Oral evidence: Fake News, HC 363
Oral evidence, 24 April 2018.
Witness
Dr Aleksandr Kogan.
Chair
Welcome to this further evidence session of the Digital, Culture, Media and Sport Committee, as part of our inquiry into disinformation and fake news. We are delighted to welcome Dr Aleksandr Kogan to give evidence to the Committee this morning. Although there is a wide range of topics that I am sure will come up, and this is a complex matter, I think there are two or three quite clear areas of interest to the Committee, which I am sure will come out during questioning. First, there is what has been called the Facebook data breach, in which you, Dr Kogan, have been seen as a central player. We want to understand more about that, and in particular Facebook’s knowledge of the work that you were doing, and the extent of that work. Secondly, we want to understand more about the use that Cambridge Analytica made of the work that you did and your relationship with them. Thirdly, we want to understand a little more about your work in general, your academic research, and the work you have done for other people. Looking at that first area—trying to understand a bit more about the way in which you worked with Facebook data, and Facebook’s knowledge of the work that you were doing—I was very interested to read the written statement that you supplied to the Committee, which the Committee has now published and is available online through the Committee’s website for anyone to read. You said: “Throughout 2013, Facebook provided me with several macro-level datasets on friendship connections and emoticon usage.” Could you explain a little more about that? This was in the period before you developed the app that is at the centre of the Facebook data breach. What sort of information did Facebook provide you with?
Aleksandr Kogan
We were collaborating with a team at Facebook—I believe they worked on protect and care. The dataset we received initially was about every friendship made in the world between every country over a number of years, and it was broken down by month. It was how many friendships were created between the United States and the UK in, say, March 2011, how many friendships were made between the UK and Canada in April 2012, and things like that. When we say “macro”, we mean at the country level or in the aggregate. Later on, they provided me with datasets about emotional expression. These are the emojis that we often see—the smiley face, the sad face, and things like that. This was also in the aggregate. It was a bit more segmented, by which I mean it was broken down by gender and age groups—how many smiley faces were used by, say, men who were 20 to 30 years old, and things like that. That was for a particular week in, I believe, August 2013.
Chair
From the information that was given to you by Facebook, was it possible to break that down in any way? Could you extrapolate information about individuals?
Aleksandr Kogan
No.
Chair
It was purely macro-level data.
Aleksandr Kogan
Exactly.
Chair
Was that data useful to you in the work that you were doing? Did it help you to design other tools? Did it help you to design apps that you went on to work on?
Aleksandr Kogan
It was useful from an academic perspective. We were using it purely for the lab, and we were studying things like whether we can predict how much money is donated during natural disasters by different countries based on the number of friendships that exist between the two nations. It was really this macro perspective.
Chair
How did you become introduced to Facebook to the extent that they were prepared to give you data like this to work with?
Aleksandr Kogan
My mentor and adviser from my undergrad days at UC Berkeley was consulting for them. He was working with a protect and care team to solve problems of people posting embarrassing photos and not taking them down. He was helping them to understand how we can better tackle that. The focus was on emotions and speaking to people on an emotional level: “Hey, this isn’t making me feel good. Would you mind taking it down?” My understanding was that this was really helpful to Facebook in combating that issue.
Chair
Who was the person or people at Facebook who you worked with at this time on this project?
Aleksandr Kogan
There were a couple of user experience researchers who were on the protect and care team.
Chair
Are you able to say who they were?
Aleksandr Kogan
I can, but I would prefer to do so privately—they are not really central to the story and I want to protect their confidence.
Chair
That is understood and perhaps we can follow up on that separately. When Facebook gave evidence to us in February, it said that it did not give out any Facebook user data, but clearly it did, even if it was in macro form.
Aleksandr Kogan
I think that is tricky. Most data that we will look at around the world, and public data stats, comes ultimately from the individual. Even World Bank data or World Health Organisation data—ultimately, any economic activity is driven, at least at the start, by people. When you deliver to somebody you are going to get it at country level or organisation level, and I think Facebook is thinking about it in the same way. We are giving you aggregated information that was generated from users—once you go up to that level it is very difficult to go back down.
Chair
I suppose that if you are giving people data on a macro level, the purpose of doing that is to help you design tools to target Facebook users on an individual basis who understand how you might do that.
Aleksandr Kogan
Not in this case. This was really an act of collaboration. The people I was working with had PhDs, so they had a strong interest in research generally. They were now working at Facebook as user experience researchers, but they maintained that academic passion. This was a chance for them to go and explore that academic passion. That was really the focus; there is really no conversation at this point about any Facebook tools or anything like that. It was really one way for us to benefit and build research.
Chair
All research has a purpose, and as you said earlier, if you were looking at how you could encourage people from one country to give money to support disaster relief in another country, the methodology for doing that could be derived from the work you were doing looking at the interactions between Facebook users in different countries.
Aleksandr Kogan
I think that is quite downstream. We are doing very early basic science. Before we get to application there are a lot of steps there. In our case we are looking just at associations. There is really interesting potential for impact, but that is quite downstream and requires quite a bit of extra research. I do not know if Facebook would really be the vehicle for achieving this. I think Governments would be.
Chair
But for downstream to exist there has to be an upstream. It has to start somewhere, and this would seem to be an area where it would start. From what you said now and in your statement, Facebook has not been clear when it discussed the way that the data it holds is used by outside organisations. It sought to draw a very clear distinction in its evidence to the Committee between Facebook user data that is accessed by developers through tools, and apps that they create and interact with, and data that Facebook holds. It said to the Committee that it does not share data that it holds on users—the only user data that is gathered by developers is done through tools that they have created. From what you said, I think it is not as clear as that.
Aleksandr Kogan
Yes. If your basis is that aggregated data is derived from users, and that in your mind would still count as user data, then I would agree with that statement.
