Ricochet is the best place on the internet to discuss the issues of the day, either through commenting on posts or writing your own for our active and dynamic community in a fully moderated environment. In addition, the Ricochet Audio Network offers over 50 original podcasts with new episodes released every day.
Fans of Brexit’s Vote Leave campaign might remember Dominic Cummings’s reflections on the uses (and abuses) of data in politics. Cummings, often hailed as the mastermind behind Vote Leave, is an eloquent advocate for how getting the data science right contributed to Vote Leave‘s success, and he has a prickly – even “psychopathic” – reputation as a man who won’t suffer data-science fools (or at least those whom he deems foolish) gladly.
No doubt Cummings is right that charlatanism infests the ranks of political “data scientists”, but a more charitable term than “charlatanism” for much iffy “data science” might be “ad-hockery”: Adventurous wunderkinds promote ad-hoc heuristics which seem to work well enough, or which work until they don’t, or which may work, but which haven’t yet been vetted by systematic scientific reasoning. Ad-hoc heuristics aren’t inherently deceptive, or incapable of delivering what they claim to deliver. They deserve to be met with plenty of skepticism, of course, but skepticism needn’t always include suspicion of fraud.
In the most purist, nerdly recesses of my heart, I agree with Cummings that ad-hockery just isn’t scientific enough to count as data science. On the other hand, the Right – especially the populist Right – has borne a reputation as the plodding, unadventurous “party of stupid” for long enough that I find the great scandal of Cambridge Analytica’s audacious adventures in data strangely cheering: Sure, Cambridge Analytica’s acquisition of the data of 50 million Facebook users was dodgy, and “psyops” targeting voters’ personality traits (such as the “Big Five”) may not work as hyped (Cummings, tweeting under @odysseanproject, seems quite confident that they don’t). But at least the stodgy Brexiteers and Republicans who employed Cambridge Analytica were willing to get out of their comfort zone and experiment! Out of their comfort zone? Blame Canada – and Bannon. More specifically, blame the gay, vegan, pink-haired Canadian fashionista Christopher Wylie, a self-taught computer prodigy who came up through the ranks of the Liberal Democrats, only to be recruited by Bannon to play the Nate-Silver-wannabe role of predictive wunderkind for Cambridge Analytica.
According to Guardian reporter Carole Cadwalladr, whose legwork on Cambridge Analytica remains impressive despite her biases,
At 24, [Wylie] came up with an idea that led to the foundation of a company called Cambridge Analytica, a data analytics firm that went on to claim a major role in the Leave campaign for Britain’s EU membership referendum, and later became a key figure in digital operations during Donald Trump’s election campaign.
Or, as Wylie describes it, he was the gay Canadian vegan who somehow ended up creating ‘Steve Bannon’s psychological warfare mind[expletive] tool’.
Wylie began his career in politics as a bright but troubled lad who had dropped out of school at age 16.
[A]t 17, he was working in the office of the leader of the Canadian opposition; at 18, he went to learn all things data from Obama’s national director of targeting, which he then introduced to Canada for the Liberal party. At 19, he taught himself to code, and in 2010, age 20, he came to London to study law at the London School of Economics.
Wylie was “studying for a PhD in fashion trend forecasting” when he
came up with a plan to harvest the Facebook profiles of millions of people in the US, and to use their private and personal information to create sophisticated psychological and political profiles. And then target them with political ads designed to work on their particular psychological makeup.
Bannon and the Mercers seemed interested not only in Wylie’s plan but also in who Wylie was. According to Wylie, both saw gay individuals like Wylie as an asset,
“[Rebekah Mercer] loved the gays. So did Steve [Bannon]. He saw us as early adopters. He figured, if you can get the gays on board, everyone else will follow. It’s why he was so into the whole Milo [Yiannopoulos] thing.”
Of Bannon, Wylie remarks,
“He’s the only straight man I’ve ever talked to about intersectional feminist theory. He saw its relevance straightaway to the oppressions that conservative, young white men feel.”
Wylie evidently warmed to Bannon’s willingness to treat politics like fashion in order to wage information warfare, as well as Bannon’s willingness to treat intersectionality as more than just the butt of right-wing jokes. Intersectionality and politics-as-fashion may be hallmarks of the SJW stereotype, but it’s possible many in the red tribe found themselves warming to Bannon’s willingness to use “SJW-like” strategies, too: Breitbart readership has fallen by half since Bannon’s departure, a sign that Bannon’s unique outlook struck a chord that’s hard to repeat.
Wylie and Bannon may have made the perfect odd couple in the info wars. But were they doing data science?
