I think Doctorow's analysis makes more sense than the typical rhetoric about Facebook and CA. But I still default to thinking that CA was largely ineffective. How many "latent klansmen" were there unaware that Trump was more their candidate than Clinton? Trump's stances were completely public, hugely well-known. What more did CA provide? I just don't really buy it. (Political advertising is very ineffective from what I've read and I just need some reason to expect CA to be more successful.)
I'm worried about the surveillance system for other reasons, like finding targets for censorship, discrimination, and hatred.
I think the proposed theory is that there is a rather broad class of could-be voters that typically don't engage in politics or vote. One example of a small category of people would be klansmen, I imagine for a group that is often sidelined they probably think their vote is wasted and therefore don't typically show up at the ballot on election day.
If your party can identify and connect with those individuals and spark a sense of outrage in them, you are likely to be able to get some portion of them to the ballot on election day.
So it is more meta than educating somebody on Trump/Clinton's politics; it is more about behavioral manipulation via outrage, fake news, etc. The goal isn't to change knowledge or opinions (which is extremely difficult), it is to fan the flame of targeted groups' collective outrage enough to bump election day participation. In many places getting just a 1-3% edge is the difference between winning and losing the election.
My personal takeaway from this is that improving general access to the ballot by making it easier to participate and vote should obviate the effectiveness of this strategy. Each step approaching 100% voter participation forces smaller and smaller returns on the strategy of motivating extremist voters.
I'm worried about the surveillance system for other reasons, like finding targets for censorship, discrimination, and hatred.