Is it possible to modify individual human brains via social media?
Brexit and Trump as AI’s Fat Man and Little Boy
While I was preparing this piece, Azeem Azhar linked to the following piece on the excellent Exponential View:
There’s a new automated propaganda machine driving global politics. How it works and what it will mean for the future…medium.com
Is this the sunrise or sunset?
During this week’s press conference with Canadian Prime Minister Justin Trudeau, Donald Trump was asked about border security. In the middle of his response, he says the following:
I’m just doing what I said I would do when we won by a very, very large electoral college vote.
And I knew that was going to happen. I knew this is what people were wanting.
“I knew that was going to happen”
On 26 January 2017, Gillian Tett from the Financial Times published the piece, “Donald Trump’s campaign shifted the odds by making big data personal.”
Why did Donald Trump win? In recent months we have been offered both sociological and economic answers.
But here is another explanation to ponder: computer science.
Tett goes on to reference a British company called Cambridge Analytica. The company started at the University of Cambridge as an experimental project mining personality traits from social network data.
The America-based Robert Mercer family purchased a majority shareholding in the company. Mercer is an ex-IBM computer scientist who worked heavily in the area of speech recognition – precursor to today’s efforts in artificial intelligence.
Cambridge Analytica sources data from social networks, creates personality profiles, and then executes behavioural micro-targetting progammes. The company uses those microtargeting programmes to “understand what people care about, why they behave the way they do, and what really drives their decision making.”
Prior to the Mercer acquisition, White House chief of staff Steve Bannon sat on the board of Cambridge Analytica. Subsequent to the Mercer acquisition, the company’s services were employed by both the Leave campaign in the EU referendum vote, and the Republican campaign of Donald Trump.
How Cambridge Analytica makes it happen
The Cambridge Analytica connection to both the Brexit Leave campaign and Trump was first reported in November 2016. Brief investigative reports established that two commonalities between both the Leave and Trump campaigns:
Each campaign looked destined to fail. Professional analysts and bookmakers alike missed the ultimate outcomes.
Each campaign worked with Cambridge Analytica and used behavioural microtargeting programmes.
Journalists and the political establishment quickly refuted the power of Cambridge Analytica’s admires and critics — “big snake oil”.
Leonid Bershidsky from Bloomberg in December 2016:
I watched Trump improve his campaigning technique and crisscross the country with a message that wasn’t microtargeted but that resonated powerfully with many people I met.
I also watched Clinton fail to connect, even though her campaign made use of data analytics before Trump got into that game. There is no scientific cure for this kind of thing, and no victory elixir.
Undaunted, technology reporters continued to pick apart Cambridge Analytica’s involvement in the election. They looked beyond the history and capabilities of the technology itself, and delved into how it was used.
In January, the Motherboard tech column of Vice magazine revealed how campaigns use targeted messages:
On the day of the third presidential 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.
“We can address villages or apartment blocks in a targeted way. Even individuals.”
Both Brexit and Trump’s victories were striking for a similar reason. Each campaign favoured to lose. Wins delivered by exactly enough votes in key districts to shift the overall totals of the races.
Welcome to the Moneyball Election. Donald Trump both lost the popular vote, and won the electoral college by one of the lowest margins. Yet he now sits in the big chair.
Trump’s detractors continually cite his loss of the popular vote as an indication he lacks a mandate to govern at best, and sits in power illegimitately at worst. Trump counters he won an electoral vote landslide, and that gives him a mandate.
Hacking the Electoral College system
The Electoral College system began as a firebreak between the people and a potential tyrant. In the event the populous could be whipped into a frenzy by an authoritarian candidate, the Electoral College is meant to reject that candidate in favour of a different candidate. The spirit of the Electoral College rejecting a candidate does not hold up the integrity of American democracy, rather it exists in the event it fails. Designed to choose between the lesser evil of usurping the will of the people versus putting a leader into office who is potentially dangerous to the republic.
The operation of the Electoral College has a very important secondary function. It reduces the mathematical complexity of modelling the votes of the presidential election.
Rather than modelling the behaviours of 128,000,000 individual voters, one first breaks the problem into 538 individual electoral college votes.
Next, discount the known likely votes either way — these votes have a high statistical probability of going to theDemocrat, and these votes have a high statistical probability of going to the Republican.
That leaves a known number of votes in “swing states,” but the reality is those are not swing states but swing districts. Demarcated in geographic information system databases, it’s trivial to create a data map of combinations required to win individual districts.
