This article is more than 1 year old
Amazon continues its ban on allowing police to use its facial-recognition software
Plus: DeepMind has been trying to gain more independence from Google, and how AI can help cosmologists
In brief Amazon promised it would refuse to allow the police to use its controversial Rekognition service for one year, and has decided to continue its ban indefinitely.
The moratorium, however, only covers police departments, and Amazon still sells access to its Rekognition technology to organizations that may well end up supplying services to cop shops, and to government agencies, too.
AI outfits jockey for schools' COVID-19 funds
Vendors are hawking AI-powered cameras to schools, claiming the expensive technology will be useful for limiting the spread of COVID-19 as students return.
Facial-recognition algorithms can be used to monitor pupils, and can detect if they’re wearing their masks or not. Software can measure if they’re keeping six-feet away from each other. These are just some of the capabilities being marketed by companies hoping to bag contracts from schools given COVID-19 relief funds in the US, according to Vice.
Motorola Solutions, Verkada, and SchoolPass are among those that have designed brochures to sell their surveillance equipment to schools, claiming they’ll make classes safer. One advert [PDF] from Motorola Solutions splashed the figure: “$82 BILLION.” And said: “Consider COVID-19 technology from Motorola Solutions for your Education Stabilization Fund dollars.”
Google denies DeepMind’s requests to be more independent
DeepMind has been fighting to have greater control and ownership over its AI technology for years though the powers that be aren't letting up.
The startup was acquired in 2014 by its Alphabet stablemate Google, which has spent hundreds of millions of dollars on funding DeepMind’s research. Although the lab has developed some of the most powerful and innovative publicly known AI software, including its reinforcement learning and search model, AlphaGo, and its protein structure predictor, AlphaFold, it hasn’t managed to reap any profits yet.
In fact, it has only continued to lose more and more money. In 2018, for instance, it racked up a $570m loss, according to Forbes.
Amid this, DeepMind wants to operate under Alphabet as a nonprofit to prevent Google from being able to control its AI research, and was engaged in serious talks to do so for years, according to the Wall Street Journal.
Last month, however, DeepMind told its staff those talks have stopped after Google put its foot down on its request to be more independent. It’s likely Google will continue to try and find ways to monetize the lab’s research. In 2018, the ad giant took over DeepMind’s health department, swallowing its data and software with it.
Google launches a new AI cloud platform
Google CEO Sundar Pichai announced the creation of Vertex AI, an online service that promises to help developers build and deploy machine-learning models in the cloud more easily, during its annual IO conference last week.
Vertex AI acts as a single hub, containing all the tools and software needed for data scientists to host their models and funnel data in and out of their systems so the software can be trained or deployed in production.
“Vertex AI requires nearly 80 per cent fewer lines of code to train a model versus competitive platforms, enabling data scientists and ML engineers across all levels of expertise the ability to implement Machine Learning Operations to efficiently build and manage ML projects throughout the entire development lifecycle,” Vertex AI director Craig Wiley claimed in a blog post.
AI to help cosmologists simulate the universe
A neural network built by academics led by Carnegie Mellon University in the US, and described in a paper this month, can seemingly be used to enhance simulations of the universe.
Cosmologists rely on computational models that require high amounts of compute to run, and can appear crude and low-resolution. This aforementioned AI-assisted approach, however, allows researchers to probe the universe in more detail. Here's the abstract:
Cosmological simulations are indispensable for understanding our Universe, from the creation of the cosmic web to the formation of galaxies and their central black holes. This vast dynamic range incurs large computational costs, demanding sacrifice of either resolution or size and often both.
We build a deep neural network to enhance low-resolution dark-matter simulations, generating superresolution realizations that agree remarkably well with authentic high-resolution counterparts on their statistical properties and are orders-of-magnitude faster. It readily applies to larger volumes and generalizes to rare objects not present in the training data. Our study shows that deep learning and cosmological simulations can be a powerful combination to model the structure formation of our Universe over its full dynamic range.
"Cosmological simulations need to cover a large volume for cosmological studies, while also requiring high resolution to resolve the small-scale galaxy formation physics, which would incur daunting computational challenges,” said Yueying Ni, one of the paper’s authors and a professor of physics and astronomy at the University of California Riverside.
We're told the neural network was able to speed up a simulation made up of 134 billion virtual particles that would take a typical supercomputer months to run to just 16 hours on a single GPU. ®