Trump's DoD strikes backs against Bezos's best laid plans ⚔️
GANs skips over hard parts of images like they don't exist
|jackie snow||Nov 1, 2019|
I love my new little logo, but think I still need to add a little “mach·in·o·cene (n): the upcoming era when machines have had such an impact on the planet that it constitutes a distinct geological age.” Next week!
Story of the week
The Defense Department gave Microsoft a huge tech contract this week worth $10 billion. The deal, called Joint Enterprise Defense Infrastructure (aka JEDI), is heavily focused on cloud computing and was expected to go to Amazon, which has the strongest cloud offerings out of the big tech companies. Too bad Jeff Bezos is also the owner of the newspaper that Trump calls the “Amazon Washington Post.”
There is a chance that Amazon will fight back and try to at least get part of the contract, an idea thrown around earlier this year over fears that the Pentagon shouldn’t be so reliant on one company. Either way, the Defense Department will likely get a lot of bespoke tools as it upgrades its tech, but we will have to see how they put to use all the built-in AI abilities that whoever they end up using offers.
This comes just as a tech panel advising the Pentagon approved a list of AI ethics principles that are meant to guide how the military should apply ethics to AI systems. The report includes 12 recommendations for ethical AI in both combat and non-combat settings. We will see how that works out the moment a hard decision has to be made.
Robo pup Aibo could be trained one day soon to turn on some smart appliances.
The ACLU is suing the FBI to find out more about how it’s using face rec tech.
If getting your license wasn’t stressful enough, now there is an AI system judging you!
DeepMind researchers taught an AI how to destroy human players at StarCraft 2.
As cities ban governmental uses of face recognition, businesses don’t want to get caught up in the next dragnet.
The NYPD swears its AI isn’t racist at all because “In terms of race, it’s essentially operating blindly.” Famous last words.
Uber got a self-driving safety and responsibility board, unfortunately after they already killed someone.
Australia might turn to face rec tech to keep kiddos off porno sites.
350 million Google searches every day should be better due to a tweak to their algorithm, the biggest change to it in a half-decade.
See what your pet would look like as a different breed or even a lion with this playful new AI tool.
U.S. Special Operations Forces were able to identify Abu Bakr al-Baghdadi with help from face rec tech (as well as fast DNA analysis).
Last week, the Senate passed the Deepfake Report Act, which would require the Department of Homeland Security to publish an annual report on the use of the tech and how it’s being used.
Santa’s little helpers this holiday season are robots.
A neural network discovered we have liking underestimating the population exposed to sea-level rise by more than a hundred million people.
Tim Cook will personally be reviewing your Siri commands and dictation with the latest iPhone software update. JK on Tim Cook, but workers in the global South will be!
GANs, or generative adversarial networks, are one of the most exciting (to researchers, at least) innovations from this current AI spring. The tech bounces between two neural networks that figure out how to create new media after seeing a bunch of examples. It’s great for synthetic data creation (useful for when you don’t have enough of the real stuff or have privacy concerns) and has advanced so far its creations are indistinguishable from actual images (check out Thispersondoesnotexist.com for proof).
But there are some things GANs can’t do, as this recent paper lays out. Given a number of photos showing different scenes with items like buildings, a sky, and trees, and a GAN *should* create new images with about the same distribution of those objects in the new images. The researchers, however, found certain things that GANs don’t produce at the same rate that they see in the training set. Those parts that GANs skip over? People, cars, palm trees, and signboards. The paper doesn’t answer why this happens but does define a research area for further exploration that could make GANs even more sophisticated in the future.
Datapoint of the week
3,800: How many AI-related applications the U.S. Patent Office received in 2018, up from roughly 3,200 in 2017 and less than 1,900 in 2014. That’s why they are looking to hire an AI specialist, according to WSJ.
“Florida has quickly become the Wild Wild West for robot cars. There is nothing stopping a Chinese tech company from beta-testing its autonomous 18-wheelers at 7:30 a.m. in an elementary school zone.”
-Dale Swope, the former president of the trial lawyer advocacy group the Florida Justice Association, in One Zero
Until The Rise of Skywalker comes out to tie up confusing loose threads,