{"id":217094,"date":"2022-07-25T12:00:19","date_gmt":"2022-07-25T11:00:19","guid":{"rendered":"https:\/\/www.transcend.org\/tms\/?p=217094"},"modified":"2022-07-25T06:34:57","modified_gmt":"2022-07-25T05:34:57","slug":"documents-reveal-advanced-ai-tools-google-is-selling-to-israel","status":"publish","type":"post","link":"https:\/\/www.transcend.org\/tms\/2022\/07\/documents-reveal-advanced-ai-tools-google-is-selling-to-israel\/","title":{"rendered":"Documents Reveal Advanced AI Tools Google Is Selling to Israel"},"content":{"rendered":"

Google employees, who have been kept in the dark about the \u201cNimbus\u201d AI project, have concerns about Israeli human rights abuses.<\/em><\/p><\/blockquote>\n

\"\"<\/a>

Illustration: Alex Hinton for The Intercept<\/p><\/div>\n

24 Jul 2022 – <\/em>Training materials reviewed by The Intercept confirm that Google is offering advanced artificial intelligence and machine-learning capabilities to the Israeli government through its controversial \u201cProject Nimbus\u201d contract. The Israeli Finance Ministry announced the contract<\/a> in April 2021 for a $1.2 billion cloud computing system jointly built by Google and Amazon. \u201cThe project is intended to provide the government, the defense establishment and others with an all-encompassing cloud solution,\u201d the ministry said in its announcement.<\/p>\n

\n
\n

Google engineers have spent the time since worrying whether their efforts would inadvertently bolster the ongoing Israeli military occupation of Palestine. In 2021, both Human Rights Watch<\/a> and Amnesty International<\/a> formally accused Israel of committing crimes against humanity by maintaining an apartheid system against Palestinians. While the Israeli military and security services already<\/a> rely on a sophisticated system<\/a> of computerized surveillance<\/a>, the sophistication of Google\u2019s data analysis offerings could worsen the increasingly<\/a> data-driven military occupation.<\/p>\n<\/div>\n

\n

According to a trove of training<\/a> documents<\/a> and videos obtained by The Intercept through a publicly accessible educational portal intended for Nimbus users, Google is providing the Israeli government with the full suite of machine-learning and AI tools available through Google Cloud Platform. While they provide no specifics as to how Nimbus will be used, the documents indicate that the new cloud would give Israel capabilities for facial detection, automated image categorization, object tracking, and even sentiment analysis that claims to assess the emotional content of pictures, speech, and writing. The Nimbus materials referenced agency-specific trainings available to government personnel through the online learning service Coursera, citing the Ministry of Defense as an example.<\/p>\n<\/div>\n

\n
\n

\"A<\/p>\n

A slide presented to Nimbus users illustrating Google image recognition technology. Credit: Google<\/strong><\/p>\n<\/div>\n<\/div>\n

\n

Jack Poulson, director of the watchdog group Tech Inquiry, shared the portal\u2019s address with The Intercept after finding it cited in Israeli contracting documents.\u201cThe former head of Security for Google Enterprise \u2014 who now heads Oracle\u2019s Israel branch \u2014 has publicly argued that one of the goals of Nimbus is preventing the German government from requesting data relating on the Israel Defence Forces for the International Criminal Court,\u201d said Poulson, who resigned in protest from his job as a research scientist at Google in 2018, in a message. \u201cGiven Human Rights Watch\u2019s conclusion that the Israeli government is committing \u2018crimes against humanity of apartheid and persecution\u2019 against Palestinians, it is critical that Google and Amazon\u2019s AI surveillance support to the IDF be documented to the fullest.\u201d<\/p>\n<\/div>\n

\n

Though some of the documents bear a hybridized symbol of the Google logo and Israeli flag, for the most part they are not unique to Nimbus. Rather, the documents appear to be standard educational materials distributed to Google Cloud customers and presented in prior training contexts elsewhere.<\/p>\n

