Last Week in AI #164: AI uses in government surveillance, AI teaches human drivers, actors union opposes AI actors, and more!

South Africa uses AI and CCTVs to identify and track people, GM explores AI program that gives feedback to human drivers, actors unions campaign against unregulated use of generated content

Top News

South Africa’s private surveillance machine is fueling a digital apartheid

South African surveillance company Vumacam has set up 6600 CCTV cameras and uses AI software to monitor these video feeds for security alerts. The software is trained on roughly 100 hours of footage that classifies “normal” behavior, and it flags anything deemed out of the ordinary. These can then be either ‘Dismissed’ or ‘Escalated’ based on the seriousness of the offense. Other security firms are using facial recognition to identify and track suspects from CCTV. South Africa is also in the process of building out a national biometric identification database called ABIS that would include the face of every resident and foreign visitor, enabling CCTV cameras to track and identify every individual even further. Due to various social, demographic, and economic factors, these cameras are typically placed in such a way that black citizens are disproportionally surveilled than white citizens. The article mentions that such unregulated deployment of AI surveillance is an example of history repeating itself, i.e. for South Africa to go back to its apartheid era.

Our Take: The facts presented certainly feel alarming but should not really come as a shock. It has been evident time and again that use of facial recognition technologies is prone to bias and inequality. Moreover, the risk when such algorithms don’t work increases dramatically when it is targeting a certain section of the population. These cameras are creating a digital equivalent of passbooks, an apartheid-era system that the government used to limit Black people’s physical movements in white areas. The aim of such technological advances should be to take society forward, keeping the interests of society upfront and not hindering their civil liberties. Until the use of such technologies is well regulated, their applications should be limited, especially for law enforcement given the immense potential for misuse. It is important to make sure that the harm does not outweigh the benefits of the technology. This is only possible if there is a collective responsibility and agenda for these private tech companies, government agencies, lawmakers, and researchers.

This article is part one of a series by Karen Hao on the inequalities caused by AI. Read the full series here.

GM Patents Autonomous Tech to Train New Drivers Sans Instructor

Autonomous vehicle makers promise safer roads, with vehicles that can navigate roads while making fewer mistakes than humans do. We have a long way to go until we reach truly “autonomous” vehicles, but what if autonomous technology could be used to help humans drive more safely? GM has filed a patent describing autonomous technology that can evaluate and train drivers using sensors and monitoring how the driver operates controls like the accelerator and wheel. By comparing a trainee’s inputs to recommended driving instructions generated by a driving algorithm, the system scores the driver. If their score meets preset thresholds, the system can selectively grant the driver with greater control of the vehicle.

Our take: GM mentions that human instructors can be prone to bias and that traditional training techniques might be more expensive and less efficient than autonomous techniques. As an idea, autonomous driver training sounds promising. But as with all technologies of this sort, it’s unclear how issues like algorithmic bias or mistakes could affect results. This is a fairly limited use of autonomous technology that can be increased or scaled back depending on how its initial rollout fares. It seems worth watching whether the technology is able to

Actors launch campaign against AI 'show stealers'

A UK actors union is campaigning to warn about the impact of deepfake technology on the livelihoods of actors. Recent improvements in AI-generated videos and speech are making it possible to synthesize a lot of content that traditionally required human actors or voice actors (e.g. commercials, audiobooks). It’s important to note that in most cases, the AI software is not generating content from scratch, but rather from samples of a real human performance. Speech synthesis software may first be trained on the voice of a specific voice actor, then the software can be used to generate different speech in the style and likeliness of that voice actor essentially for free. The union is primarily campaigning for greater transparency, control, and revenue sharing for the actors for the latter case so that an actor’s likeness is not used without consent.

Our take: Generative algorithms will be able to automate away content generation for many domains. This will be disruptive to the creative industries, but it is not clear exactly what the long-term implications are. For example, we may see short-form podcasts or generic audiobooks being machine-narrated, but people might want to pay higher premiums for real human programming. Regardless, it is clear that actors whose voice and performances power the training data for generative algorithms should be fairly compensated, and current copyright regulations have yet to adapt to these use cases.

Other News

Research

Rock On: Scientists Use AI to Improve Sequestering Carbon Underground - "A new neural operator architecture named U-FNO simulates pressure levels during carbon storage in a fraction of a second while doubling accuracy on certain tasks, helping scientists find optimal injection rates and sites."

In-depth insights into Alzheimers disease by using explainable machine learning approach - "Alzheimers disease is still a field of research with lots of open questions. The complexity of the disease prevents the early diagnosis before visible symptoms regarding the individuals cognitive capabilities occur."

UC Berkeley & Intel’s Photorealistic Denoising Method Boosts Video Quality on Moonless Nights - "a research team from UC Berkeley and Intel Labs leverages a GAN-tuned, physics-based noise model to represent camera noise under low light conditions and trains a novel denoiser that, for the first time, achieves photorealistic video denoising in starlight."

