25 April 2018

Kogan: “I don’t think Facebook has a developer policy that is valid”


A Cambridge University academic at the center of a data misuse scandal involving Facebook user data and political ad targeting faced questions from the UK parliament this morning.

Although the two-hour evidence session in front of the DCMS committee’s fake news enquiry raised rather more questions than it answered — with professor Aleksandr Kogan citing an NDA he said he had signed with Facebook to decline to answer some of the committee’s questions (including why and when exactly the NDA was signed).

TechCrunch understands the NDA relates to standard confidentiality provisions regarding deletion certifications and other commitments made by Kogan to Facebook not to misuse user data — after the company learned he had user passed data to SCL in contravention of its developer terms.

Asked why he had a non disclosure agreement with Facebook Kogan told the committee it would have to ask Facebook. He also declined to say whether any of his company co-directors (one of whom now works for Facebook) had been asked to sign an NDA. Nor would he specify whether the NDA had been signed in the US.

Asked whether he had deleted all the Facebook data and derivatives he had been able to acquire Kogan said yes “to the best of his knowledge”, though he also said he’s currently conducting a review to make sure nothing has been overlooked.

A few times during the session Kogan made a point of arguing that data audits are essentially useless for catching bad actors — claiming that anyone who wants to misuse data can simply put a copy on a hard drive and “store it under the mattress”.

(Incidentally, the UK’s data protection watchdog is conducting just such an audit of Cambridge Analytica right now, after obtaining a warrant to enter its London offices last month — as part of an ongoing, year-long investigation into social media data being used for political ad targeting.)

Your company didn’t hide any data in that way did it, a committee member asked Kogan? “We didn’t,” he rejoined.

“This has been a very painful experience because when I entered into all of this Facebook was a close ally. And I was thinking this would be helpful to my academic career. And my relationship with Facebook. It has, very clearly, done the complete opposite,” Kogan continued.  “I had no interest in becoming an enemy or being antagonized by one of the biggest companies in the world that could — even if it’s frivolous — sue me into oblivion. So we acted entirely as they requested.”

Despite apparently lamenting the breakdown in his relations with Facebook — telling the committee how he had worked with the company, in an academic capacity, prior to setting up a company to work with SCL/CA — Kogan refused to accept that he had broken Facebook’s terms of service — instead asserting: “I don’t think they have a developer policy that is valid… For you to break a policy it has to exist. And really be their policy, The reality is Facebook’s policy is unlikely to be their policy.”

“I just don’t believe that’s their policy,” he repeated when pressed on whether he had broken Facebook’s ToS. “If somebody has a document that isn’t their policy you can’t break something that isn’t really your policy. I would agree my actions were inconsistent with the language of this document — but that’s slightly different from what I think you’re asking.”

“You should be a professor of semantics,” quipped the committee member who had been asking the questions.

A Facebook spokesperson told us it had no public comment to make on Kogan’s testimony. But last month CEO Mark Zuckerberg couched the academic’s actions as a “breach of trust” — describing the behavior of his app as “abusive”.

In evidence to the committee today, Kogan told it he had only become aware of an “inconsistency” between Facebook’s developer terms of service and what his company did in March 2015 — when he said he begun to suspect the veracity of the advice he had received from SCL. At that point Kogan said GSR reached out to an IP lawyer “and got some guidance”.

(More specifically he said he became suspicious because former SCL employee Chris Wylie did not honor a contract between GSR and Eunoia, a company Wylie set up after leaving SLC, to exchange data-sets; Kogan said GSR gave Wylie the full raw Facebook data-set but Wylie did not provide any data to GSR.)

“Up to that point I don’t believe I was even aware or looked at the developer policy. Because prior to that point — and I know that seems shocking and surprising… the experience of a developer in Facebook is very much like the experience of a user in Facebook. When you sign up there’s this small print that’s easy to miss,” he claimed.

“When I made my app initially I was just an academic researcher. There was no company involved yet. And then when we commercialized it — so we changed the app — it was just something I completely missed. I didn’t have any legal resources, I relied on SCL [to provide me with guidance on what was appropriate]. That was my mistake.”