Once those districts are modelled, the next step is to source the population of those districts. That creates a total number of voters.
Once that number is known, the process repeats. Establish the temperament of individuals to be Democrat, Republican, or swing. Once those individuals have been tagged, targeted messaging begins.
From a computer science perspective, this is a relatively simplistic problem once the data map has been set up, and the individual entities tagged. Solve the equation that produces the greatest likelihood of winning the most Electoral College votes by targeting the underlying districts. Target the swing voters in those districts with ads designed to shift their opinion toward the Republican candidate.
Through iterative processes, the machine delivers hundreds of thousands (or millions) of tailored data streams to social media users. Primary polling and other informal polls mine the views of individuals in known geographies. Together, a data-driven feedback loop targeted to win the requisite number of Electoral College votes.
As a logician would say, “Complex, but not difficult.”
It happens because Steve Bannon makes it happen
The groundwork for Trump’s Presidential victory may have been laid on 1 March 2012. Brietbart news founder Andrew Brietbart collapsed in Brentwood, California. The following day he was pronounced dead of heart failure at 43 years of age.
Brietbart started Brietbart.com in 2007 as a conservative news network that aggregated stories from newswire services. It quickly became one of the leading feeders to the Drudge Report. From there, it blossomed into a global conservative digital news channel. Monitoring every click on every story and using that data to power an advertising model.
When Andrew Brietbart passed away, leadership at the news service was taken over by former Goldman Sachs banker Steve Bannon.
Ever the opportunist, Bannon likely saw beyond the fundamental purpose of Andrew Brietbart’s vision of a “pro-freedom, pro-Isreal news website”. Bannon may well have seen a technology platform with a strong brand name and massive global research. A platform that did three things:
- sourced content
- delivered content to audiences
- measured audience engagement with content
Rather than a news site, Bannon may well have seen a feedback loop that distributed words and images, and then measured individual people’s engagement with those words and images.
Building the meta-feedback loop
Brietbart.com’s reach would only extend to its direct and indirect consumers. Once Bannon saw the possibility not only of the data feedback loop to report content, he would have also seen the power of the loop to shape content.
Rather than using content to drive advertising metrics, could it be used to shape opinions reported elsewhere? Could the content pushed out on Brietbart be used to shape, for example, voting behaviour?
Now, I do not know Steve Bannon and I am not aware of his thoughts. Rather, I’m connecting dots between the day he took over at Brietbart in 2012, and the day he entered the White House as Donald Trump’s chief of staff in 2017.
From time at Goldman Sachs, Bannon would have been exposed the some of the most advanced financial trading technology in the world. Systems sourcing data from around the world to predict asset values and stock prices. One key component of those systems is a rudimentary level of ecosystem awareness. How trading one stock has an effect on another stock. How the Goldman Sachs trading system itself modifies the largest system of interconnected financial markets.
Take that same technology an apply it to people. Change the feedback input from ad revenue to shifts in voting patterns. Target the system to source content and produce shifts in wins for Republican candidates.
In addition to shifting voting patterns, such a system would also report in detail specifically what content shifted voting patterns.
Yet the Brietbart content delivery system has two distinct limitations:
First, capabilities to track individual people are at best rudimentary.
Secondly, content is delivered through specific downstream channels, where the consumer effectively opt-in by clicking on a story. Consumers likely reaching Brietbart through a partner brand. To truly create a meta-feedback system capable of creating shifts in opinion, such a system needs to reach beyond Brietbart’s existing base.
Unwitting collaborators — Facebook and Twitter
Fortunately for Steve Bannon and Donald Trump, they found unwitting accomplices in Jack Dorsey and Mark Zuckerberg.
In August 2016, Zuckerberg said about Facebook: “We are a tech company, not a media company.”
Then there’s the technology version of the Sad Keanu meme. The world’s technological elite sitting in Trump Tower. Political eunaches in the ascending king’s court.
Barak Obama visited Silicon Valley. Campus by campus, company by company. He wanted to see the people building the future. Now, they sit together in Trump Tower understanding the true nature of the California bubble. They run public companies. And for grand commitments to diversity and transparency, ultimately they lead the companies of their own creation at the behest of their shareholders. Shareholders who answer to the Securities and Exchange Commission — a government department now led by the man sitting before them in Trump Tower.
Both Facebook and Twitter (amongst others) pitch themselves as platforms that do not generate content, rather they distribute it between users. Each company has taken great pains to distance itself from the idea the companies have their own views and shape content. Facebook and Twitter are simply highly-advanced distribution networks. They do not generate primary content, nor do the platforms (actively) censor content.