Google did not respond to a request for comment.<\/p>\n

The documents obtained by The Intercept detail for the first time the Google Cloud features provided through the Nimbus contract. With virtually nothing publicly disclosed about Nimbus beyond its existence, the system\u2019s specific functionality had remained a mystery even to most of those working at the company that built it.\u00a0In 2020, citing the same AI tools, U.S Customs and Border Protection tapped Google Cloud to <\/a>process imagery<\/a> from its network of border surveillance towers.<\/p>\n

Many of the capabilities outlined in the documents obtained by The Intercept could easily augment Israel\u2019s ability to surveil people and process vast stores of data \u2014 already prominent features of the Israeli occupation.<\/p>\n

\u201cData collection over the entire Palestinian population was and is an integral part of the occupation,\u201d Ori Givati of Breaking the Silence, an anti-occupation advocacy group of Israeli military veterans, told The Intercept in an email. \u201cGenerally, the different technological\u00a0developments we are seeing in the Occupied Territories all direct to one central element which is more control.\u201d<\/p>\n

The Israeli security<\/u> state has for decades benefited from the country\u2019s thriving research and development sector<\/a>, and its interest in using AI to police and control Palestinians isn\u2019t hypothetical. In 2021, the Washington Post reported on the existence of Blue Wolf,<\/a> a secret military program aimed at monitoring Palestinians through a network of facial recognition-enabled smartphones and cameras.<\/p>\n

\u201cLiving under a surveillance state for years taught us that all the collected information in the Israeli\/Palestinian context could be securitized and militarized,\u201d said Mona Shtaya, a Palestinian digital rights advocate at 7amleh-The Arab Center for Social Media Advancement, in a message. \u201cImage recognition, facial recognition, emotional analysis, among other things will increase the power of the surveillance state to violate Palestinian right to privacy and to serve their main goal, which is to create the panopticon feeling among Palestinians that we are being watched all the time, which would make the Palestinian population control easier.\u201d<\/p>\n

The educational materials obtained by The Intercept show that Google briefed the Israeli government on using what\u2019s known as sentiment detection, an increasingly controversial and discredited form of machine learning. Google claims that its systems can discern inner feelings from one\u2019s face and statements, a technique commonly rejected as invasive and pseudoscientific, regarded as being little better than phrenology<\/a>. In June, Microsoft announced that it would no longer offer<\/a> emotion-detection features through its Azure cloud computing platform \u2014 a technology suite comparable to what Google provides with Nimbus \u2014 citing the lack of scientific basis.<\/p>\n<\/div>\n

\n

Google does not appear to share Microsoft\u2019s concerns. One Nimbus presentation touted<\/a> the \u201cFaces, facial landmarks, emotions\u201d-detection capabilities of Google\u2019s Cloud Vision API, an image analysis toolset. The presentation then offered a demonstration using the enormous grinning face sculpture at the entrance of Sydney\u2019s Luna Park. An included screenshot of the feature ostensibly in action indicates that the massive smiling grin is \u201cvery unlikely\u201d to exhibit any of the example emotions. And Google was only able to assess that the famous amusement park is an amusement park with 64 percent certainty, while it guessed that the landmark was a \u201cplace of worship\u201d or \u201cHindu Temple\u201d with 83 percent and 74 percent confidence, respectively.<\/p>\n<\/div>\n

\n
\n

\"A<\/p>\n

A slide presented to Nimbus users illustrating Google AI\u2019s ability to detect image traits. <\/strong>Credit: Google<\/strong><\/p>\n<\/div>\n<\/div>\n

\n

Google workers who reviewed the documents said they were concerned by their employer\u2019s sale of these technologies to Israel, fearing both their inaccuracy and how they might be used for surveillance or other militarized purposes.\u201cVision API is a primary concern to me because it\u2019s so useful for surveillance,\u201d said one worker, who explained that the image analysis would be a natural fit for military and security applications. \u201cObject recognition is useful for targeting, it\u2019s useful for data analysis and data labeling. An AI can comb through collected surveillance feeds in a way a human cannot to find specific people and to identify people, with some error, who look like someone. That\u2019s why these systems are really dangerous.\u201d<\/p>\n<\/div>\n

\n
\n

\"A<\/p>\n

A slide presented to Nimbus users outlining various AI features through the company\u2019s Cloud Vision API. Credit: Google<\/strong><\/p>\n<\/div>\n<\/div>\n

