Uses and Abuses of AI in Election Campaigns

Uses and abuses of AI in election campaigns

Alistair Knott

1 / 29 The company’s (somewhat) new angle: Mine the internet for information about voters’ personalities; Use this information to for ‘microtargeting’ of campaign material to voters.

Both those steps involve AI techniques.

Overview

This talk centres on the role of a data analytics company, Cambridge Analytica, in the Brexit and Trump election campaigns.

2 / 29 Both those steps involve AI techniques.

Overview

This talk centres on the role of a data analytics company, Cambridge Analytica, in the Brexit and Trump election campaigns.

The company’s (somewhat) new angle: Mine the internet for information about voters’ personalities; Use this information to for ‘microtargeting’ of campaign material to voters.

2 / 29 Overview

This talk centres on the role of a data analytics company, Cambridge Analytica, in the Brexit and Trump election campaigns.

The company’s (somewhat) new angle: Mine the internet for information about voters’ personalities; Use this information to for ‘microtargeting’ of campaign material to voters.

Both those steps involve AI techniques.

2 / 29 Overview

There will be three parts to the talk:

Models for inferring people’s personality from their online footprint From Cambridge University to Cambridge Analytica Cambridge Analytica’s role in the Brexit and Trump campaigns

3 / 29 The dimensions are: Openness to new experiences Conscientiousness Extraversion Agreeableness Neuroticism

These dimensions mainly emerge from factor analysis on lengthy self-assessment personality questionnaires.

Background: the OCEAN personality measure

The OCEAN measure is a set of personality dimensions used by many social psychologists. It emerged in the 80s and 90s.

4 / 29 These dimensions mainly emerge from factor analysis on lengthy self-assessment personality questionnaires.

Background: the OCEAN personality measure

The OCEAN measure is a set of personality dimensions used by many social psychologists. It emerged in the 80s and 90s.

The dimensions are: Openness to new experiences Conscientiousness Extraversion Agreeableness Neuroticism

4 / 29 Background: the OCEAN personality measure

The OCEAN measure is a set of personality dimensions used by many social psychologists. It emerged in the 80s and 90s.

The dimensions are: Openness to new experiences Conscientiousness Extraversion Agreeableness Neuroticism

These dimensions mainly emerge from factor analysis on lengthy self-assessment personality questionnaires.

4 / 29 Kosinski and colleagues’ 2015 paper

In 2015, Michal Kosinski and colleagues from Cambridge University’s Psychometric Centre published a paper that gathered a lot of attention. [Youyou et al., 2015]

5 / 29 Subjects each completed a personality questionnaire, from which an OCEAN-based measure of personality was computed. The researchers built a model that predicted subjects’ personality from their Facebook ‘Likes’. (With permission.) With 65 Likes, the model was as accurate as a ‘friend’. . . With 300 Likes, it was as accurate as the subject’s spouse.

For a demo, see https://applymagicsauce.com/demo.html .

Kosinski and colleagues’ 2015 paper

Data for the study came from 86,000 Facebook users who used Facebook’s ‘myPersonality’ App.

6 / 29 The researchers built a model that predicted subjects’ personality from their Facebook ‘Likes’. (With permission.) With 65 Likes, the model was as accurate as a ‘friend’. . . With 300 Likes, it was as accurate as the subject’s spouse.

For a demo, see https://applymagicsauce.com/demo.html .

Kosinski and colleagues’ 2015 paper

Data for the study came from 86,000 Facebook users who used Facebook’s ‘myPersonality’ App.

Subjects each completed a personality questionnaire, from which an OCEAN-based measure of personality was computed.

6 / 29 With 65 Likes, the model was as accurate as a ‘friend’. . . With 300 Likes, it was as accurate as the subject’s spouse.

For a demo, see https://applymagicsauce.com/demo.html .

Kosinski and colleagues’ 2015 paper

Data for the study came from 86,000 Facebook users who used Facebook’s ‘myPersonality’ App.