“Why I think this is still not Facebook’s policy is that we were advised [by an IP lawyer] that Facebook’s terms for users and developers are inconsistent. And that it’s not actually a defensible position for Facebook that this is their policy,” Kogan continued. “This is the remarkable thing about the experience of an app developer on Facebook. You can change the name, you can change the description, you can change the terms of service — and you just save changes. There’s no obvious review process.

“We had a terms of service linked to the Facebook platform that said we could transfer and sell data for at least a year and a half — nothing was ever mentioned. It was only in the wake of the Guardian article [in December 2015] that they came knocking.”

Kogan also described the work he and his company had done for SCL Elections as essentially worthless — arguing that using psychometrically modeled Facebook data for political ad targeting in the way SCL/CA had apparently sought to do was “incompetent” because they could have used Facebook’s own ad targeting platform to achieve greater reach and with more granular targeting.

“It’s all about the use-case. I was very surprised to learn that what they wanted to do is run Facebook ads,” he said. “This was not mentioned, they just wanted a way to measure personality for many people. But if the use-case you have is Facebook ads it’s just incompetent to do it this way.

“Taking this data-set you’re going to be able to target 15% of the population. And use a very small segment of the Facebook data — page likes — to try to build personality models. When do this when you could very easily go target 100% and use much more of the data. It just doesn’t make sense.”

Asked what, then, was the value of the project he undertook for SCL, Kogan responded: “Given what we know now, nothing. Literally nothing.”

He repeated his prior claim that he was not aware that work he was providing for SCL Elections would be used for targeting political ads, though he confirmed he knew the project was focused on the US and related to elections.

He also said he knew the work was being done for the Republican party — but claimed not to know which specific candidates were involved.

Pressed by one committee member on why he didn’t care to know which politicians he was indirectly working for, Kogan responded by saying he doesn’t have strong personal views on US politics or politicians generally — beyond believing that most US politicians are at least reasonable in their policy positions.

“My personal position on life is unless I have a lot of evidence I don’t know. Is the answer. It’s a good lesson to learn from science — where typically we just don’t know. In terms of politics in particular I rarely have a strong position on a candidate,” said Kogan, adding that therefore he “didn’t bother” to make the effort to find out who would ultimately be the beneficiary of his psychometric modeling.

Kogan told the committee his initial intention had not been to set up a business at all but to conduct not-for-profit big data research — via an institute he wanted to establish — claiming it was Wylie who had advised him to also set up the for-profit entity, GSR, through which he went on to engage with SCL Elections/CA.

“The initial plan was we collect the data, I fulfill my obligations to SCL, and then I would go and use the data for research,” he said.

And while Kogan maintained he had never drawn a salary from the work he did for SCL — saying his reward was “to keep the data”, and get to use it for academic research — he confirmed SCL did pay GSR £230,000 at one point during the project; a portion of which he also said eventually went to pay lawyers he engaged “in the wake” of Facebook becoming aware that data had been passed to SCL/CA by Kogan — when it contacted him to ask him to delete the data (and presumably also to get him to sign the NDA).

In one curious moment, Kogan claimed not to know his own company had been registered at 29 Harley Street in London — which the committee noted is “used by a lot of shell companies some of which have been used for money laundering by Russian oligarchs”.

Seeming a little flustered he said initially he had registered the company at his apartment in Cambridge, and later “I think we moved it to an innovation center in Cambridge and then later Manchester”.

“I’m actually surprised. I’m totally surprised by this,” he added.

Did you use an agent to set it up, asked one committee member. “We used Formations House,” replied Kogan, referring to a company whose website states it can locate a business’ trading address “in the heart of central London” — in exchange for a small fee.

“I’m legitimately surprised by that,” added Kogan of the Harley Street address. “I’m unfortunately not a Russian oligarch.”

Later in the session another odd moment came when he was being asked about his relationship with Saint Petersburg University in Russia — where he confirmed he had given talks and workshops, after traveling to the country with friends and proactively getting in touch with the university “to say hi” — and specifically about some Russian government-funded research being conducted by researchers there into cyberbullying.

Committee chair Collins implied to Kogan the Russian state could have had a specific malicious interest in such a piece of research, and wondered whether Kogan had thought about that in relation to the interactions he’d had with the university and the researchers.