\n

The employee \u2014 who, like other Google workers who spoke to The Intercept, requested anonymity to avoid workplace reprisals \u2014 added that they were further alarmed by potential surveillance or other militarized applications of AutoML, another Google AI tool offered through Nimbus. Machine learning is largely the function of training software to recognize patterns in order to make predictions about future observations, for instance by analyzing millions of images of kittens today in order to confidently claim that it\u2019s looking at a photo of a kitten tomorrow. This training process yields what\u2019s known as a \u201cmodel\u201d \u2014 a body of computerized education that can be applied to automatically recognize certain objects and traits in future data.<\/p>\n

Training an effective model from scratch is often resource intensive, both financially and computationally. This is not so much of a problem for a world-spanning company like Google, with an unfathomable volume of both money and computing hardware at the ready. Part of Google\u2019s appeal to customers is the option of using a pre-trained model, essentially getting this prediction-making education out of the way and letting customers access a well-trained program that\u2019s benefited from the company\u2019s limitless resources.<\/p>\n<\/div>\n

\n
\u201cAn AI can comb through collected surveillance feeds in a way a human cannot to find specific people and to identify people, with some error, who look like someone. That\u2019s why these systems are really dangerous.\u201d<\/strong><\/em><\/div>\n<\/blockquote>\n
\n

Cloud Vision is one such pre-trained model, allowing clients to immediately implement a sophisticated prediction system. AutoML, on the other hand, streamlines the process of training a custom-tailored model, using a customer\u2019s own data for a customer\u2019s own designs. Google has placed some limits on Vision \u2014 for instance limiting it to face\u00a0detection, or whether it sees a face, rather than recognition that would identify a person. AutoML, however,\u00a0would allow Israel to leverage Google\u2019s computing capacity to train new models with its own government data for virtually any purpose it wishes. \u201cGoogle\u2019s machine learning capabilities along with the Israeli state\u2019s surveillance infrastructure poses a real threat to the human rights of Palestinians,\u201d said Damini Satija, who leads Amnesty International\u2019s Algorithmic Accountability Lab. \u201cThe option to use the vast volumes of surveillance data already held by the Israeli government to train the systems only exacerbates these risks.\u201dCustom models generated through AutoML, one presentation noted, can be downloaded for offline \u201cedge\u201d use \u2014 unplugged from the cloud and deployed in the field.<\/p>\n

That Nimbus lets<\/u> Google clients use advanced data analysis and prediction in places and ways that Google has no visibility into creates a risk of abuse, according to Liz O\u2019Sullivan, CEO of the AI auditing startup\u00a0Parity and a member of the U.S. National Artificial Intelligence Advisory Committee. \u201cCountries can absolutely use AutoML to deploy shoddy surveillance systems that only seem like they work,\u201d O\u2019Sullivan said in a message. \u201cOn edge, it\u2019s even worse \u2014 think bodycams, traffic cameras, even a handheld device like a phone can become a surveillance machine and Google may not even know it\u2019s happening.\u201d<\/p>\n

In one Nimbus webinar reviewed by The Intercept, the potential use and misuse of AutoML was exemplified in a Q&A session following a presentation. An unnamed member of the audience asked the Google Cloud engineers present on the call if it would be possible to process data through Nimbus in order to determine if someone is lying.<\/p>\n

\u201cI\u2019m a bit scared to answer that question,\u201d said the engineer conducting the seminar, in an apparent joke. \u201cIn principle: Yes. I will expand on it, but the short answer is yes.\u201d Another Google representative then jumped in: \u201cIt is possible, assuming that you have the right data, to use the Google infrastructure to train a model to identify how likely it is that a certain person is lying, given the sound of their own voice.\u201d Noting that such a capability would take a tremendous amount of data for the model, the second presenter added that one of the advantages of Nimbus is the ability to tap into Google\u2019s vast computing power to train such a model.<\/p>\n<\/div>\n

\n
\u201cI\u2019d be very skeptical for the citizens it is meant to protect that these systems can do what is claimed.\u201d<\/strong><\/em><\/div>\n<\/blockquote>\n
\n