Distribution networks that today reach billions of people.
While Dorsey and Zuckerberg represent the idealistic generation of Millennial businessmen reshaping the world through Silicon Valley’s higher values. The same transparency they distilled into the platforms also mean those platforms accurately report information on billions of individual people to anyone using the service, to the businesses paying for access to that data.
While Dorsey and Zuckerberg are not responsible for how customers use the advanced data analytics of their services, it is unlikely that the potential engineering of both Brexit and the American election could have occured in the matter it may well have done without Dorsey and Zuckerberg’s attachment to democratic communication and radical transparency.
Both Facebook and Twitter played huge roles in the presidential campaign, and now Twitter has become the primary distribution channel for Trump’s thoughts. One of Trump’s primary reasons to discredit traditional media and news channels is to drive the public toward his own new channel — the @realdonaldtrump on Twitter.
Since the election, Donald Trump’s reach on Twitter expands 10,000 people a day – doubling every six months.
The more he can break the attachment of the electorate to traditional media brands, the greater his own channel rises in power. No longer dependant upon the fourth estate to carry his message to the people – complete with analysis, context, and opinion.
He speaks to 25M directly, whenever he desires.
More importantly, he possesses full analytics on what people do with that message. Whether the like, share, or respond.
There’s a specific reason he continues to use his private Twitter account over an official government account.
Access to rich analytics without public transparency is a very good reason.
Welcome to Lean Presidency
Most importantly, the overall messages and directions are being iterated.
Why is Donald Trump’s Twitter account largely one-way? He (prolifically) posts, generates hundreds of thousands of responses, and then claims to be isolated from the media.
Those Twitter accounts are not only primary channels for communicating outside of the media, but also for data collection.
It is not difficult to imagine the data processing loops sitting behind Donald Trump’s Twitter account. He writes a message and distributes it.
Silently, in data centres around the world, systems run machine learning processes on the content of the tweet, and then watch as it diseminates. The data shows who supports the message, who replies against it. Where it goes. Who shares it.
If one were to entertain the data map at its most thorough possibility, it indicates what language can Trump use that maximises reach amongst both supporters and detractors.
Growth hack the city on the hill
Every time he tweets and a flurry of responses fire back, threatening everything from civil disobedience to impeachment, he may well be effectively crowdsourcing a strategy against his opponents. And they’re freely giving it to him.
Reading every line of the constitution. Threatening him with every potential scenario of legal and political action. Tweeting their hates. Copying their friends.
Creating a deep data lake of how individuals around the world feel about his messages, catalogued in real time by unique individual identifier.
Everyone tweeting at Donald Trump is personally identifying themselves to his data scientists. Mapping where ideas start, and how they transmit.
To be clear, Donald Trump doesn’t seem to care about the individual responses, rather he seems most interested in generating responses.
Generating data for machine learners on everyone who interacts with a Trump tweet online.
The potential effects become quite chilling if this regime does become authoritarian. In reaction to the Trump purge of the United States civil service, a raft of shadow — or rogue — accounts have popped up.
One in particular claims to be RoguePOTUSStaff — an operative or operatives inside the White House, serving the President himself. Claiming to be the resistance. Yet, there is no proof of identity.
A theory emerges from Louise Mensch that RoguePOTUS is a baited account run by Steve Bannon. If that’s the case, he’s catalogued 900,000 people so far who are de facto against his administration.
Something wicked this way comes
Two of the most important concepts in both politics and civil life are the following:
Silicon Valley focuses on big questions of generalised artificial intelligence. How theoretical future societies handle AI.
Right now, we may well be witnessing real applications of machine learning which potentially deny millions of people of lawful rights.
It may be a stretch to speak of “weaponised machine learning”. It is obvious that something is going on behind the scenes with regard to Donald Trump’s presidency and data.
He “knew he was going to win”. He possesses a cocksureness of the unassailability of his administration despite a growing body of decisions constitutional scholars deem illegal.
The Trump administration begins to look beyond concepts of legitimacy, as if that legitimacy does not extend from the people of the United States, but from some other, unseen place.
He cycles through political topics and talking points in the same manner Facebook and Twitter cycle through ads. Constantly searching for the exact stream of content that produces the greatest data signal.
Data fiction or data fact?
To fight a weaponisation of data requires taking a reductionist view of human cognition. We must consider human beings to be simple input/output processes.
Despite the infinitely-complex structures of neurons within each human brain, that infinitely complexity can now be externally-understandable enough so that someone even possessing a modicum of power to remotely map stimulus/responseses on an individual basis.