Chair
And Facebook clearly had a programme of working with academics and outside institutions, in many ways to try to help it understand the aggregated data based on its users.
Aleksandr Kogan
Yes, I would not go as far as ascribing that much intent, as far as its desire, but it is well documented that Facebook collaborates with researchers. You can just go and look at the public records and publications of Facebook researchers and academics working together in publications—there are a number of them.
Chair
But it is interesting because in some ways people sought to portray you as some sort of rogue operator, but nevertheless you are someone who before the development of the app—we will come on to talk about that—was already working with Facebook and using Facebook data that had been given to you by Facebook.
Aleksandr Kogan
Yes, sir.
Chair
Given some of the things that have been said about you, some people might find that quite surprising. Certainly, based on the evidence the Committee has received from Facebook, I think that is surprising. At the end of this particular pre-project that you were working on, what happened to the data that had been used to run these tests and analysis at the end of the process?
Aleksandr Kogan
We were going to hold on to it indefinitely, because we were writing many papers. Throughout that year—2013 and into 2014—we were writing about 10 different publications on it with different ideas. Normally, we would hold on to the dataset to generate further publications in the future. Once December 2015 comes along and Facebook reaches out, they asked me to delete everything, including this academic dataset, which I did.
Chair
Okay. So there was no requirement from Facebook, when they gave you that data in the first place, that you should destroy it or give it back.
Aleksandr Kogan
No. In fact, there was not even a signed agreement initially. They gave me the dataset without any agreement signed. It was just, “Here’s an email. Here’s the dataset.” Sometime later—maybe even as far as a year later—we actually had a signed agreement. If you recall, there was a pretty big scandal about a “Facebook is trying to make people sad” publication—I think it was the fall of 2013—and in the wake of that they started to look more carefully at academic collaborations, and have them formalised a bit better.
Chair
What was Facebook trying to achieve out of this, do you think?
Aleksandr Kogan
In terms of—
Chair
What was the value to Facebook of you doing this work?
Aleksandr Kogan
I think it makes their employees happy. My perception was that management tolerated this. It was not a focus, because obviously it takes away time from their employees working on how to make Facebook a better platform. It was something they gave their employees to stimulate them—to allow them to have this relationship.
Chair
Okay, so they are saying to their employees, “You can take macro-level Facebook user data and give it to academics who don’t work for us without any kind of contract or licence, and let them play with it and just see what happens.”
Aleksandr Kogan
Yes, sir.
Chair
As simple as that.
Aleksandr Kogan
As simple as that.
Chair
It seems pretty open, doesn’t it?
Aleksandr Kogan
Yes. The company is very open.
Chair
Maybe too open.
Aleksandr Kogan
Arguably so.
Chair
Open not with their own information, but with their users’ data.
Aleksandr Kogan
Given the last month, there is a lot of credibility to that statement.
Chair
Thank you very much.
Christian Matheson
Dr Kogan, good morning. With the Chairman’s permission, I might need to leave a little bit before the end of the evidence session, but no disrespect is meant. You are an academic psychologist.
Aleksandr Kogan
Yes, sir.
Christian Matheson
But you have also developed several apps that managed to harvest a large amount of data.
Aleksandr Kogan
Yes, sir.
Christian Matheson
And that data then had a financial value of its own.
Aleksandr Kogan
Yes, sir.
Christian Matheson
Which came first—the academic research or the idea that you might make a business out of this?
Aleksandr Kogan
Certainly the academic research.
Christian Matheson
And at what point did you suddenly think, “Ah, there’s money to be made here”?
Aleksandr Kogan
If we stop and take a step back, the app was initially created to help us with the relationship we had with Facebook on the publications. As I said, we were writing 10 papers. I am a psychologist, so we were working at an individual level, typically, rather than the macro level that an economist or a sociologist would be interest in. We created this app to add data to the papers we were writing. Initially, it was that we would create the app, collect data at an individual level, and add it, with the Facebook data we have just discussed—
Christian Matheson
Can I just check which app we are talking about at this point?
Aleksandr Kogan
This is the very app we have been dealing with throughout the story.
Christian Matheson
And what is it called?
Aleksandr Kogan
It was called the CPW lab app. CPW is my lab’s name—the Cambridge Prosociality and Well-Being lab. I registered it under my personal account on Facebook—you have to do that as a developer. It was very simple to make. It is just a little bit of code, you get a log-in button, and then we would embed it in the studies we were running as a lab to collect the Facebook data. This was at some point in 2013. The conversation about a potential financial project came later, once I was introduced to SCL. This was spring 2014.
Christian Matheson
So it was SCL that put the idea in your mind about maybe monetising the app.
Aleksandr Kogan
I think that is fair. It is slightly more nuanced, in that the conversation with SCL initially was not at all about Facebook. When I was first introduced to them, they just wanted consulting services and survey design. In our conversations over the next couple of months, we talked about other projects and other datasets. There was this interesting Facebook data that grew out of that.
Christian Matheson
But you had, on a drive somewhere, a large bunch of data.
Aleksandr Kogan
Yes, sir. In my lab.
Christian Matheson
Do you sell the data, or do you sell it as part of a package with your skills?
Aleksandr Kogan
We never sell that data. The data we collect as a lab is never sold.
Christian Matheson
So is it transferred to—
Aleksandr Kogan
No. That would be a very serious breach of ethics. No. 1, that data never left my servers. It was used exclusively with my PhD students. No. 2, when we collected that data, we collected it by telling our participants it was for academic research, so we had no right to use it for anything else.
Christian Matheson
So you would sell the app and somebody else would collect data using the app.
Aleksandr Kogan
Let me return to that in a second, so that it is clear—I know this has been a point of confusion. Thirdly, we did not collect names when we did the academic work. Even if we threw out the ethics and we threw out any concerns we did not even have the ability to do that. When we get to SCL, we collected new data and we used the app to collect new data and we stored it in a different server in a different location and under different terms of service.
