in cases where machine learning methods have been used to remove all gender ind”
Propaganda-as-a-service may be on the horizon if large language models are abused - “AI-powered large language models (LLMs) like OpenAI’s GPT-3 have enormous potential in the enterprise. “
The AI Economy Is Focused on the Usual Superstar Cities. Let’s Fix It Now - “There are things cities, states, companies and the federal government can do to make sure urban areas get a piece of the next big thing”
Machine Learning Most Acquired Skill In 2021 - “Coursera Report finds that Machine Learning was the most acquired skill among Indians in 2021 and that most engineers are not employable.”
Better images of AI - “BBC R&D is collaborating on a project called “Better images of AI” to develop and distribute images representing artificial intelligence (AI) and machine learning (ML) for anyone to use.”
100 years of robots: How technology – and our lives – have changed - “It was 100 years ago that we first heard the term “robot,” so to mark the occasion, 24/7 Tempo has compiled a list of how technology — and our lives — have changed over 100 years of robots.”
What Does It Mean for AI to Understand? - “It’s simple enough for AI to seem to comprehend data, but devising a true test of a machine’s knowledge has proved difficult.”
FTC Signals It May Conduct Privacy, AI, & Civil Rights Rulemaking - “The Federal Trade Commission announced Friday that it is considering using its rulemaking authority “to curb lax security practices, limit privacy abuses, and ensure that algorithmic decision-making does not result in unlawful discrimination.”
Toronto considers banning sidewalk robots - “The Toronto City Council is considering banning mobile robots from sidewalks and bike paths. The provision was originally put forward by the Toronto Accessibility Advisory Committee.”
U.N. chief urges action on ‘killer robots’ as Geneva talks open - “United Nations Secretary-General Antonio Guterres called on Monday for new rules covering the use of autonomous weapons as a key meeting on the issue opened in Geneva.”
U.S. Set to Ban American Investment in Some Chinese Companies Over Surveillance - “A draft Treasury Department announcement says the companies, including a large drone maker, help China’s mass surveillance of Muslim ethnic groups”
5 things lawyers should know about artificial intelligence - “before advising clients on AI issues, lawyers should have some basic technical knowledge to answer questions about legal compliance.”
A Nobel Prize-Winning Economist Explains Why Good AI Will Always Outsmart Humans - “Economist Daniel Kahneman on why the human brain makes avoidable mistakes when solving problems.”
Why Computers Don’t Need to Match Human Intelligence - “With continuing advances in machine learning, it makes less and less sense to compare AI to the human mind.”
California’s AV Testing Rules Apply to Tesla’s “FSD” - “As I emphasized in 2016, California’s rules for “autonomous technology” necessarily apply to inchoate automated driving systems that, in the interest of safety, still use human drivers during on-road testing.”
Do large language models understand us? - “Large language models (LLMs) represent a major advance in artificial intelligence (AI), and in particular toward the goal of human-like artificial general intelligence (AGI).”
Copyright © 2021 Skynet Today, All rights reserved.