A broad body of research, however, has shown that the very notion of a \u201clie detector,\u201d whether the simple polygraph or \u201cAI\u201d-based analysis of vocal changes or facial cues, is junk science. While Google\u2019s reps appeared confident that the company could make such a thing possible through sheer computing power, experts in the field say that any attempts to use computers to assess things as profound and intangible as truth and emotion are faulty to the point of danger.<\/p>\n

One Google worker who reviewed the documents said they were concerned that the company would even hint at such a scientifically dubious technique. \u201cThe answer should have been \u2018no,\u2019 because that does not exist,\u201d the worker said. \u201cIt seems like it was meant to promote Google technology as powerful, and it\u2019s ultimately really irresponsible to say that when it\u2019s not possible.\u201d<\/p>\n

Andrew McStay, a professor of digital media at Bangor University in Wales and\u00a0head of the Emotional AI Lab<\/a>, told The Intercept that the lie detector Q&A exchange was \u201cdisturbing,\u201d as is Google\u2019s willingness to pitch pseudoscientific AI tools to a national government. \u201cIt is [a] wildly divergent field, so any technology built on this is going to automate unreliability,\u201d he said. \u201cAgain, those subjected to them will suffer, but I\u2019d be very skeptical for the citizens it is meant to protect that these systems can do what is claimed.\u201d<\/p>\n

According to some critics, whether these tools work might be of secondary importance to a company like Google that is eager to tap the ever-lucrative flow of military contract money. Governmental customers too may be willing to suspend disbelief when it comes to promises of vast new techno-powers. \u201cIt\u2019s extremely telling that in the webinar PDF that they constantly referred to this as \u2018magical AI goodness,\u2019\u201d said Jathan Sadowski, a scholar of automation technologies and research fellow at Monash University, in an interview with The Intercept. \u201cIt shows that they\u2019re bullshitting.\u201d<\/p>\n<\/div>\n

\n
\n

\"FILE-<\/p>\n

Google CEO Sundar Pichai speaks at the Google I\/O conference in Mountain View, Calif. Google pledges that it will not use artificial intelligence in applications related to weapons or surveillance, part of a new set of principles designed to govern how it uses AI. Those principles, released by Pichai, commit Google to building AI applications that are \u201csocially beneficial,\u201d that avoid creating or reinforcing bias and that are accountable to people. Photo: Jeff Chiu\/AP<\/strong><\/p>\n<\/div>\n<\/div>\n

\n

Google, like Microsoft,<\/u> has its own public list of \u201cAI principles<\/a>,\u201d a document the company says is an \u201cethical charter that guides the development and use of artificial intelligence in our research and products.\u201d Among these purported principles is a commitment to not \u201cdeploy AI \u2026 that cause or are likely to cause overall harm,\u201d including weapons, surveillance, or any application \u201cwhose purpose contravenes widely accepted principles of international law and human rights.\u201dIsrael, though, has set up its relationship with Google to shield it from both the company\u2019s principles and any outside scrutiny. Perhaps fearing the fate of the Pentagon\u2019s Project Maven<\/a>, a Google AI contract felled by intense employee protests, the data centers that power Nimbus will reside on Israeli territory, subject to Israeli law<\/a>\u00a0and insulated from political pressures. Last year, the Times of Israel reported<\/a> that Google would be contractually barred from shutting down Nimbus services or denying access to a particular government office even in response to boycott campaigns.<\/p>\n

Google employees interviewed by The Intercept lamented that the company\u2019s AI principles are at best a superficial gesture. \u201cI don\u2019t believe it\u2019s hugely meaningful,\u201d one employee told The Intercept, explaining that the company has interpreted its AI charter so narrowly that it doesn\u2019t apply to companies or governments that buy Google Cloud services. Asked how the AI principles are compatible with the company\u2019s Pentagon work, a Google spokesperson told Defense One<\/a>, \u201cIt means that our technology can be used fairly broadly by the military.\u201d<\/p>\n<\/div>\n

\n
\u201cGoogle is backsliding on its commitments to protect people from this kind of misuse of our technology. I am truly afraid for the future of Google and the world.\u201d<\/strong><\/em><\/div>\n<\/blockquote>\n
\n