Chair
For clarity, I think Chris was asking about the CPW app, for which you are saying you held the data and it was never shared. Obviously, the My Digital Life app that we will talk about later, is separate from this.
Aleksandr Kogan
That is separate. It is not the app that was used for the project with SCL. Only a couple of hundred people ever used the thisisyourdigitallife app and that data was not given to SCL. The app was actually called the GSR app. Once we get to 2014 and we are going to do the project commercially, we changed the name of the GSR app and changed its description. Christopher Wylie provided us with commercial terms of service that we put on the Facebook platform. At that point, once we have made it commercial and separate from the university, we hook it up to a different dataset with a different database and different servers and the data goes there. So it is never in the same place. It can never be mixed up and new data is collected.
Christian Matheson
Is the technical side of the app—the programming—the same?
Aleksandr Kogan
It is the same.
Christian Matheson
You said you were introduced to SCL. One of the aspects of this inquiry that we have encountered is the interchangeability between SCL and Cambridge Analytica. You are clear you were working for SCL.
Aleksandr Kogan
I never had a contract with Cambridge Analytica.
Christian Matheson
Did you ever have any contact with them?
Aleksandr Kogan
That is tricky in the sense that the folks who are involved in one are very much involved in the other. I worked with Alexander Nix and met him a couple of times. He is obviously involved with both. The data science team is very similar on both sides. Alexander Tayler is currently the head of research. I interacted with him. I interacted with all of them in their capacity as SCL Elections, but obviously they are now part of Cambridge Analytica and I am not even entirely clear what the distinction is.
Christian Matheson
Were you clear at the time when you were working with them?
Aleksandr Kogan
I did not hear the name Cambridge Analytica when we started working with them. It did not come up. I only probably heard about it in January 2015, after we had already collected all the data.
Christian Matheson
I want to ask you a question to which the specific answer is probably commercially confidential, but you can give us a ball-park figure. When you are selling an app to a company such as SCL, if I was to do the same thing at the same time, roughly how much might I be negotiating for? I am sure you do not want to give away—
Aleksandr Kogan
I can be fully transparent. To provide a bit of context: the app itself is not very expensive, because you can write this app in three or four days. It is not technically challenging in any way. Facebook explains how to do it, so there is great documentation on this. Do not think of the app itself as something that is valuable, whereas the data that comes in is valuable. For the project that we did, the goal was to recruit about 200,000 people to authorise the app. There are various way to recruit that many people. There has been some confusion about this. Some folks believe there was a honeypot where we made a viral quiz, put it on Facebook, and people clicked—that is not the case. Those things are hard to do and it was not our avenue. Instead, we basically did a market research project where we asked a company that specialises in recruiting people to recruit 200,000 people for us to do a survey about personality and various other items. As part of the survey there was a log in button that they clicked and that authorised the data. The money here, the expensive bit, is paying those users because each person costs about $3 to $4. Obviously, we were recruiting about 200,000 people so we were in that $600,000 to $800,000 range, just to pay the users. That was indeed what Cambridge, or SCL in this case, paid for during that first step of the project. Later on, they came to us and they wanted to get predictions. We use the data to drive predictions about people’s personalities. For that, they paid us £230,000. I think that I testified to that in the written submission.
Christian Matheson
That goes to you, your company or the university?
Aleksandr Kogan
The company.
Christian Matheson
Did you have any day-to-day operational contact with them once you handed over the app?
Aleksandr Kogan
We ran the app. The way the project ran is that they would give me the survey questions, and I would be in charge of running the survey and collecting the data. This arrangement makes sense, because this way I control the data—I receive the data and have the data—and I can do the models on it and then deliver what we were contractually obligated to do.
Christian Matheson
Just to be clear, unlike the academic data, this had names and all the identifying—
Aleksandr Kogan
Yes, sir.
Christian Matheson
Can I ask one more question? Just to take you to the introduction with SCL, you suggested that Chris Wylie had given you the terms of service.
Aleksandr Kogan
Yes, sir.
Christian Matheson
Did you deal directly with Mr Wylie?
Aleksandr Kogan
Yes, sir.
Christian Matheson
Were there any lawyers involved? Did you have a lawyer?
Aleksandr Kogan
I did not. Chris Wylie said he was a data expert and an expert in data law. He guided us. This is well documented, and I can provide the Committee with evidence of that.
Christian Matheson
Thank you.
Ian C. Lucas
When Facebook gave you information, could you identify individuals from that information?
Aleksandr Kogan
Do you mean the dataset they gave during the research project?
Ian C. Lucas
Yes.
Aleksandr Kogan
No.
Ian C. Lucas
Was there other data where you could identify individuals given to you by Facebook?
Aleksandr Kogan
Not through the API; directly by the company—is that correct? You mean not through the API and not through the app, but directly by the company?
Ian C. Lucas
Yes.
Aleksandr Kogan
No.
Ian C. Lucas
So at no stage did Facebook give you information that could identify individuals.
Aleksandr Kogan
No.
Ian C. Lucas
I want to ask you about your relationship with Cambridge University. I note that you are a research associate and university lecturer at the University of Cambridge in the department of psychology. You were appointed to that in 2012.
Aleksandr Kogan
August 2012.
Ian C. Lucas
Do you receive a salary in that post?
Aleksandr Kogan
I do.
Ian C. Lucas
Did you apply for the job?
Aleksandr Kogan
I did.
Ian C. Lucas
Are you associated with an individual college?
Aleksandr Kogan
Yes, I am.
Ian C. Lucas
Which college?
Aleksandr Kogan
Magdalene College.
Ian C. Lucas
Do you teach individual students in tutorials and so on?
Aleksandr Kogan
I do.
Ian C. Lucas
We have been talking also about the various companies that you have set up. There are a number of different companies.
Aleksandr Kogan
There are two.
Ian C. Lucas
What are the names of those companies?