6 / 29 For a demo, see https://applymagicsauce.com/demo.html .

Kosinski and colleagues’ 2015 paper

Data for the study came from 86,000 Facebook users who used Facebook’s ‘myPersonality’ App.

Subjects each completed a personality questionnaire, from which an OCEAN-based measure of personality was computed. The researchers built a model that predicted subjects’ personality from their Facebook ‘Likes’. (With permission.) With 65 Likes, the model was as accurate as a ‘friend’. . . With 300 Likes, it was as accurate as the subject’s spouse.

6 / 29 Kosinski and colleagues’ 2015 paper

Data for the study came from 86,000 Facebook users who used Facebook’s ‘myPersonality’ App.

For a demo, see https://applymagicsauce.com/demo.html .

6 / 29 Kosinski and colleagues’ 2015 paper

7 / 29 After a couple of weeks, Facebook Likes became private by default.

[Graessegger and Krogerus, 2017]

Repercussions of Kosinki’s paper

The day Kosinski published these ﬁndings, he had two phone calls:

one threatening a lawsuit one offering a job.

Both came from Facebook.

8 / 29 Repercussions of Kosinki’s paper

The day Kosinski published these ﬁndings, he had two phone calls:

one threatening a lawsuit one offering a job.

Both came from Facebook.

After a couple of weeks, Facebook Likes became private by default.

[Graessegger and Krogerus, 2017]

8 / 29 Aside: the power of Facebook Likes

The power of Facebook Likes has been appreciated for some time. Around 2010, Google realised that Likes provided a proprietary source of information about web search rankings that could be exploited in a rival search engine. That’s why they developed Google+.

9 / 29 Kogan started a company, Global Science Research (GSR) that reimplemented the model. SCL ran a project that used Mechanical Turk to access a large pool of Facebook proﬁles, hoovering up tens of thousands of individuals’ demographic data—names, locations, birthdays, genders—as well as their Facebook Likes. The same information was captured for each subject’s Facebook friends (on average 350 of them).

[Davies, 2015]

Aleksandr Kogan and GSR

Kosinski was approached by Aleksandr Kogan (from Cambridge’s Psychology department) who wanted to license Kosinski’s model for SCL to use. Kosinski said no.

10 / 29 Aleksandr Kogan and GSR

Kosinski was approached by Aleksandr Kogan (from Cambridge’s Psychology department) who wanted to license Kosinski’s model for SCL to use. Kosinski said no.

Kogan started a company, Global Science Research (GSR) that reimplemented the model. SCL ran a project that used Mechanical Turk to access a large pool of Facebook proﬁles, hoovering up tens of thousands of individuals’ demographic data—names, locations, birthdays, genders—as well as their Facebook Likes. The same information was captured for each subject’s Facebook friends (on average 350 of them).

[Davies, 2015]

10 / 29 Aleksandr Kogan and GSR

In 2015, Kogan’s business partner claimed to ‘own a massive data pool of 40+ million individuals across the US—for each of whom they had generated detailed personality proﬁles’.

[Davies, 2015]

11 / 29 Cambridge Analytica emerged from the alignment of two companies: Strategic Communication Laboratories (SCL): a UK company Renaissance Technologies: a US hedge fund.

Where Cambridge Analytica came from

12 / 29 Where Cambridge Analytica came from

Cambridge Analytica emerged from the alignment of two companies: Strategic Communication Laboratories (SCL): a UK company Renaissance Technologies: a US hedge fund.

12 / 29 Core business is data mining and data analysis on a speciﬁed audience. Based on results, communications will be speciﬁcally targeted to key audience groups to modify behavior in accordance with the goal of the client. [Wikipedia, 2017]

It was founded in the 90s by Nigel Oakes (ex-Saatchi & Saatchi).

SCL

SCL is a behavioural research and strategic communication company.

13 / 29 It was founded in the 90s by Nigel Oakes (ex-Saatchi & Saatchi).