Amazon opens MASSIVE AI speech dataset so Alexa can speak your language - "Amazon released an open-source speech dataset supporting 51 languages on Wednesday, encouraging developers to build more third-party apps and services for its AI speaker device Alexa."

Scientists develop new computational approach to reduce noise in X-ray data - "Scientists from the National Synchrotron Light Source II (NSLS-II) and Computational Science Initiative (CSI) at the U.S. Department of Energy's (DOE) Brookhaven National Laboratory have helped to solve a common problem in synchrotron X-ray experiments: reducing the noise, or meaningless information, present in data. "

Applications

What a Rise in AI Means for Agriculture - "As artificial intelligence (AI) and machine learning start to infiltrate agriculture, it's wise to take a moment to remember that the physical world is not data, and plants and animals are not machines."

How AI is helping address the climate crisis - "Emerging technologies are playing a pivotal role in the global effort against climate change, helping the world decarbonize energy production, boost the efficiency of industrial systems, and reduce the carbon footprint of everyday life."

Business

Alphabet-owned Intrinsic is acquiring fellow robotic software firm Vicarious - "Alphabet X-birthed Intrinsic made its big debut last September. The subsidiary looks to buck its parent company’s somewhat spotty robotics record with a software-first approach."

Digital twin generator Unlearn nets $50M to bolster clinical trials with AI models - "Unlearn's goal is to help drugmakers and academic researchers complete clinical trials while requiring fewer overall participants."

Videoverse offers AI tool to help enterprise videos go viral - "Its AI models are trained to detect moments as they occur and render them into new micro-content video assets, auto resized and optimized for different social platforms and one-click publishing."

Startups Join AI Acquisition Rush - "Venture-backed companies spent $8 billion acquiring an estimated 72 artificial intelligence developers last year"

Israeli Construction Tech Startup Ception Unveils AI System For Heavy Equipment - "Israeli AI-based construction tech startup Ception, unveiled its new product MineCept, a real-time high-definition 3D visual mapping system to address heavy equipment challenges in the construction and development sector, the company announced on Thursday."

Motorola Solutions buys London-based video-AI start-up Calipsa - "Motorola Solutions (MSI.N) has acquired London-based start-up Calipsa to boost its ability to use AI-powered analytics across any video security solution, helping its customers verify alarms and detect tampering in real time."

Elon Musk releases more details on Tesla's upcoming robotaxi electric car : 'will cost less per mile than bus ticket' - "The vehicle will focus on cost per mile and Tesla is aiming to reach volume production of the new vehicle in 2024. Musk said that Tesla is going to provide “by far the lowest cost per that customers have ever experienced”."

Concerns

A Tesla vehicle using ‘Smart Summon’ appears to crash into a $3.5 million private jet - "The Reddit user (who did not immediately respond to a request for comment) said the incident took place at an event sponsored by aircraft manufacturer Cirrus at Felts Field in Spokane, Washington."

AI’s inequality problem -"New digital technologies are exacerbating inequality. Here’s how scientists creating AI can make better choices.”

This AI clone of Reddit’s Am I The Asshole forum will give you the best bad advice - "First, a necessary disclaimer: don’t use artificial intelligence language generators to solve your ethical quandaries. Second: definitely go tell those quandaries to this AI-powered simulation of Reddit because the results are fascinating."

Analysis

Is AI Art Real Art? - "Generative Art uses techniques from Machine Learning and Machine Vision to generate images. And it has produced a lot of headlines around the subject of what is and isn’t real art. To me, the answer is clear: AI-generated art is art. Of course, it is."

A.I. Is Mastering Language. Should We Trust What It Says? - "OpenAI’s GPT-3 and other neural nets can now write original prose with mind-boggling fluency — a development that could have profound implications for the future."

Can A.I. All but End Car Crashes? The Potential Is There. - "Well before self-driving cars become a reality, there are simpler approaches that can make roads much safer. This article is the last in a limited series on artificial intelligence’s potential to solve everyday problems."

Policy

The military wants ‘robot ships’ to replace sailors in battle - "The program is a direct response to countries like China, which have been building sophisticated missile technology to target ships that approach their shores. But experts warn the autonomous ships could fuel an AI naval arms race and have difficulty replicating a sailor’s workload."

AI and GDPR: A tight affair - "Introduction This time last year, the European Union took a decisive first step in the direction of regulating lawful, safe and trustworthy artificial intelligence technologies by publishing the so-called AI Act—officially known as the “Proposal for a Regulation laying down harmonized rules on A"

Expert Opinions

On NYT Magazine on AI: Resist the Urge to be Impressed - "What’s needed is not something out of science fiction — it’s regulation, empowerment of ordinary people and empowerment of workers."

---

Copyright © 2022 Skynet Today, All rights reserved.