Aleksandr Kogan
There is Global Science Research, which is a UK entity, and then there is Philometrics, which was a US entity.
Ian C. Lucas
When did you set up GSR?
Aleksandr Kogan
GSR was set up some time in the spring of 2014.
Ian C. Lucas
And what about Philometrics?
Aleksandr Kogan
I believe we set up Philometrics in the summer of 2015.
Ian C. Lucas
Does the intellectual property relating to those businesses belong to you, the companies or the university?
Aleksandr Kogan
They belong to me.
Ian C. Lucas
They belong to you as an individual.
Aleksandr Kogan
They belong to the company, but the university makes no claim on the intellectual property that was brought over when I set up the company. My understanding in talking to the university is that since the only thing here that is valuable is code or know-how, the university makes no claim on those things. It encourages its faculty members to start companies to commercialise their work.
Ian C. Lucas
On the Cambridge University website, you are listed both as Aleksandr Kogan and Aleksandr Spectre. Can you explain that?
Aleksandr Kogan
Yes, sir. In 2015, I was married. My wife and I decided that it did not make sense for me to take her name or for her to take my name, so we said, “Let’s choose a new last name.” Since we are both religious and scientists, we thought the idea of light made a lot of sense. We were looking for something relating to light. My father was sadly sick at the time, and one of his surgeons was named Jason Spectre. We thought, “That is a really cool-sounding name”, and it also nails down the theme of light because of spectrum. We decided on Spectre and a derivative of spectrum as a symbol of us going forward as a family.
Ian C. Lucas
You know that Spectre is the evil organisation in the Bond films.
Aleksandr Kogan
It is an unfortunate coincidence.
Ian C. Lucas
Did you know that at the time?
Aleksandr Kogan
I did not.
Ian C. Lucas
You are not a James Bond fan.
Chair
He’s a Spectre fan, though.
Aleksandr Kogan
Exactly.
Ian C. Lucas
Going back to the app you mentioned earlier, we were talking about individual data from Facebook. Can you identify individuals through the API?
Aleksandr Kogan
Yes sir. When you collect data through the API, a user authorises the app, so you can easily identify that user. Depending on what other data you gather, you could also identify their friends.
Ian C. Lucas
You can identify their friends.
Aleksandr Kogan
Yes, sir.
Ian C. Lucas
As well as the people who consented.
Aleksandr Kogan
Correct.
Ian C. Lucas
Has that always been the case?
Aleksandr Kogan
Yes, sir.
Ian C. Lucas
Did that change at all in 2014 with the terms of service from Facebook?
Aleksandr Kogan
What changed was that it was hard to not gather friend data. In terms of the user, you could still always identify them. On Facebook, your name, location, gender and birthday, I believe, are considered your public profile that anybody could search for. I do not have Facebook anymore but if I did, I could search for you, sir. Then I could look at your profile and I would probably be able to see your name and some of this other information, even if we are not friends.
Ian C. Lucas
I think what’s difficult for people—well for me and, I think, others too—to understand is the distinction you draw between knowledge about individuals that is given to you by Facebook and knowledge that you derive through the API. Can you expand on that? What’s the difference?
Aleksandr Kogan
The difference is that when you work with Facebook directly—with the datasets we got—they took millions of people and they averaged them together, so you lose the individuality and the differences. You just have a summary. That is what we are working with. From a summary, you can’t scroll back. For example, if we took the room here and averaged everybody’s age together, and let’s say the average was 45. Taking the number 45, I cannot know your age because you are just part of the group. That’s the problem: there is no way to go back once you’ve summarised. Facebook gave me a summary. With the API you are getting your information, so I know exactly how old you are, if you have given your birthday.
Ian C. Lucas
Essentially, you are deriving the same information: you are deriving personal information through the API?
Aleksandr Kogan
The API gives you the personal information; that is correct. You never get personal information from the data Facebook gave me through our academic link.
Ian C. Lucas
So from my personal perspective, as an individual who is on Facebook, what really is the difference between personal information being handed over by Facebook, and you deriving that information from the API?
Aleksandr Kogan
They’re just not giving us your personal information. They never say: “Ian Lucas, 40 years old”. I never get that; I get: “England; average age is 42”. At that point, there is just no real case for this being personal information because there are no people. It is a summary; it’s a country at this point. Whereas, obviously, me getting information about you—and having a record in my database that says, “Ian Lucas, 42”—is a very different case.
Chair
I’ll ask a couple of questions about work at Cambridge University on this sort of analysis. Other tools have been created at the university that help analyse Facebook likes and preferences to determine information about their data. I believe Cubeyou was one of those. Do you know much about the work of Cubeyou? Were you involved at all?
Aleksandr Kogan
I wasn’t involved with Cubeyou. I know something about the collaborator they worked with, which is the Psychometrics Centre.
Chair
On the website of the Psychometrics Centre, there is also an API called Apply Magic Sauce. Do you know who owns the licence for Apply Magic Sauce?
Aleksandr Kogan
I would believe the Psychometrics Centre.
Chair
Do you know who owns the underlying data. Would that be owned by the Psychometrics Centre as well?
Aleksandr Kogan
I believe the myPersonality dataset was collected by David Stillwell before he joined the university. I believe he would himself own the underlying dataset. The tools I believe were built during his time in the university as part of the Psychometrics Centre, so I believe the centre would in theory—actually, let me back up. My understanding of the university’s position is that the person who writes the code often owns it, unless it is contractually stated that they don’t. I don’t know who wrote that code and what contractual obligations and statements they have. It is possible they had a developer that did assign it to the university and the Psychometrics Centre. Or it is possible they still own the code itself in theory.
Chair
With a project like that, who would benefit? Just the developer himself, or would it be more widely available to researchers at the university?