Kogan described it as a “big leap” to connect the piece of research to Kremlin efforts to use online platforms to interfere in foreign elections — before essentially going on to repeat a Kremlin talking point by saying the US and the UK engage in much the same types of behavior.

“You can make the same argument about the UK government funding anything or the US government funding anything,” he told the committee. “Both countries are very famous for their spies.

“There’s a long history of the US interfering with foreign elections and doing the exact same thing [creating bot networks and using trolls for online intimidation].”

“Are you saying it’s equivalent?” pressed Collins. “That the work of the Russian government is equivalent to the US government and you couldn’t really distinguish between the two?”

“In general I would say the governments that are most high profile I am dubious about the moral scruples of their activities through the long history of UK, US and Russia,” responded Kogan. “Trying to equate them I think is a bit of a silly process. But I think certainly all these countries have engaged in activities that people feel uncomfortable with or are covert. And then to try to link academic work that’s basic science to that — if you’re going to down the Russia line I think we have to go down the UK line and the US line in the same way.

“I understand Russia is a hot-button topic right now but outside of that… Most people in Russia are like most people in the UK. They’re not involved in spycraft, they’re just living lives.”

“I’m not aware of UK government agencies that have been interfering in foreign elections,” added Collins.

“Doesn’t mean it’s not happened,” replied Kogan. “Could be just better at it.”

During Wylie’s evidence to the committee last month the former SCL data scientist had implied there could have been a risk of the Facebook data falling into the hands of the Russian state as a result of Kogan’s back and forth travel to the region. But Kogan rebutted this idea — saying the data had never been in his physical possession when he traveled to Russia, pointing out it was stored in a cloud hosting service in the US.

“If you want to try to hack Amazon Web Services good luck,” he added.

He also claimed not to have read the piece of research in question, even though he said he thought the researcher had emailed the paper to him — claiming he can’t read Russian well.

Kogan seemed most comfortable during the session when he was laying into Facebook’s platform policies — perhaps unsurprisingly, given how the company has sought to paint him as a rogue actor who abused its systems by creating an app that harvested data on up to 87 million Facebook users and then handing information on its users off to third parties.

Asked whether he thought a prior answer given to the committee by Facebook — when it claimed it had not provided any user data to third parties — was correct, Kogan said no given the company provides academics with “macro level” user data (including providing him with this type of data, in 2013).

He was also asked why he thinks Facebook lets its employees collaborate with external researchers — and Kogan suggested this is “tolerated” by management as a strategy to keep employees stimulated.

Committee chair Collins asked whether he thought it was odd that Facebook now employs his former co-director at GSR, Joseph Chancellor — who works in its research division — despite Chancellor having worked for a company Facebook has said it regards as having violated its platform policies.

“Honestly I don’t think it’s odd,” said Kogan. “The reason I don’t think it’s odd is because in my view Facebook’s comments are PR crisis mode. I don’t believe they actually think these things — because I think they realize that their platform has been mined, left and right, by thousands of others.

“And I was just the unlucky person that ended up somehow linked to the Trump campaign. And we are where we are. I think they realize all this but PR is PR and they were trying to manage the crisis and it’s convenient to point the finger at a single entity and try to paint the picture this is a rogue agent.

At another moment during the evidence session Kogan was also asked to respond to denials previously given to the committee by former CEO of Cambridge Analytica Alexander Nix — who had claimed that none of the data it used came from GSR and — even more specifically — that GSR had never supplied it with “data-sets or information”.

“Fabrication,” responded Kogan. “Total fabrication.”

“We certainly gave them [SCL/CA] data. That’s indisputable,” he added.

In written testimony to the committee he also explained that he in fact created three apps for gathering Facebook user data. The first one — called the CPW Lab app — was developed after he had begun a collaboration with Facebook in early 2013, as part of his academic studies. Kogan says Facebook provided him with user data at this time for his research — although he said these datasets were “macro-level datasets on friendship connections and emoticon usage” rather than information on individual users.

The CPW Lab app was used to gather individual level data to supplement those datasets, according to Kogan’s account. Although he specifies that data collected via this app was housed at the university; used for academic purposes only; and was “not provided to the SCL Group”.