SCL

SCL is a behavioural research and strategic communication company. Core business is data mining and data analysis on a speciﬁed audience. Based on results, communications will be speciﬁcally targeted to key audience groups to modify behavior in accordance with the goal of the client. [Wikipedia, 2017]

13 / 29 SCL

It was founded in the 90s by Nigel Oakes (ex-Saatchi & Saatchi).

13 / 29 SCL

It started in the 90s with commercial work: targeted advertising. It then moved into political communications (particularly in the developing world), and military psyops. It has inﬂuenced elections in Italy, Latvia, Ukraine, Albania, Romania, South Africa, Nigeria, Kenya, Mauritius, India, Indonesia, Thailand, Taiwan, Colombia (. . . ) It has worked on ‘counter-radicalisation’ in Pakistan and Yemen. It has helped Nato counter Russian propaganda in eastern Europe. (‘Recoding the mass consciousness to turn patriotism into collaborationism’)

[Wikipedia, 2017]

[Doward and Gibbs, 2017]

[Cadwalladr, 2017]

14 / 29 It was one of the ﬁrst successful funds to use ‘quantitative trading’. (From 1994 through mid-2014 it averaged a 71.8% annual return.)

The most inﬂuential people in the company are computational linguists, recruited in the 90s from a group at IBM. Peter Brown and Bob Mercer (co-CEOs since 2009) David Magerman

At IBM, they pioneered statistical natural language processing.

Renaissance

Renaissance is a hedge fund, founded in 1982.

15 / 29 The most inﬂuential people in the company are computational linguists, recruited in the 90s from a group at IBM. Peter Brown and Bob Mercer (co-CEOs since 2009) David Magerman

At IBM, they pioneered statistical natural language processing.

Renaissance

Renaissance is a hedge fund, founded in 1982.

It was one of the ﬁrst successful funds to use ‘quantitative trading’. (From 1994 through mid-2014 it averaged a 71.8% annual return.)

15 / 29 At IBM, they pioneered statistical natural language processing.

Renaissance

Renaissance is a hedge fund, founded in 1982.

It was one of the ﬁrst successful funds to use ‘quantitative trading’. (From 1994 through mid-2014 it averaged a 71.8% annual return.)

The most inﬂuential people in the company are computational linguists, recruited in the 90s from a group at IBM. Peter Brown and Bob Mercer (co-CEOs since 2009) David Magerman

15 / 29 Renaissance

Renaissance is a hedge fund, founded in 1982.

It was one of the ﬁrst successful funds to use ‘quantitative trading’. (From 1994 through mid-2014 it averaged a 71.8% annual return.)

The most inﬂuential people in the company are computational linguists, recruited in the 90s from a group at IBM. Peter Brown and Bob Mercer (co-CEOs since 2009) David Magerman

At IBM, they pioneered statistical natural language processing.

15 / 29 Renaissance

16 / 29 Some techniques: Hidden Markov models of language Statistical parsers Statistical machine translation

Statistical NLP

Natural language processing (NLP) is now basically applied statistics. You can tell a lot about a document from the words it contains. Structures of words are more informative than single words. The way words in a document are organised into structures can be characterised statistically.

17 / 29 Statistical NLP

Some techniques: Hidden Markov models of language Statistical parsers Statistical machine translation

17 / 29 Some guesses: Sentiment analysis, information extraction on news items Good models of many languages Well-designed complex statistical models

What quant methods make Renaissance so good?

18 / 29 Sentiment analysis, information extraction on news items Good models of many languages Well-designed complex statistical models

What quant methods make Renaissance so good?

Some guesses:

18 / 29 Good models of many languages Well-designed complex statistical models

What quant methods make Renaissance so good?

Some guesses: Sentiment analysis, information extraction on news items

18 / 29 Well-designed complex statistical models

What quant methods make Renaissance so good?

Some guesses: Sentiment analysis, information extraction on news items Good models of many languages

18 / 29 What quant methods make Renaissance so good?