Aleksandr Kogan
The Psychometrics Centre is a bit of an unusual entity in the university, in that it is involved in both research and commercial activities under the same umbrella. That is more rare. I’d split mine up. I had my university position, which was very separate from my company activity. This was more under the same umbrella. My understanding is that the Psychometrics Centre uses its data for both academic work and commercial work and also just for fun—for people to get results about what their page likes say.
Chair
Do you believe there was a commercial application from APIs like Apply Magic Sauce?
Aleksandr Kogan
I believe they certainly tried to find a commercial application. I am aware they were talking to many commercial entities. I just don’t know where those conversations went.
Chair
Okay. What sort of commercial application would that have been? Would that have been using the data that they gathered as a predictive model for targeting Facebook users?
Aleksandr Kogan
I don’t know if it is quite that specific. My understanding of the way that they envisioned it is a company would bring in its own Facebook data, with its page likes, pass it through this Apply Magic Sauce API, get back predictions about various psychological traits, which then that company can use for whatever purpose its wants, such as market targeting or whatever else.
Brendan O'Hara
Dr Kogan, could you tell the Committee under what circumstances you first met SCL?
Aleksandr Kogan
I was introduced to SCL by a PhD student at Cambridge. I knew the student because I had given him statistics advice on his PhD from time to time.
Brendan O'Hara
What was the purpose of the meeting?
Aleksandr Kogan
My understanding was that the student was part-time consulting for SCL Elections and he had a friend there, named Chris Wylie, who he thought would be great for me to meet and see where the conversations go.
Brendan O'Hara
When was that meeting?
Aleksandr Kogan
I believe it was January or February 2014.
Brendan O'Hara
You met with SCL in January or February 2014.
Aleksandr Kogan
Correct, sir.
Brendan O'Hara
And that meeting was initiated by someone working for SCL.
Aleksandr Kogan
Correct, sir.
Brendan O'Hara
What was the outcome of that meeting?
Aleksandr Kogan
If my recollection is right, we talked about potentially collaborating and me providing them some consulting services. We also talked about them providing me with some of the commercial datasets they had acquired so I could set up a big data science institute.
Brendan O'Hara
Was your decision to set up GSR directly related to your meeting with SCL?
Aleksandr Kogan
GSR was set up entirely to do the project with SCL—I actually relied on their guidance on how to do it appropriately—but it came a little bit later. At this point, we are not talking about commercial entities. Even the Facebook project is still a little bit off from coming up.
Brendan O'Hara
Just for the record, you met with SCL in January 2014 and you set up GSR in May 2014, specifically to work for SCL.
Aleksandr Kogan
Correct, sir.
Brendan O'Hara
In June 2014, the director of analytics at SCL wrote to Cambridge University to say that Dr Kogan first introduced them to the possibility of using online social media data to score and predict human personality traits at a meeting in London, as we have said, in January 2014. Upon agreeing to explore the viability of that idea, Dr Kogan then introduced them to Dr David Stillwell at a meeting in Cambridge in February to discuss the myPersonality project and whether it could be used in partnership with SCL. Why would the director of analytics at SCL write to Cambridge University?
Aleksandr Kogan
Initially, when the project was envisioned, there was going to be a collaboration between myself and the Psychometrics Centre. The idea was that since the Psychometrics Centre had already set up an API and models for personality, we would just use that to generate the predictions. On my side, I would use the app that I had built to collect the data. So, at some point during these negotiations, SCL floated a rough budget of about $2 million. I think it was something that Chris Wylie mentioned during the conversation. I thought, all right, maybe we will give the Psychometrics Centre $500,000 for the modelling aspect of this. As I just mentioned, I was trying to set up a big data science institute—that was my intention. So there was a lot of money—I think a couple of hundred thousand dollars—allocated for server costs and running that. The Psychometrics Centre was fine with that and they liked the idea. As the conversations continued, SCL backed off; they really clawed. They said, “We are not going to pay for anything other than data collection at this point.” That was the $800,000 to pay the participants to answer the survey. They said, “We don’t see why we should be paying $500,000 for the models that the Psychometrics Centre is providing.” At that point, I went to the Psychometrics Centre and said, “They are not willing to pay it. How about $100,000?” I was trying to broker a middle road. The Psychometrics Centre refused. So, SCL instructed me to remove them from the project. At that point, Dr Stillwell and Dr Kosinski, the two people at the Psychometrics Centre I was talking to about this, went and informed Professor Rust of the situation, because he had never been involved with this. Professor Rust wrote a complaint letter to the university saying that “I think Alex is trying to swindle us out of this. He is going to get $1 million. We were promised $500,000 and now there is going to be $100,000.” So, he made some serious accusations. SCL wrote a letter to the university in my defence to state that that was just not true. The Committee has already seen the contract that we ultimately signed with SCL, so you also know that is not true.
Brendan O'Hara
Okay. So the timeline was, you met SCL in January 2014. They then engaged Cambridge University in February 2014. Cambridge University decided not to go ahead with it for whatever reason. Then you set up two months later GSR.
Aleksandr Kogan
That’s not quite right. The problem with that account is that it is not Cambridge University. It is this one lab at the Psychometrics Centre.
Brendan O'Hara
But the director of SCL in his email wrote to Cambridge University. I am trying to work out who he spoke to, why he spoke to Cambridge University, then for nothing to come of it, and then for you to set up within weeks GSR.
Aleksandr Kogan
The plan was already to set up GSR. That was not something new. The intention throughout this process was to set up GSR. Because there was going to be this collaboration between GSR—a private entity that I held—and the Psychometrics Centre, which is a lab in the university, which as I said is a bit unusual in that it works both in a commercial and academic space, so we were going to collaborate. The breakdown occurred because the Psychometrics Centre refused to move from this price tag of $500,000, so SCL kicked them off the project. At that point, the Psychometrics Centre made a complaint to the university, to their legal team, “Hey, we think this is inappropriate and we were promised $500,000.” So SCL wrote to the university legal team to explain that what the Psychometrics Centre had alleged was untrue.
