Comment by PaulHoule | Hacker News Reader

Comment by PaulHoule | Hacker News Reader

Parent Full thread

Every machine learning project I worked on that failed failed because people (sometimes me) didn't figure out a way to get good training data.

Classification on a binary basis ("that person will like that video or not") is on a much better ontological basis than multiple class classification ("out of 50 links, that person picked will pick the 39th") For one thing it is possible to calibrate a yes/no decision, TikTok knows that there is an 80% chance that you're going to like this video and a 30% chance that you are going to like that video and make a rational decision about what to show you. Multi-class classification is a shot in the dark in comparison.

Google and everyone else in social media wants to look like the image in

https://tvtropes.org/pmwiki/pmwiki.php/Funny/Idiocracy

They give you 50 things to click on and make a big deal about having a privacy policy and how "data is the new oil" but the clickthrough data is absolutely worthless because out of 50 things there were 5 things you might have clicked on and you could only pick one. In particular, they can't come to any conclusion that you didn't like any of the other 49 things.

The HR office hires one black woman for their DEI initiative to work in "AI Ethics" to claim this technology is so powerful that it's actually dangerous and when they get into some spat with management and get fired they are doing their job because it legitimizes the idea that this "artificial intelligence" is so powerful it's dangerous but the web brought to you by Google, Facebook, Twitter and such is artificial stupidity inspired by Idiocracy just as Facebook's metaverse is inspired by Sword Art Online.

TikTok on the other hand really is dangerous, not because it's algorithms are better (they are, see https://en.wikipedia.org/wiki/Multi-armed_bandit) but because their whole approach gives them meaningful data that can be put to work. It vindicates the idea of looking at one piece of content a time or a "Tinder for documents" that I talked about here https://ontology2.com/essays/ClassifyingHackerNewsArticles/

If you thought this comment was insightful, remember to vote it up because it helps the Hacker News algorithm.

Why does the multi-armed bandit problem imply Tiktok's algorithms are better?

It's a disciplined approach to explore vs exploit.

Just like the Kelley criterion tells you how to turn predictions into bets, the multi-armed bandit tells you how to keep a user engaged in the short term by showing them things you know they'll like but also keep them engaged in the long term by showing them things they might like occasionally which, if they pass the test, will expand the range of things that you know they like.

Multi-armed bandits are used all over the industry. They were in use before TikTok existed.

I never really understood the "dangers" of big corporations knowing you. What are they going to do, serve you better ads? I've thought the same for 20 years, and so far I've been vindicated - there's no widespread personal harm in having an "open" internet persona vs using incognito all the time.

Your personal data is of no value (unless you are of high-profile 0.01% of people).

However, the data of your city, state, country, gender, race is very valuable because then, for example, companies/government/foreign actors can know exactly what is this town, state, country, gender, race is thinking about any given topic. This is enormous value and this is why companies like Cambridge Analytica are so profitable.

I agree, but how is this different from polls, fidelity programs at department stores, statistics on various topics, etc - all of which have existed for decades?

It's different because department store fidelity programs weren't generating terabytes of raw data per day, feeding it into ML systems, correlating output against millions of other data points available for purchase, and then using the resulting insights to manipulate political affairs on a large scale.

I still see no problem with it. They're getting value, I'm getting value.

Mass scale political manipulation? See Brexit, for instance.

It appears democracy has changed fundamentally in the past few years. These days, whenever votes against your personal favourite, they have been manipulated. Its basically the same story on the left and on the right. Both claim the voting result could only have happened because the other side was tampering with poor voters.

Main problem with brexit was bubble. Even now, after so many years, you're saying that it's manipulation, simply because you don't personally know people who wanted it to happen. Who exist, I assure you.

> What I mean is – well, take creationists. According to Gallup polls, about 46% of Americans are creationists. Not just in the sense of believing God helped guide evolution. I mean they think evolution is a vile atheist lie and God created humans exactly as they exist right now. That’s half the country. > And I don’t have a single one of those people in my social circle. It’s not because I’m deliberately avoiding them; I’m pretty live-and-let-live politically, I wouldn’t ostracize someone just for some weird beliefs. And yet, even though I probably know about a hundred fifty people, I am pretty confident that not one of them is creationist. Odds of this happening by chance? 1/2^150 = 1/10^45 = approximately the chance of picking a particular atom if you are randomly selecting among all the atoms on Earth.

source: https://slatestarcodex.com/2014/09/30/i-can-tolerate-anythin...

-------

This being said, guarding my gps history like it's treasure does absolutely zilch to stop political manipulation, when you have stuff like corporate control over mainstream media and PR firms. See NYT's coverage of FTX, for a contemporary example.

Or, well... https://www.youtube.com/watch?v=ksb3KD6DfSI

No one doubts Brexiteers exist; they won the referendum.

How they came to that opinion is where the manipulation comes in.

You're really saying that after reading the second half of my comment? Mainstream media is overwhelmingly progressive and not shy in being unsubtle, but somehow brexit did it all by being so freaking brilliant that it manipulated even more... somehow.

We can pretend “mainstream media” in the UK doesn’t have things like The Sun in it if you like; no one asserts The Guardian was part of the Brexit full-court press.

You do these things with social media bots, trolls, misinformation, etc.

Brexit wasn’t brilliant. It was an empty slate people could project their worries on. That’s why actual implementation, where details mattered more than slogans, is where the difficulty lay.

You know, the attitude of "plebes are plebes, they should be protected and carefully told what to think so they don't to stupid things like Brexit" is very un-democratic.

They had access to both sides' discourse. They voted. They made a choice.

Instead of taking this information and learning from it, you're just blaming the very progressive-leaning social media.

Oh, why am I even trying. It's ok, it was Big Corp, they're the ones to blame.

You make me feel like a weirdo for hating ads and not wanting to see any of them. And if they're specifically aimed at me and make me more likely to spend money, even worse. I get tempted and then I feel like an idiot for spending the money I worked for on balls shaving machines and flexible pants that look like jeans on the outside.

No, I totally get that. Well, not balls shaving machines, that's between you and the algorithm, I get that ads are a fight for your attention. But that's just a regular tradeoff, nothing huge.

To me, people stating that they don't care about corporation having their data because they're just one person are equally as valid as those who think voting is useless because they're just one ballot.

Those people then typically complain about election results or corporation bending them over, without ever questioning their own responsibility in the outcome.

Are Facebook/Instagram Reels and YouTube Shorts then equally dangerous?

Reply on news.ycombinator.com