Some guesses: Sentiment analysis, information extraction on news items Good models of many languages Well-designed complex statistical models

18 / 29 Cambridge Analytica is an offshoot of SCL, created in 2013 to participate in US politics.

Bob Mercer was one of the main investors in the company.

Politically, Mercer is ultra-conservative. He’s one of the main funders of the far-right Breitbart website. Steve Bannon (until recently exec chairman of Breitbart) was a VP on the board of Cambridge Analytica. Its CEO is Alexander Nix.

SCL + Bob Mercer: Cambridge Analytica

19 / 29 Bob Mercer was one of the main investors in the company.

SCL + Bob Mercer: Cambridge Analytica

Cambridge Analytica is an offshoot of SCL, created in 2013 to participate in US politics.

19 / 29 Politically, Mercer is ultra-conservative. He’s one of the main funders of the far-right Breitbart website. Steve Bannon (until recently exec chairman of Breitbart) was a VP on the board of Cambridge Analytica. Its CEO is Alexander Nix.

SCL + Bob Mercer: Cambridge Analytica

Cambridge Analytica is an offshoot of SCL, created in 2013 to participate in US politics.

Bob Mercer was one of the main investors in the company.

19 / 29 He’s one of the main funders of the far-right Breitbart website. Steve Bannon (until recently exec chairman of Breitbart) was a VP on the board of Cambridge Analytica. Its CEO is Alexander Nix.

SCL + Bob Mercer: Cambridge Analytica

Cambridge Analytica is an offshoot of SCL, created in 2013 to participate in US politics.

Bob Mercer was one of the main investors in the company.

Politically, Mercer is ultra-conservative.

19 / 29 Steve Bannon (until recently exec chairman of Breitbart) was a VP on the board of Cambridge Analytica. Its CEO is Alexander Nix.

SCL + Bob Mercer: Cambridge Analytica

Cambridge Analytica is an offshoot of SCL, created in 2013 to participate in US politics.

Bob Mercer was one of the main investors in the company.

Politically, Mercer is ultra-conservative. He’s one of the main funders of the far-right Breitbart website.

19 / 29 Its CEO is Alexander Nix.

SCL + Bob Mercer: Cambridge Analytica

Cambridge Analytica is an offshoot of SCL, created in 2013 to participate in US politics.

Bob Mercer was one of the main investors in the company.

19 / 29 SCL + Bob Mercer: Cambridge Analytica

Cambridge Analytica is an offshoot of SCL, created in 2013 to participate in US politics.

Bob Mercer was one of the main investors in the company.

19 / 29 SCL + Bob Mercer: Cambridge Analytica

20 / 29 From a presentation given by Nix:

The company ‘buys personal data from a range of different sources, like land registries, automotive data, shopping data, bonus cards, club memberships, what magazines you read, what churches you attend.’ [Nix displays the logos of globally active data brokers like Acxiom and Experian—in the US, almost all personal data is for sale. For example, if you want to know where Jewish women live, you can simply buy this information, phone numbers included.]

[Graessegger and Krogerus, 2017]

Cambridge Analytica’s database

Cambridge Analytica took Aleksandr Kogan (SCL)’s Mechanical Turk Facebook data and ampliﬁed it from many sources.

21 / 29 [Nix displays the logos of globally active data brokers like Acxiom and Experian—in the US, almost all personal data is for sale. For example, if you want to know where Jewish women live, you can simply buy this information, phone numbers included.]

[Graessegger and Krogerus, 2017]

Cambridge Analytica’s database

Cambridge Analytica took Aleksandr Kogan (SCL)’s Mechanical Turk Facebook data and ampliﬁed it from many sources.

From a presentation given by Nix:

21 / 29 Cambridge Analytica’s database

Cambridge Analytica took Aleksandr Kogan (SCL)’s Mechanical Turk Facebook data and ampliﬁed it from many sources.

From a presentation given by Nix:

[Graessegger and Krogerus, 2017]

21 / 29 ‘Nix likes to boast that Analytica’s personality model has allowed it to create a personality proﬁle for every adult in the U.S—220 million of them, each with up to 5,000 data points.’