Brendan O'Hara
But that is not quite the account in this email. As I said, the then director of SCL in February wanted to discuss the myPersonality project and whether it could be used in partnership with SCL, having already looked at online social media data to score and predict human personality traits. I am struggling to see how one went from this meeting—what you described as a fairly informal meeting—with an associate of SCL in the January, all the way through February with this great plan that never came to fruition, then you setting up GSR and working with SCL.
Aleksandr Kogan
Let me walk you through that in a bit more detail. I met SCL—Chris Wylie in particular—and we started these conversations. They were trying to pull out many datasets. They were buying up many datasets from data brokers—at least, that is what I was led to believe. I said, “Well, listen, if you are interested in datasets, and I know you are interested in personality”—the interest in personality was already there—“let me introduce you to the Psychometrics Centre, who hold a very big dataset on personality.” That was the myPersonality project. I facilitated a meeting for them to strike that conversation. SCL are at this point interested in acquiring this entire dataset, but David Stillwell, who owns the dataset as we have already discussed, decides not to sell it to them. The grounds were that he had collected that data by telling people it was going to be for academic work, so he felt it would be inappropriate to sell it. All fine, so far. At this point, David, Michal Kosinski—who is also part of the Psychometrics Centre—and I propose an alternative plan. Rather than using the dataset they already had, we would use just the models they had built to make predictions, but we still need data to feed through these models. So I would collect the data, through GSR—that was always the plan—and feed it through these models that the Psychometrics Centre own, which is the Apply Magic Sauce, and generate these predictions about personality. The bit that never came to fruition was this passing the data through their models, because, like I said, they wanted half a million dollars. The only thing that changed was that GSR was not only responsible for the data collection but also for the modelling aspect and the predictions.
Brendan O'Hara
Okay. So you then set up GSR on your own.
Aleksandr Kogan
With one of my associates.
Brendan O'Hara
So there are two of you. You are the two company directors of GSR.
Aleksandr Kogan
Yes, sir.
Brendan O'Hara
And you have this relationship with SCL. Did SCL pay you for your services?
Aleksandr Kogan
At this point they paid only for the data collection. Later on, in January 2015, when we delivered them the second tranche of data, they paid GSR £230,000.
Brendan O'Hara
How much did they pay initially, before the payment of £230,000 in January 2015? How much were you paid by SCL or any of their associates then?
Aleksandr Kogan
In terms of me, or my company?
Brendan O'Hara
The company.
Aleksandr Kogan
We netted zero. The money that went through—the £800,000 I think you are referring to—is the cost for the participants. That is paid almost entirely directly to Qualtrics. That money did not even pass through us. So we would get an invoice from Qualtrics and give it to SCL, and SCL would pay that invoice directly. SCL gave us maybe £10,000 or £20,000, if I recall correctly, to pay for the servers we were running on Amazon Web Services. All that money was entirely for that and we had to match anything we spent with an invoice. So the company at that point was not paid any money at all.
Brendan O'Hara
The only money you made as a company came in 2015, and that was a £230,000 payment from SCL.
Aleksandr Kogan
Correct, sir.
Brendan O'Hara
Was your contract exclusive to SCL or could you approach, or were you approached by, any other data analytics companies?
Aleksandr Kogan
It was not exclusive. The way my contract was set up was that I owned the dataset we collected. We were approached by a company called Eunoia, again in summer 2014. This was a company that Chris Wylie set up immediately after leaving SCL.
Brendan O'Hara
So when you were approached by Christopher Wylie, did you work with Christopher Wylie, did you sell Christopher Wylie or Eunoia any of your data?
Aleksandr Kogan
We did, but it was not for a monetary exchange. The agreement was that we would give them our dataset in exchange for the datasets he purported to have. He did not honour that agreement. He never delivered his datasets, so we moved to cancel that contract by a legal letter in, I believe, March or February of 2015.
Brendan O'Hara
Just bringing this to an end, Christopher Wylie maintained that he refused to work with you when he realised you were not setting up an academic institute as he believed you had originally claimed, but that you were setting up this private company with Alexander Nix. How do you respond to that claim?
Aleksandr Kogan
That is a fabrication.
Brendan O'Hara
In what way?
Aleksandr Kogan
In all ways. Chris Wylie helped me to set up the for-profit entity. It was literally he who said to me, “Please set up a for-profit entity”. Then, after he left SCL, he tried to get me to work with his company. I was not interested, after I had witnessed how he interacted with SCL. He still wanted to have a partnership, so we agreed to do a fair commercial agreement. I gave his company the dataset. The only time this broke down was when he did not honour his side of the agreement and did not deliver the data and we moved to cancel it in February or March 2015. Up to that point, I would say that Mr Wylie was highly enthusiastic about working with me and my commercial entity.
Brendan O'Hara
What was your personal relationship with SCL, and were you paid personally by SCL ever?
Aleksandr Kogan
Never.
Brendan O'Hara
Or any of their associates?
Aleksandr Kogan
Never.
Brendan O'Hara
Okay. What is the status of GSR now?
Aleksandr Kogan
It has been closed.
Brendan O'Hara
When was it closed, and why?
Aleksandr Kogan
I believe it was closed maybe a year ago. I would have to double check the date for you. It was closed because— [Interruption.]
Brendan O'Hara
From my reading the company was dissolved in October 2017. Why was it dissolved?
Aleksandr Kogan
Initially when I set up the company, I was not really thinking about making a company. I was really interested in setting up a non-profit big data institute. I wanted to call it “Mutual Science”. Mr Wylie advised me, “Oh you should also set up a for-profit entity.” I think the reason was that there would be tax benefits. That was Global Science Research. The initial plan was that we would collect the data, I would fulfil my obligations to SCL, and then I would use the data for research. After the project, we decided to give the whole company thing a go. Another associate joined the company who actually had some business experience, because I had none, to try to make a company of it. We tried to develop a couple of products—one was brand reports—but it was just never successful. We used the £230,000 that SCL eventually paid us to invest in trying to build this product. We bought some Twitter data, hired some developers, and tried to sell it, but we never got a single client, so the money ran out. We closed the company just because it failed.