Most Americans today not only have a unique physical identifier (in the form of a mobile phone), but also now have unique cognitive identifiers – in the form of social media accounts. Not only is one’s physical location traceable, but now also one’s mental state.
The American President today possesses access to systems that tracks each individual’s location in real-time. He also possesses a system can firstly derive an individual’s mental state. The ultimate question is, can that system of byte-streams modify individual mental states?
One wonders about the fates of Jack Dorsey and Mark Zuckerberg, heading publicly-traded companies collectively mapping the minds and moods of Americans in real-time. There would perhaps no more valuable databases for an authoritarian regime than these.
Most disconcertingly, far from humanising technology, our human leaders now look driven by machine logic.
The Westminster Prison Experiment
One consistent theme across Dr. Siddhartha Mukherjee’s brilliant book The Gene reflects the truism that, generally, experience proceeds understanding. It took five generations of scientists to move from Charles Darwin’s initial observations — his experience of gene expression — to a gene-level understanding of how individual genes express.
Similarly, what we can observe today is the observation of how machine learning and social networks modify human decision-making and governance. Equally, we are very far from understanding how text strings on a mobile device modify neurotransmitter release. Further, how modifying neurotransmitter release in one individual can cascade and catalyse neurotransmitter releases in other individuals. How data now readily manifests as social contagion.
While we do not understand the mechanisms of action, we nonetheless can recognise that action is present.
To return to Brexit briefly, rather than the British government shaping Brexit, Brexit now profoundly shapes the British government.
In the Stanford Prison Experiment, Philip Zimbardo created a context which readily turned students into aggressive prison guards that completely dehumanised their fellow prisoner students. The shift was so swift and profound that Zimbardo was forced to halt the experiment. His review was the power of context to shape behaviour. Given absolute power, dark subconscious human traits readily manifest.
We may now see the same thing today, with Theresa May’s ascension to British PM. She began as a moderate remain campaigner, aligned with David Cameron’s view of the United Kingdom as a leading player in a reformed Europe. Less than a year later, not only is she stonchly focused on guiding Britain out of the European Union, but doing so regardless of the costs.
After a relatively quiet retirement, Tony Blair identified the UK government’s function has become so focused around Brexit, it is incapable of seeing anything else. Similar to the Stanford Prison Experiment students, who shed their roles and relationships with each other, transforming from friends and colleagues into new identities that bore little connection to identities possessed only days prior.
Rather than Theresa May shaping Brexit, the ‘mandate’ given by the British voters at Brexit now completely shapes Theresa May.
If we are at the point of weaponised artificial intelligence, while both current leaderships and electorates readily observe the fundamental shifts technology is driving, neither is capable of understanding the complex drivers behind those shifts.
Perhaps we have entered a new version of the Stanford Prison Experiment on the scale of the western world. Where ordinary businessmen and politicians become authoritarian prison guards, and persecute people who were days before collegues and friends.
Cambridge Analytica optimises for a single variable — delivering the voting results specified by its clients. What human experience it unlocks in its social media targets getting to those results are inconsequential.
If the feedback loop says that the EU Referendum vote can be shifted by generating racist overtones and social strife, the system is then intentionally optimised to generate racism and social strife.
A solution is quite simple. These systems must be configured to solve problems based on multiple constraints. Help achieve the target goal for the customer, without also generating secondary effects like racism. In word, regulated.
Science or science fiction?
To close, the technology components as presented here exist. Today, the core components of artificial intelligence and machine learning are freely downloadable, and both individuals and companies around the world are using those tools to effect change in every individual area of human life.
The fundamental question as to whether or not Donald Trump is a legitimate President or not, comes down to one’s personal belief if the technology systems in place today are powerful enough to allow the minds of voters to be manipulated and/or hacked?
Is it possible for today’s systems to deliver byte-streams of content that modify the human brain on both mass scales and an individual basis simultaneously?
If it is, then not only do we have a constitutional crisis coming, but a fundamental crisis in the very nature of human governance. We have achieved a level of technological advancement so powerful, that the fundamental inequalities of existing systems will be pushed to their extremes.
If that is the case, we have reached the limits of this format and implementation of democratic rule. Maintaining democracy moving forward requires regulation and transparency of tools capable of cognitive manipulation.
We cannot ascribe legitimacy to a theory of governance based on individual rationality, if a small number of actors now have the potential to modify the rationality of the electorate at large, without the latter’s conscious awareness of the former.