[Anderson and Horvath, 2017]

Cambridge Analytica’s database

Now Cambridge Analytica aggregates this data with the electoral rolls of the Republican party and online data and calculates an [OCEAN] personality proﬁle. Digital footprints suddenly become real people with fears, needs, interests, and residential addresses.

[Graessegger and Krogerus, 2017]

22 / 29 Cambridge Analytica’s database

[Graessegger and Krogerus, 2017]

‘Nix likes to boast that Analytica’s personality model has allowed it to create a personality proﬁle for every adult in the U.S—220 million of them, each with up to 5,000 data points.’

[Anderson and Horvath, 2017]

22 / 29 In November 2015, Leave.EU (founded by Arron Banks, supported by Nigel Farage’s UKIP) announced it had commissioned Cambridge Analytica to support its online campaign using OCEAN-based microtargeting. (No money changed hands.)

‘Everywhere he went, Kosinski had to explain that he had nothing to do with this company.’

Meanwhile, Aleksandr Kogan moved to Singapore, married, and changed his name to ‘Aleksandr Spectre’.

[Graessegger and Krogerus, 2017]

Analytica in the Brexit campaign

23 / 29 ‘Everywhere he went, Kosinski had to explain that he had nothing to do with this company.’

Meanwhile, Aleksandr Kogan moved to Singapore, married, and changed his name to ‘Aleksandr Spectre’.

[Graessegger and Krogerus, 2017]

Analytica in the Brexit campaign

In November 2015, Leave.EU (founded by Arron Banks, supported by Nigel Farage’s UKIP) announced it had commissioned Cambridge Analytica to support its online campaign using OCEAN-based microtargeting. (No money changed hands.)

23 / 29 Meanwhile, Aleksandr Kogan moved to Singapore, married, and changed his name to ‘Aleksandr Spectre’.

[Graessegger and Krogerus, 2017]

Analytica in the Brexit campaign

‘Everywhere he went, Kosinski had to explain that he had nothing to do with this company.’

23 / 29 Analytica in the Brexit campaign

‘Everywhere he went, Kosinski had to explain that he had nothing to do with this company.’

Meanwhile, Aleksandr Kogan moved to Singapore, married, and changed his name to ‘Aleksandr Spectre’.

[Graessegger and Krogerus, 2017]

23 / 29 In 2015, Analytica became known as the data analysis company for Ted Cruz’ presidential campaign. (Prior to this, several other US political campaigns.)

In 2016, Analytica started to work for Trump’s campaign.

Trump talks a lot about ‘another Brexit’ during his campaign. Jared Kushner and Jason Miller [on his campaign team] told Andy Wigmore of Leave.EU (in 2015, prior to Analytica) that ‘the holy grail is Artiﬁcial Intelligence’.

[Cadwalladr, 2017]

Analytica in the Trump campaign

24 / 29 In 2016, Analytica started to work for Trump’s campaign.

[Cadwalladr, 2017]

Analytica in the Trump campaign

In 2015, Analytica became known as the data analysis company for Ted Cruz’ presidential campaign. (Prior to this, several other US political campaigns.)

24 / 29 Trump talks a lot about ‘another Brexit’ during his campaign. Jared Kushner and Jason Miller [on his campaign team] told Andy Wigmore of Leave.EU (in 2015, prior to Analytica) that ‘the holy grail is Artiﬁcial Intelligence’.

[Cadwalladr, 2017]

Analytica in the Trump campaign

In 2015, Analytica became known as the data analysis company for Ted Cruz’ presidential campaign. (Prior to this, several other US political campaigns.)

In 2016, Analytica started to work for Trump’s campaign.

24 / 29 Jared Kushner and Jason Miller [on his campaign team] told Andy Wigmore of Leave.EU (in 2015, prior to Analytica) that ‘the holy grail is Artiﬁcial Intelligence’.

