Amazon's neural net offer to border cops, Waymo charges fares, the first AI portrait sold at auction, and more

Roundup Let's kick Monday off with a bunch of bits and bytes you may have missed last week from the world of AI – alleged intelligence or artificial intelligence, depending on where you stand.

Amazon and ICE: A leaked email has shown that Amazon tried to sell its facial recognition software to the US Immigration and Customs Enforcement to help the government identify people from video footage.

Documents published by the Project on Government Oversight (POGO), a non-profit organization investigating the US federal government, show an email sent from an Amazon employee to a representative at ICE, in June earlier this year.

“Thank you for your time with Amazon Web Services on Tuesday at the McKinsey Redwood City, CA Offices. We are ready and willing to support the vital HSI mission,” it reads. There is also a list of “action items” that Amazon and ICE discussed. One of the bullet points is “Reckognition Video tagging/analysis, scalability, custom object libraries” as well as a list of AI algorithms AWS uses.

Amazon has already been criticised for selling its facial technology, Reckognition, to police departments and Palantir, a data analytics business.

A previous investigation by the American Civil Liberties Union (ACLU) found that Rekognition was struggled to identify people of colour correctly. The ACLU trained the system on a database of criminal mugshots and matched them against a list of innocent US congress members.

Out of the 535 images, 28 were incorrectly identified. Nearly 40 per cent of the wrong matches were of people of color, when they only make up about 20 per cent of Congress.

It’s not entirely clear how the technology is being used by ICE, the ACLU have filed an Freedom of Information act request to Department of Homeland Security. The non-profit group wants to know “where and how it intends to use facial recognition as well as who it has purchased the technology from.”

IBM Watson Health chief is leaving: The saga following the fallout of IBM Watson Health continues as the head of IBM’s health division steps down.

Deborah DiSanzo is leaving her post, according to Stat News. IBM Watson Health has struggled to sell its AI software used to help doctors treat cancer patients to hospitals, after it was revealed to be ineffective. The systems were trained on simulated data instead of real medical data.

She will join the strategy team for IBM Cognitive Solutions after three years as the general manager of IBM Watson Health. John Kelly, the senior vice president for Cognitive Solutions and IBM Research will take her place.

Waymo money:Waymo is charging customers for its robo-taxi service in Arizona the Google announced in an earning’s call on Friday.

The company are currently testing its cars in Phoenix for passengers that were accepted for its early rider program. Ruth Porat, chief financial officer, said it was looking at ways to commercialize and that “some customers” have already been paying for rides, according to the Financial Times. The exact costs were not revealed.

Porat also said Waymo is planning to launch a wider, commercial program in Phoenix by the end of the year. Its self driving cars are currently tested in cities across Washington, California, Arizona, Texas, Georgia and Michigan.

AI art sold at for how much?!: The first portrait produced by an AI algorithm has sold for a whopping $432,500, beating initial estimates of over $7,000.

The AI hype has reached the art world. Generative adversarial networks (GANs), made up of a generator and discriminator networks, have been used by researchers to create all sorts of content for a while.

Now, the first AI ‘painting’ has been sold by Christie’s, an auction house in New York City, for $432,500 (~£337,000). It was estimated to sell for around $7,000 to $10,000 (~£5,454 to £7,792).

The artwork features an abstract picture named “Edmond de Belamy, from La Famille de Belamy” and depicts a man dressed in black against a dark backgroung. It’s a bit smudgy and blurry, and doesn’t fill the whole frame.

The painting was created by a group of French students from Obvious, a group focused on democratizing AI through art. People over the internet have reportedly spotted an uncanny resemblence between the style of the painting and the work of Robbie Barrat, an AI artist and researcher at a bioinformatics lab at Stanford University

Barrat’s GAN code was posted publicly on his GitHub, and Obvious admitted they borrowed bits of it. Experts have warned about the potential danger of lifting open source code and repurposing it for malicious uses.

left: the "AI generated" portrait Christie's is auctioning off right now

right: outputs from a neural network I trained and put online *over a year ago*.

Does anyone else care about this? Am I crazy for thinking that they really just used my network and are selling the results? pic.twitter.com/wAdSOe7gwz

— Robbie Barrat (@DrBeef_) October 25, 2018

Obvious said it didn't just copy and paste the pretrained model, but retrained it with new training data to get its own results.

Here is a video of the training process of Edmond, showing we did not use pre-trained models. pic.twitter.com/SfYYx5KZjI

— Obvious (@obv_ious) October 25, 2018

But Barrat is upset that his work was commercialized without any credit. Obvious posted a screenshot of a private Twitter exchange with Barat, where he said he was “100% okay” with his code being used.

Barrat, later said he didn’t know what it was for but believed it was for an open source project. He asked Obvious to credit him for his work, and the group said they did on Medium posts and in interviews.

i had no idea what you were doing with it - "democratized" sounds a lot like you were doing some open source project.

Conveniently cutting out the part where I ask you for credit a few weeks later after I see you posting the images for the first time for sale. pic.twitter.com/GbmoavWhkf

— Robbie Barrat (@DrBeef_) October 25, 2018

Barrat updated his GitHub license to say: “NO OUTPUTS OF THE PRE-TRAINED MODELS MAY BE SOLD OR USED FOR-PROFIT OTHERWISE,” after it had been used by Obvious. ®