Ian C. Lucas
Why was GSR registered at the 29 Harley Street address?
Aleksandr Kogan
Is that in Cambridge?
Ian C. Lucas
I think it is in London. Do you know what the registered office of GSR was?
Aleksandr Kogan
Initially it was just my apartment in Cambridge. I think later we moved it to an innovation centre in Cambridge, and later Manchester. I am legitimately surprised by this.
Ian C. Lucas
Did you use an agent to set it up?
Aleksandr Kogan
We used Formations House.
Ian C. Lucas
Who is we?
Aleksandr Kogan
Myself and my associate who I made the company with.
Ian C. Lucas
Who was that?
Aleksandr Kogan
I want to keep the name private—again, in confidence.
Ian C. Lucas
Are they on the company’s register?
Aleksandr Kogan
They are.
Ian C. Lucas
Was that Joseph Chancellor?
Aleksandr Kogan
It was.
Ian C. Lucas
So you don’t know anything about the Harley Street address?
Aleksandr Kogan
I am legitimately surprised by that.
Ian C. Lucas
It is used by a lot of shell companies, some of which have been used for money laundering for Russian oligarchs. Did you know that?
Aleksandr Kogan
I did not. I’m unfortunately not a Russian oligarch.
Ian C. Lucas
Have you met any Russian oligarchs?
Aleksandr Kogan
I have not.
Ian C. Lucas
Did SCL ever tell you that they would be using the information that they were receiving in connection with these projects for political campaigning?
Aleksandr Kogan
The official name of the company was SCL Elections, so it would be hard to miss that.
Ian C. Lucas
So you knew it was going to be used for political campaigning.
Aleksandr Kogan
Yes, sir.
Ian C. Lucas
Did they specify where it was going to be used for political campaigning—which countries? [Interruption.]
Aleksandr Kogan
What is that?
Ian C. Lucas
It is the start of business. When it is really exciting, it is a vote, but it is not a vote at the moment. It is just the start of business.
Aleksandr Kogan
Okay. It was going to be in the United States, given that we were collecting data for the United States.
Ian C. Lucas
Was there any discussion about political campaigning in the UK?
Aleksandr Kogan
Later. After we had done this project, there was some discussion about it, but it never went anywhere.
Ian C. Lucas
Was that discussion with Alexander Nix at any time?
Aleksandr Kogan
Yes, it was.
Ian C. Lucas
So Alexander Nix talked to you about political campaigning in the UK?
Aleksandr Kogan
He did.
Chair
Obviously Joseph Chancellor’s name has now come up. Can you explain his role as a director in GSR? Was he your partner? Did you work on everything together?
Aleksandr Kogan
We did. When the company was set up, we were 50:50 partners. Eventually another co-founder came in, so we became equal third partners.
Chair
So the work you were doing for SCL, the work you were doing looking at Facebook data, he was your partner across all of that?
Aleksandr Kogan
Yes, sir.
Chair
He has worked for Facebook since November 2015, hasn’t he?
Aleksandr Kogan
Yes, he has.
Chair
So when Facebook’s response from their deputy general counsel, describing your work as a “scam” and a “fraud”, as data-harvesting—and they singled you out to say, you have lied to us and violated our platform policies, those remarks must apply to Joseph Chancellor as well.
Aleksandr Kogan
If you want to push on the spirit, I would agree. Personally I am very glad that they have not moved on Joe—I think it would be petty, personally.
Chair
But you say you work on it together, so if it applies to you, it applies to him in the same way.
Aleksandr Kogan
I think that would be a fair characterisation.
Chair
So do you not think it is odd that they employ someone that they regard as a scam and a fraud, and who has lied to them and violated their platform policies?
Aleksandr Kogan
Honestly, I don’t think it is odd.
Chair
But I think any normal person would think it is odd. It may not be odd in Facebook-land, but—
Aleksandr Kogan
The reason I don’t think it is 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 realise that the platform has been mined left and right by thousands of others. I was just the unlucky person that ended up somehow linked to the Trump campaign, and we are where we are. I think they realise all this, but PR is PR, and they are trying to manage the crisis. It is convenient to point the finger at a single entity and try to paint the picture that this is a rogue agent.
Chair
But do you not think it is surprising that Facebook employs someone who knew all about this, all about what was being done—that this Facebook data was being gathered to be used by SCL in elections in America—yet did very little to investigate that, even after the story was made public by The Guardian newspaper at the end of 2015, when Joseph Chancellor was already working for the company?
Aleksandr Kogan
Yes, sir. In fairness to Facebook, my perception is that they did a pretty thorough job of investigating the issue. They talked to all the parties involved; they demanded that all the data be deleted. The difficulty here is compliance. I have seen that Facebook has been challenged—as in, why didn’t you audit anyone? The problem with an audit is that the only thing you are going to catch is people who are trying to do the right thing, but missed a few files—because that is possible. You’re trying to delete everything, but maybe you missed a file here and there, because you have don’t have forensic resources. What an audit cannot do is catch people who are trying to be bad actors, because you can always put the data on a hard drive and stick it under your mattress. So in many ways that’s a futile effort. Once the data is off the system, you really are on the honour system in terms of ever trying to put that genie back in the bottle.
Chair
Sandy Parakilas gave evidence to the Committee a few weeks ago. He basically said that once the data is gone, it’s gone. Even if the original dataset is destroyed, the derived value of the data can never be recovered, because it then exists in a new form. Would you agree with that?