[Cadwalladr, 2017]

Analytica in the Trump campaign

In 2015, Analytica became known as the data analysis company for Ted Cruz’ presidential campaign. (Prior to this, several other US political campaigns.)

In 2016, Analytica started to work for Trump’s campaign.

Trump talks a lot about ‘another Brexit’ during his campaign.

24 / 29 Analytica in the Trump campaign

In 2015, Analytica became known as the data analysis company for Ted Cruz’ presidential campaign. (Prior to this, several other US political campaigns.)

In 2016, Analytica started to work for Trump’s campaign.

[Cadwalladr, 2017]

24 / 29 Some caveats:

All election data analysis companies like to claim they have the magic sauce when their politician wins.

People speaking on behalf of Cambridge Analytica are salesmen.

Hedge fund computer programs are ‘the blackest of black boxes’.

How was AI used in the Brexit and Trump campaigns?

25 / 29 All election data analysis companies like to claim they have the magic sauce when their politician wins.

People speaking on behalf of Cambridge Analytica are salesmen.

Hedge fund computer programs are ‘the blackest of black boxes’.

How was AI used in the Brexit and Trump campaigns?

Some caveats:

25 / 29 People speaking on behalf of Cambridge Analytica are salesmen.

Hedge fund computer programs are ‘the blackest of black boxes’.

How was AI used in the Brexit and Trump campaigns?

Some caveats:

All election data analysis companies like to claim they have the magic sauce when their politician wins.

25 / 29 Hedge fund computer programs are ‘the blackest of black boxes’.

How was AI used in the Brexit and Trump campaigns?

Some caveats:

All election data analysis companies like to claim they have the magic sauce when their politician wins.

People speaking on behalf of Cambridge Analytica are salesmen.

25 / 29 How was AI used in the Brexit and Trump campaigns?

Some caveats:

All election data analysis companies like to claim they have the magic sauce when their politician wins.

People speaking on behalf of Cambridge Analytica are salesmen.

Hedge fund computer programs are ‘the blackest of black boxes’.

25 / 29 1. Campaign actions targeted to voters’ personality types.

How was AI used in the Brexit and Trump campaigns?

26 / 29 How was AI used in the Brexit and Trump campaigns?

1. Campaign actions targeted to voters’ personality types.

26 / 29 How was AI used in the Brexit and Trump campaigns?

1. Campaign actions targeted to voters’ personality types.

In a presentation, Nix showed how psychographically categorized voters can be differently addressed, based on the example of gun rights. ‘For a highly neurotic and conscientious audience the threat of a burglary—and the insurance policy of a gun.’ [An image shows the hand of an intruder smashing a window.] ‘Conversely, for a closed and agreeable audience: people who care about tradition, and habits, and family.’ [An image shows a man and a child standing in a ﬁeld at sunset, both holding guns, shooting ducks.]

[Graessegger and Krogerus, 2017]

26 / 29 How was AI used in the Brexit and Trump campaigns?

1. Campaign actions targeted to voters’ personality types.

On the day of the third debate between Trump and Clinton, Trump’s team tested 175,000 different ad variations for his arguments, in order to ﬁnd the right versions above all via Facebook. The messages differed for the most part only in microscopic details, in order to target the recipients in the optimal psychological way: different headings, colors, captions, with a photo or video. This ﬁne-tuning reaches all the way down to the smallest groups, Nix explained in an interview. ‘We can address villages or apartment blocks in a targeted way. Even individuals.’

[Graessegger and Krogerus, 2017]

26 / 29 How was AI used in the Brexit and Trump campaigns?

1. Campaign actions targeted to voters’ personality types.

‘Trump’s striking inconsistencies, his much-criticized ﬁckleness, and the resulting array of contradictory messages, suddenly turned out to be his great asset: a different message for every voter.’

[Graessegger and Krogerus, 2017]

26 / 29 How was AI used in the Brexit and Trump campaigns?

1. Campaign actions targeted to voters’ personality types.

[Graessegger and Krogerus, 2017]

(Analytica’s targeting was not always microscopic. For many purposes it divided the US population into just 32 personality types.)

