Divisiveness is efficient for watch time, and watch time leads to ads

The videos in favour of Remain weren't divisive, because being in favour of Remain is the moral, right-thinking stance.

Anti-Brexit is the status quo. It's by definition not controversial since they can't do anything other than advocate for how things currently are.

Pro-Brexit is by definition controversial since they were advocating for a complete upheaval in the relationship with the EU constructed over decades. And it was always going to be tinged with xenophobia and nationalism since that was the entire premise for the Brexit vote to begin with. And it doesn't get any more divisive than xenophobia.

Not necessarily.

For example, recommending videos that are somehow categorized "pro-evolutionary theory" would almost certainly be less divisive than recommending Creationist videos that attempt to "debunk" evolutionary theory. The pro-evolutionary theory videos would typically be substantive scientific talks or lectures that require patience and critical thinking in order to understand. The Creationist videos, on the other hand, typically use lots of underhanded tactics. They will attempt to mislead the viewer and typically throw out so many different arguments from so many disparate fields that it's impossible for any one person to debunk them all out of hand.

If you recommend the "pro-evolution" videos you probably end up with a lot of bored viewers who don't even make it through a single video. If you recommend the "anti-evolution" videos then you'll get a biologist debunking point 1 of 100 and a troglodyte responding "what about point 2-100?"

I have no clue whether the pro/anti Brexit videos fit that bill. But it's certainly a possibility.

Brexit may not be the best example, because (perhaps, arguably) there could be legitimate arguments for the pro-Brexit stance.

Vaccination might be a better illustration.

Promulgating the view that "vaccines are a hoax, and vaccinating your child will cause autism, or let the government achieve mind control" is both incorrect (scientifically, factually) and wrong (morally). Those convinced by this view put their child and other children around their child at grave risk.

A system that promotes this view over the established, normative, scientifically validated consensus view that we should vaccinate our children, to the point that it changes opinions and behavior, is harmful to public discourse (and harmful to children).

This is true whether automated/algorithmic, or intentionally devised and operated by people with malevolent intent.

At scale, these phonomena become harmful to democracy itself. Democracies are diminished when their subjects believe false information, especially objectively harmful false information whose belief causes disease, stuctural collapse of infrastructure, etc.

Since we are now observing this happening to various extents in various areas of society, we should use our power as democratic citizens to compel operators such as YouTube to provide real data on what their algorithms are selecting, so that we can understand the effect on our society.

At least, that's what I got out of his tweet.

The engineers (and friends) who built these systems are wildly successful, and as often happens in these cases, they now feel guilty about that success, especially when they're surrounded by so much human suffering (in a place like SF), which pushes them in a progressive direction.

But now they've discovered that the systems they built had unintended consequences and may have even led to the very success of the political ideologies they despise.

The cognitive dissonance working at a place like Twitter must be palpable. Trump probably drives more revenue to the company than any other person, and yet probably 90+% of the employees who work there despise the man and everything he stands for.

So you can see why so many people are on-board with trying to do something about it. But the same poorly-thought-out, gung-ho attitude which led to algorithms no one understands anymore, will backfire when they try to "correct" them.

Social engineering is a dangerous game. This country is a time bomb just waiting to go off and when Silicon Valley finally pulls the plug on Trump there's no telling where it will end.

(though I do think things have calmed down a little in the last couple years... we'll see where they end up after the midterms...)

> If they just chose more responsible business models, this wouldn't be happening.

No effort on social change that requires that corporations voluntary forego profit (or that requires that individuals forego status) will ever work, nor should it. Everyone talks about what behavior should be, but nobody wants to talk about the sort of incentive structure that would produce that behavior.

Fundamentally, there is a lot of content out there. So much so, that it is necessary to have some algorithm assist in curation. Any algorithm you choose will favor certain types of content over others (unless you literally choose randomly, in which case you aren't really solving the original problem). The question is what should that algorithm be optimizing for. We have seen what problems optimizing for ad revenue causes; but I see no reason to think that optimizing for some other metric wouldn't produce a comparable class of problems.

Ultimatly, the problem of algorithmic curation is a new problem, and no one has a good answer for it. Whatever answer we pick will have significant influence on the public discourse, because so much of said discourse is filtered through the algorithm.

Working inside of YouTube, he was in a position to see that the algorithms heavily promoted Brexit. However, for people not directly involved in YouTube's operations, this is not necessarily apparent — and is essentially impossible to check.

His point is that the ultimate outcomes of this algorithmic selection has potentially massive societal repercussions, and therefore the citizenry deserves to know what is being selected for promotion.

I'd bet something along the lines of "absolute_value(upvotes divided by total votes - 50%)" is a feature somewhere. That's how I interpreted OP.

They have to at least believe in the desire to be informed. I agree with your point and optimizing for eyeballs means you're gonna optimize for what people want to see. But as we've seen lately, if that machine driven optimization goes against the wishes of the squeakier of the wheels, it is considered intentional favoritism or at best negligence. Internet advertising is the popular scapegoat these days when cognitive dissonance amongst the loud is all that remains.

However I will keep it here as a prediction of future events.