Moreover, this employee added that Google lacks both the ability to tell if its principles are being violated and any means of thwarting violations. \u201cOnce Google offers these services, we have no technical capacity to monitor what our customers are doing with these services,\u201d the employee said. \u201cThey could be doing anything.\u201d Another Google worker told The Intercept, \u201cAt a time when already vulnerable populations are facing unprecedented and escalating levels of repression, Google is backsliding on its commitments to protect people from this kind of misuse of our technology. I am truly afraid for the future of Google and the world.\u201d<\/p>\n

Ariel Koren, a Google employee who claimed earlier this year that she faced retaliation<\/a> for raising concerns about Nimbus, said the company\u2019s internal silence on the program continues. \u201cI am deeply concerned that Google has not provided us with any details at all about the scope of the Project Nimbus contract, let alone assuage my concerns of how Google can provide technology to the Israeli government and military (both committing grave human rights abuses against Palestinians daily) while upholding the ethical commitments the company has made to its employees and the public,\u201d she told The Intercept in an email. \u201cI joined Google to promote technology that brings communities together and improves people\u2019s lives, not service a government accused of the crime of apartheid by the world\u2019s two leading human rights organizations.\u201d<\/p>\n

Sprawling tech\u00a0companies have published ethical AI charters to rebut critics who say that their increasingly powerful products are sold unchecked and unsupervised. The same critics often counter that the documents are a form of \u201cethicswashing\u201d \u2014 essentially toothless self-regulatory pledges that provide only the appearance of scruples, pointing to examples like the provisions in Israel\u2019s contract with Google that prevent the\u00a0company from shutting down its products. \u201cThe way that Israel is locking in their service providers through this tender and this contract,\u201d said Sadowski, the Monash University scholar, \u201cI do feel like that is a real innovation in technology procurement.\u201d<\/p>\n

To Sadowski, it matters little whether Google believes what it peddles about AI or any other technology. What the company is selling, ultimately, isn\u2019t just software, but power. And whether it\u2019s Israel and the U.S. today or another government tomorrow, Sadowski says that some technologies amplify the exercise of power to such an extent that even their use by a country with a spotless human rights record would provide little reassurance. \u201cGive them these technologies, and see if they don\u2019t get tempted to use them in really evil and awful ways,\u201d he said. \u201cThese are not technologies that are just neutral intelligence systems, these are technologies that are ultimately about surveillance, analysis, and control.\u201d<\/p>\n

_____________________________________________<\/p>\n

\"\"<\/a><\/em><\/p>\n

 <\/p>\n

Sam Biddle – <\/em><\/a>sam.biddle@\u200btheintercept.com<\/a><\/em><\/p>\n

 <\/p>\n

 <\/p>\n

Go to Original – theintercept.com<\/a><\/p>\n<\/div>\n<\/div>\n","protected":false},"excerpt":{"rendered":"

24 Jul 2022 – Google employees, who have been kept in the dark about the \u201cNimbus\u201d AI project, have concerns about Israeli human rights abuses.<\/p>\n","protected":false},"author":4,"featured_media":217096,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[60],"tags":[462,88,1109,911,2841],"class_list":["post-217094","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-whistleblowing-surveillance","tag-google","tag-israel","tag-spying","tag-surveillance","tag-tyranny"],"_links":{"self":[{"href":"https:\/\/www.transcend.org\/tms\/wp-json\/wp\/v2\/posts\/217094","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.transcend.org\/tms\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.transcend.org\/tms\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.transcend.org\/tms\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/www.transcend.org\/tms\/wp-json\/wp\/v2\/comments?post=217094"}],"version-history":[{"count":0,"href":"https:\/\/www.transcend.org\/tms\/wp-json\/wp\/v2\/posts\/217094\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.transcend.org\/tms\/wp-json\/wp\/v2\/media\/217096"}],"wp:attachment":[{"href":"https:\/\/www.transcend.org\/tms\/wp-json\/wp\/v2\/media?parent=217094"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.transcend.org\/tms\/wp-json\/wp\/v2\/categories?post=217094"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.transcend.org\/tms\/wp-json\/wp\/v2\/tags?post=217094"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}