[Graessegger and Krogerus, 2017]

26 / 29 How was AI used in the Brexit and Trump campaigns?

1. Campaign actions targeted to voters’ personality types.

Facebook was the key to the whole Brexit campaign, said Andy Wigmore [Leave.EU’s communications director].

A Facebook Like was their most ‘potent weapon’. ‘Because using artiﬁcial intelligence, as we did, tells you all sorts of things about that individual and how to convince them with what sort of advert.’

[Cadwalladr, 2017]

26 / 29 How was AI used in the Brexit and Trump campaigns?

2. Clever methods for delivering targeted messages to voters.

27 / 29 How was AI used in the Brexit and Trump campaigns?

2. Clever methods for delivering targeted messages to voters.

(i) ‘Dark posts’: sponsored news-feed-style ads in Facebook timelines that can only be seen by users with speciﬁc proﬁles.

[Graessegger and Krogerus, 2017]

27 / 29 How was AI used in the Brexit and Trump campaigns?

2. Clever methods for delivering targeted messages to voters.

(ii) Canvassing apps. From July 2016, Trump’s canvassers were provided with an app with which they could identify the political views and personality types of the inhabitants of a house. It was the same app provider used by Brexit campaigners.

[Graessegger and Krogerus, 2017]

27 / 29 How was AI used in the Brexit and Trump campaigns?

2. Clever methods for delivering targeted messages to voters.

(iii) Twitter bots. (Useful resource: http://politicalbots.org/)

27 / 29 How was AI used in the Brexit and Trump campaigns?

2. Clever methods for delivering targeted messages to voters.

(iii) Twitter bots. (Useful resource: http://politicalbots.org/)

In the Brexit campaign: Of 1.5 million tweets with hashtags related to the referendum sampled between 5 June and 12 June, 54 per cent were pro-Leave and 20 per cent were pro-Remain. But a third—half a million tweets—were generated by just 1 per cent of the 300,000 sampled accounts. This level of activity suggests that many of these are scripted bots. Throughout the period, the Brexit bots were much more active, tweeting more than three times as often as the Remain bots.

[Howard and Kollanyi, 2016]

27 / 29 How was AI used in the Brexit and Trump campaigns?

2. Clever methods for delivering targeted messages to voters.

(iii) Twitter bots. (Useful resource: http://politicalbots.org/)

In the Trump campaign: [During the first presidential debate,] we find that (1) Twitter traffic on pro-Trump hashtags was roughly double that of the pro-Clinton hashtags, (2) about one third of the pro-Trump twitter traffic was driven by bots and highly automated accounts, compared to one fifth of the pro-Clinton Twitter traffic.

[Kollanyi et al., 2016a]

27 / 29 How was AI used in the Brexit and Trump campaigns?

2. Clever methods for delivering targeted messages to voters.

(iii) Twitter bots. (Useful resource: http://politicalbots.org/)

In the Trump campaign: Not only did the pace of pro-Trump activity increase over time, but the gap between pro-Trump and pro-Clinton activity widened from 4:1 during the ﬁrst debate to 5:1 by election day. The use of automated accounts was deliberate and strategic throughout the election, most clearly with pro-Trump campaigners and programmers who carefully adjusted the timing of content production during the debates, strategically colonized pro-Clinton hashtags, and then disabled activities after Election Day.

[Kollanyi et al., 2016b]

27 / 29 David Magerman (one of Renaissance’s IBM computational linguists) has just been sacked for publicly criticising Mercer’s politics.

[Durden, 2017]

Cambridge Analytica is not willing to comment on alleged ongoing talks with Theresa May.

[Graessegger and Krogerus, 2017]

Michal Kosinki: ‘This is not my fault. I did not build the bomb. I only showed that it exists.’

[Graessegger and Krogerus, 2017]

The fallout

28 / 29 Cambridge Analytica is not willing to comment on alleged ongoing talks with Theresa May.