Later, once Kogan had set up GSR and was intending to work on gathering and modeling data for SCL/Cambridge Analytica, the CPW Lab app was renamed to the GSR App and its terms were changed (with the new terms provided by Wylie).

Thousands of people were then recruited to take this survey via a third company — Qualtrics — with Kogan saying SCL directly paid ~$800,000 to it to recruit survey participants, at a cost of around $3-$4 per head (he says between 200,000 and 300,000 people took the survey as a result in the summer of 2014; NB: Facebook doesn’t appear to be able to break out separate downloads for the different apps Kogan ran on its platform — it told us about 305,000 people downloaded “the app”).

In the final part of that year, after data collection had finished for SCL, Kogan said his company revised the GSR App to become an interactive personality quiz — renaming it “thisisyourdigitallife” and leaving the commercial portions of the terms intact.

“The thisisyourdigitallife App was used by only a few hundred individuals and, like the two prior iterations of the application, collected demographic information and data about “likes” for survey participants and their friends whose Facebook privacy settings gave participants access to “likes” and demographic information. Data collected by the thisisyourdigitallife App was not provided to SCL,” he claims in the written testimony.

During the oral hearing, Kogan was pressed on misleading T&Cs in his two commercial apps. Asked by a committee member about the terms of the GSR App not specifying that the data would be used for political targeting, he said he didn’t write the terms himself but added: “If we had to do it again I think I would have insisted to Mr Wylie that we do add politics as a use-case in that doc.”

“It’s misleading,” argued the committee member. “It’s a misrepresentation.”

“I think it’s broad,” Kogan responded. “I think it’s not specific enough. So you’re asking for why didn’t we go outline specific use-cases — because the politics is a specific use-case. I would argue that the politics does fall under there but it’s a specific use-case. I think we should have.”

The committee member also noted how, “in longer, denser paragraphs” within the app’s T&Cs, the legalese does also state that “whatever that primary purpose is you can sell this data for any purposes whatsoever” — making the point that such sweeping terms are unfair.

“Yes,” responded Kogan. “In terms of speaking the truth, the reality is — as you’ve pointed out — very few if any people have read this, just like very few if any people read terms of service. I think that’s a major flaw we have right now. That people just do not read these things. And these things are written this way.”

“Look — fundamentally I made a mistake by not being critical about this. And trusting the advice of another company [SCL]. As you pointed out GSR is my company and I should have gotten better advice, and better guidance on what is and isn’t appropriate,” he added.

“Quite frankly my understanding was this was business as usual and normal practice for companies to write broad terms of service that didn’t provide specific examples,” he said after being pressed on the point again.

“I doubt in Facebook’s user policy it says that users can be advertised for political purposes — it just has broad language to provide for whatever use cases they want. I agree with you this doesn’t seem right, and those changes need to be made.”

At another point, he was asked about the Cambridge University Psychometrics Centre — which he said had initially been involved in discussions between him and SCL to be part of the project but fell out of the arrangement. According to his version of events the Centre had asked for £500,000 for their piece of proposed work, and specifically for modeling the data — which he said SCL didn’t want to pay. So SCL had asked him to take that work on too and remove the Centre from the negotiations.

As a result of that, Kogan said the Centre had complained about him to the university — and SCL had written a letter to it on his behalf defending his actions.

“The mistake the Psychometrics Centre made in the negotiation is that they believed that models are useful, rather than data,” he said. “And actually just not the same. Data’s far more valuable than models because if you have the data it’s very easy to build models — because models use just a few well understood statistical techniques to make them. I was able to go from not doing machine learning to knowing what I need to know in one week. That’s all it took.”

In another exchange during the session, Kogan denied he had been in contact with Facebook in 2014. Wylie previously told the committee he thought Kogan had run into problems with the rate at which the GSR App was able to pull data off Facebook’s platform — and had contacted engineers at the company at the time (though Wylie also caveated his evidence by saying he did not know whether what he’d been told was true).

“This never happened,” said Kogan, adding that there was no dialogue between him and Facebook at that time.  “I don’t know any engineers at Facebook.”


Read Full Article

No comments:

Post a Comment