As far as I can tell, Dominic Cummings would say they weren’t. The Brexiteers were a fractious lot, with the overall organization of a clown car stuffed with stabby sorority sisters, if Cummings is to be believed. Cummings asserts Cambridge Analytica was “100% irrelevant” to Vote Leave’s success, referring darkly to a January coup, which may have involved different Brexit factions going their separate ways over data analysis. Cummings tweeted,
whole point of the Jan16 coup & Farage trying to get me fired was my refusal to work w/ those bozos, the exact opposite of [Guardian reporter Carole Cadwalladr’s] conspiracy [theory]
“Those bozos”, in context, appear to include Cambridge Analytica. Nearly getting fired for refusing to work with “those bozos” is by itself plenty motivation for supposing “those bozos” incapable of real data science, so Cummings may harbor a bias against Cambridge Analytica’s competence. Complicating the picture, Cummings did work with AggregateIQ, a firm which apparently shared software with Cambridge Analytica, and which Cummings characterizes as,
AIQ are NOT a data science or data analytics company. They did not do DS/modelling as I have explained repeatedly.
Perhaps Cummings prefers to be unreasonably rigorous in his definition of what counts as analytics because it suits his politics. A man who boasted of bringing in real physicists to create the analytics which eked out Vote Leave’s spectacular Brexit victory may shy away from being associated even indirectly with the “data scientists” employed by the scruffier politicians (Cummings doesn’t seem to think highly of the UKIP crowd), some of whom developed their data-manipulation skills through ad-hoc training in fashion and politics, as Wylie did. Perhaps Cummings’s desire for more rigor in political data analysis is inseparable from a desire to be seen as more respectable politically, physicists being more respectable to the mainstream than data-jockeying fashionistas and political operatives smacking of the “alt right”.
Reporters describe Wylie as brilliant, but brilliance doesn’t guarantee truthful analysis, especially when you’re young, arrogant, and inexperienced. Still, it’s hard not to admire Wylie’s gumption in pursuing his crazy “psyop” ideas as far as those fostering him would allow.
One way data science advances is through upstarts being willing to stress-test their crazy heuristics until they find one that doesn’t fail. That’s the hacker’s way of doing science, neither as satisfying nor as respectable as carefully building heuristics up from first principles, and those hacking their way through the data with one ad-hockery after another should be shot down, whenever possible, by those with a better grasp of the data’s scientific context. But saying hackish ad-hockery deserves to be attacked by the full majesty of respectable science isn’t saying hackish ad-hockery shouldn’t exist. Rather, it’s saying that both ad-hockery and systematic scientific thinking benefit from one another, though they’re not equal partners in what really counts as science.
Disentangling the gumption to try crazy ideas from the gullibility to succumb to crazy ideas is tough. Cummings emphasizes that politicians and reporters alike are too easily gulled into treating theories as “science” when they aren’t. It’s tempting for those (like reporters and politicians) who excel at spinning compelling narratives (rather than vetting narratives for strict scientific accuracy) to believe that the world of data science is a world of data science fiction – a sensational, spy-novel world where “psyops” must surely work, where “information warfare” is something fractious and limited humans have cracked the code of waging, and so on. Some bias toward conflating the sensational with the scientific is not only understandable but perhaps even necessary: reporters incapable of writing about science as if it were thrilling, chilling science fiction would bore most readers out of reading, after all.
Cambridge Analytica is busting up. Its CEO has been suspended. Its headquarters have been raided. Its wunderkind Christopher Wylie has been cooperating with reporters for a while now on exposés to bring Cambridge Analytica down. Did Cambridge Analytica succeed in becoming a highly-effective force for sinister political manipulation? Doubtful. Did it have much power at all, other than the power to bamboozle the more populist politicians and their wealthy donors into financing its adventures?
Robert Mercer was the political donor backing Cambridge Analytica, and Mercer, himself an AI expert, is no dummy. It seems possible Cambridge Analytica partially delivered on its promise to influence a non-negligible portion of voters to either get out and vote on election day or stay home (influencing voters to change who they’d vote for in the first place is much more difficult, and usually beyond the scope of such operations). This by itself is no great scandal, though many would like it to be. It’s also possible Cambridge Analytica advanced some useful analytic techniques in the process, even if much of the “data science” it was peddling turns out to be mere science fiction.
Cambridge Analytica isn’t the first outfit attempting to use big data or psychology to influence voters, and it won’t be the last. It’s not the first to sell as data “science” what might merely be science fiction, and it won’t be the last. What’s most remarkable about Cambridge Analytica isn’t what it tried to do, but who it tried to do it for. It’s one thing for hip, cosmopolitan politicians like Obama to exploit big data. That seems trendy. That seems natural – tribally “in character.” It’s another when “deplorable” politicians do – not because “deplorable” politicians will obviously be more successful at exerting evil influence thereby, but simply because it’s incongruous – deplorables have a reputation for being plodding technophobes! Perhaps Cambridge Analytica’s meteoric rise and fall are so shocking because it’s the best-publicized evidence we’ve seen so far that politicians representing the red tribe can be just as willing to experiment with novel ideas and technology as those who represent the blue tribe.
Would it be even better if more right-leaning politicians did as the Vote Leave campaign, under the flinty eye of Cummings and his allies, did, tempering their adventures in data with considerable rigor, skepticism, and transparency? Yes, it would be. But, as Reagan said, half a loaf is better than none, and you can come back to get the second half later.