[Graessegger and Krogerus, 2017]

Michal Kosinki: ‘This is not my fault. I did not build the bomb. I only showed that it exists.’

[Graessegger and Krogerus, 2017]

The fallout

David Magerman (one of Renaissance’s IBM computational linguists) has just been sacked for publicly criticising Mercer’s politics.

[Durden, 2017]

28 / 29 Michal Kosinki: ‘This is not my fault. I did not build the bomb. I only showed that it exists.’

[Graessegger and Krogerus, 2017]

The fallout

David Magerman (one of Renaissance’s IBM computational linguists) has just been sacked for publicly criticising Mercer’s politics.

[Durden, 2017]

Cambridge Analytica is not willing to comment on alleged ongoing talks with Theresa May.

[Graessegger and Krogerus, 2017]

28 / 29 The fallout

David Magerman (one of Renaissance’s IBM computational linguists) has just been sacked for publicly criticising Mercer’s politics.

[Durden, 2017]

Cambridge Analytica is not willing to comment on alleged ongoing talks with Theresa May.

[Graessegger and Krogerus, 2017]

Michal Kosinki: ‘This is not my fault. I did not build the bomb. I only showed that it exists.’

[Graessegger and Krogerus, 2017]

28 / 29 References

Anderson, B. and Horvath, B. (2017). The rise of the weaponized AI propaganda machine. https://www.scout.ai/story/the-rise-of-the-weaponized-ai-propaganda-machine. Undated. Accessed March 8 2017. Cadwalladr, C. (2017). Robert Mercer: the big data billionaire waging war on mainstream media. https://www.theguardian.com/politics/2017/feb/26/ robert-mercer-breitbart-war-on-media-steve- bannon-donald-trump-nigel-farage. Feb 27 2017. Accessed March 8 2017. Davies, H. (2015). Ted Cruz using firm that harvested data on millions of unwitting facebook users. https://www.theguardian.com/us-news/2015/dec/11/ senator-ted-cruz-president-campaign-facebook-user-data. Dec 11 2015. Accessed March 8 2017. Doward, J. and Gibbs, A. (2017). Ted cruz using firm that harvested data on millions of unwitting Facebook users. https://www.theguardian.com/politics/2017/mar/04/nigel-oakes-cambridge-analytica- what-role-brexit-trump. Mar 4 2017. Accessed March 9 2017. Durden, T. (2017). Mutiny breaks out inside the world’s most successful hedge fund over Donald Trump. http://www.zerohedge.com/news/2017-02-23/mutiny-breaks-out-inside-worlds-most-successful- hedge-fund-over-donald-trump. Feb 24th 2017. Accessed March 8 2017. Graessegger, H. and Krogerus, M. (2017). The data that turned the world upside down. https://motherboard.vice.com/en_us/article/ how-our-likes-helped-trump-win. Jan 29 2017. Accessed March 8 2017. Howard, P. and Kollanyi, B. (2016). Bots, strongerin, and brexit: Computational propaganda during the uk-eu referendum. https://arxiv.org/abs/1606.06356. Accessed March 9 2017. Kollanyi, B., Howard, P., and Wooley, S. (2016a). Bots and automation over Twitter during the first U.S. presidential debate. http://politicalbots.org/wp-content/uploads/2016/ 10/Data-Memo-First-Presidential-Debate.pdf. Oct 14 2016. Accessed March 8 2017. Kollanyi, B., Howard, P., and Wooley, S. (2016b). Bots and automation over Twitter during the US election. http://politicalbots.org/wp-content/uploads/2016/11/Data-Memo-US-Election.pdf. Nov 17 2016. Accessed March 8 2017. Wikipedia (2017). Scl. https://en.wikipedia.org/wiki/SCL_Group. Accessed March 8 2017. Youyou, W., Kosinski, M., and Stillwell, D. (2015). Computer-based personality judgments are more accurate than those made by humans. PNAS, 112(4):1036–1040.

29 / 29