Aleksandr Kogan
Facebook asked everybody to delete the data and its derivatives. So I think if you are careful in scope and really make sure that everything that could be considered derivative is deleted, you could delete that. The problem is that you are still on the honour system, because if somebody decides not to—and I think in this case there is now evidence to suggest that somebody decided not to be honourable in terms of all the parties involved, and didn’t delete the data—it is very hard to know that until there is a PR crisis or somebody sees that and they get a tip.
Chair
Yes. But, certainly, the data you acquired was used during the Ted Cruz campaign, wasn’t it?
Aleksandr Kogan
I believe so, yes. SCL admitted as much to me.
Chair
Do you believe that the derived data was also used in the Trump campaign in part as well?
Aleksandr Kogan
I can’t know.
Chair
But do you believe it would be likely?
Aleksandr Kogan
I would actually think it would be unlikely, personally.
Chair
Given that they have already said that they used data from the Cruz campaign in the Trump campaign?
Aleksandr Kogan
The reason I think it is unlikely is that, from talking to others who knew a bit more—because I was not involved at this point; this is much after my time—my understanding is that the Ted Cruz campaign was part unhappy with the product that they thought that SCL delivered to them. Also, SCL at this point is under enormous pressure from Facebook to clean up their act and delete all the data. If I am a company, I don’t know that I would risk the legal liability of somebody like Facebook suing me to keep a dataset that has apparently failed with the last client that it was used for. I don’t know that this was their thinking, but it does not necessarily make sense to me that they would keep this around.
Chair
Can I just ask a couple more questions on this before bringing in colleagues: the political work that SCL was engaged in, which you were supporting—this included the work for Ambassador Bolton’s super PAC, didn’t it?
Aleksandr Kogan
That I don’t know.
Chair
Because there are emails that we published, given to us by Chris Wylie, which reference you working on gathering the data for that work. So you were gathering data on state-specific work for voters in Arkansas, North Carolina and New Hampshire to support Ambassador Bolton’s political activities in those states—the campaigns he was supporting.
Aleksandr Kogan
I would say that the details of who the candidates were was obfuscated for me. What I was told was, “We need these 11 states,” and I was given the surveys that they wanted to ask in the collection. In terms of the details of who they were working with, that was not really ever told to me. Ambassador Bolton is a name I have recently heard. It was not a name that I can ever recall hearing at that time.
Chair
In this email it says—this email was sent by someone on the SCL team—“I also recall your very good idea that Kogan could model 2/3 of what he had of that last round of Bolton issues testing, thereby enabling faster further modelling by the data team to be done first on that sample, then the final model once all data was received. Kogan gets data daily so 2/3 of the data must already be available, even with the delay due to Qualtric’s mistake. Was this done and can the preliminary findings be shared”? This sounds like a fairly dynamic process that you are working on. It is feeding directly into the work for the Bolton super PAC.
Aleksandr Kogan
It is less dynamic than it sounds, because the way the project ran is we did a few waves of data collection and for each wave of data collection SCL would provide me “This is the survey we want to ask”. I wasn’t instructed about why it was being asked, just “This is the survey we want to ask.” Then we would run the study, collect the data, and then we would apply the models we were building to generate predictions about each of these questions. The modelling process does not really even require us to know what the question is. It is just a number. We are just predicting that.
Chair
So what you described earlier on—the survey is being set up; you have got a company going out, getting the bulk audience you need. To some people, when they heard you talk about it earlier, that may sound like a reasonably indiscriminate process, but this would suggest that people are being deliberately targeted in certain states, because that is useful to the commercial project that SCL are working on. So are you directing the surveys team and saying, “We need more people from New Hampshire to respond to this survey, to add into the model”?
Aleksandr Kogan
It wasn’t quite that. We actually did collect data from all 50 states. It was closer to indiscriminate, because the goal was to make predictions on as many people as we could. In this case, with those 11 states, I think we delivered a couple of million people—2 or 3—but the thesis was, rather than a story just focusing on those states, it is better to go in all states, to get the data faster, because, let’s imagine you live in Virginia; you might have friends in West Virginia, and so we would have got that as well. So we got the data, actually, across America. I don’t recall we ever actually, when we instructed Qualtrics to collect the data, there were any restrictions.
Chair
You can take a moment to read the notes, if that would be helpful.
Aleksandr Kogan
Just to make something else clear, the emails that you are talking about I was never involved with. I saw those emails actually for the first time yesterday when it was pointed out that this information has been submitted.
Chair
Yes, but you are clearly interacting with the people that are in the email chain, because you are working for them.
Aleksandr Kogan
Yes—so I am chatting with some of the SCL team. If you can imagine, SCL is in between here, and I guess the Bolton super PAC is on one side instructing them. SCL is instructing me, but there is no communication between these two entities, and quite honestly not even awareness of this side.
Chair
From our point of view, that doesn’t really matter to us, whether you had direct interface with the Bolton team, because this is clearly a project that SCL are undertaking on his behalf, and you are clearly playing a vital role because you are gathering data that has been inputted into the tools that are used to support that campaign. So presumably you must have known or been told that there was a priority around processing data linked to certain states, because the Bolton super PAC project was focused at that time on specific races in specific states, not looking to build up a survey of the whole country—not at that point, anyway.
Aleksandr Kogan
We were told that we were focusing on specific states, but why we were focusing on those states, I have no recollection of ever being informed.
Chair
But you said earlier that you knew it was elections.
Aleksandr Kogan
It was elections. Certainly it was elections in the nine or 11 states that we were contractually obligated to seek out.
Chair
Okay. You did not necessarily know who the candidates were. You just knew that those were the state races you were working on.
Aleksandr Kogan
Exactly.
Chair
Finally, I want to cover some of the evidence that Alexander Nix gave to the Committee. Can I read to you the questions I asked Alexander Nix and the answers he gave, and can you tell me whether you agree with the answers he gave?
Aleksandr Kogan
Sure.