Hacker News Reader

Comment by mcv | Hacker News Reader

> "For years it has been next to impossible to get a result that is faithful to the search you actually typed in."

Good lord, yes. If I type two words, I want preference for sites that contain both of them, yet the first results all have either one or the other, because surely I must be more interested in a popular site that uses only one of these, right? Google is sometimes too smart, trying to interpret exact words I type as vaguely related words. Sometimes that's relevant, but often it's not.

> "For ten years after I met my wife the most relevant ads Google could think of was dating sites. Not toys, not family holidays,"

They have a tendency to show you ads for exactly the thing you don't need anymore because you already found it. I don't think AI is in any danger of taking over the world just yet. Except with bad advertising, apparently.

View on HN

The "AI" that Google's search engine seems to have is definitely feeling more "human" over time, but not in a good way --- it's like a stupid salesperson who has trouble understanding what you're trying to find. An analogy I have is that you go into a pet shop and ask for a black cat, and instead the salesperson shows you black dogs, white cats, and green gerbils (because they're absolutely cool these days and you wouldn't want to miss out on a great deal, no?)

They should have a checkbox to disable AI. :)

Something as simple as Apache Lucene would beat Google's algorithm any day for relevance.

But I guess Google doesn't care for relevance, they just care to show you ads.

> "They have a tendency to show you ads for exactly the thing you don't need anymore because you already found it. I don't think AI is in any danger of taking over the world just yet."

There's an eschatological trait to targeted advertising, as it seems to be all about past sins. So I'm not too sure about your evaluation and AI's own claims…

> I don't think AI is in any danger of taking over the world just yet.

The scary thing about AI is that, even as the algorithms have greater and greater intelligence, we're still not much closer to teaching them to do what we want them to do. They can game the system better than ever, and then the universe is tiled with surgical masks.

Maybe this is how AC finally reversed entropy.

https://www.multivax.com/last_question.html

Insufficient data for an answer.

So if AI ever takes over the world and kills us all, it will probably be because it failed to understand what we actually wanted.

https://wiki.lesswrong.com/wiki/Paperclip_maximizer

https://www.decisionproblem.com/paperclips/index2.html

Or perhaps because it was rotten to the core, as it was initially created for a malicious purpose (most likely advertising).

If "what we actually want" even comes into consideration, we did orders of magnitude better than the current industry standard. (Poor consolation, I know.) Right now, the vast majority of AIs don't even have a concept of "human desire" – probably none of them, to be honest, though some that are good at manipulating their handlers might've come close to a particularly stupid dog's understanding. This is at the core of the Friendly AI problem: https://wiki.lesswrong.com/wiki/Friendly_artificial_intellig...

Just because we created the AI doesn't mean it'll care about us. That's like saying a maths problem will try to make your favourite number its answer, just because you wrote it. No, you are the one who must make its answer the number you want. It won't happen by chance.

Corollary: you can't patch broken FAI designs. Reinforcement learning (underlying basically all of our best AI) is known to be broken; it'll game the system. Even if they were powerful enough to understand our goals, they simply wouldn't care; they'd care less than a dolphin. https://vkrakovna.wordpress.com/2018/04/02/specification-gam...

And there are far too many people in academia who don't understand this, after years of writing papers on the subject.

> They have a tendency to show you ads for exactly the thing you don't need anymore because you already found it

Say you searched for a TV a month ago. Now you're seeing lots of ads about TVs. Stupid Google.

But is it? A substantial fraction of those people are returning their TV because something is wrong with it. Now they are looking for another TV set.

Sure, the majority keeps their TV. But it is still profitable to target all those TV buyers, because they have self-selected into the set of people who really want a TV now, and they are willing to pay.

Reaching the fraction of those who need another one is probably[1] very lucrative.

[1] I'm sure Google has run the numbers

Agreeing: This thread of thought comes up semi-regularly here, I've argued similarly to you.

People will rebuy good products, or be stimulated to replace other similar products (bought a new TV for the kitchen, now the lounge TV seems dated, new boots feel awesome get another pair for when they wear out, new $thing is fun buy one for friends birthday.).

There's also a big place for brand enforcement. Show Sony stuff, to remind someone ['s subconscious!] they bought Sony.

A tertiary effect is what I call the "Starbucks Purposeful Bad Naming effect" - you get ads for the exact TV you bought -- beyond the brand reinforcement, etc., you also get to tell everyone you meet a weird story about how "internet advertising is broken ..." and "yes, my new Sony TV is great thanks, you should get one".

Those ad agencies aren't stupid; they have metrics for their metrics and have tracking that can tell you to the second when your gut bacteria burps ...

> Those ad agencies aren't stupid; they have metrics for their metrics and have tracking that can tell you to the second when your gut bacteria burps ...

Stupid they aren't, but I don't think they're smart in the way they are.

Ad attribution is a hard problem. Or, in other words, it's hard to estimate which $ spent on which advertising activities generated how many $ of profit where. That gap is a huge opportunity to scam the product vendor out of their money.

So the ad agency has a metric for their metric, and their reports overflow with numbers and various charts shaped like food or aquatic mammals. But does that mean anything at all? It might not. Statistics is hard, and as long as the vendor isn't better at it than the agency, money can be made. I used to work next desk to a group of content marketers who had no fucking clue about what their numbers mean, but their customers didn't have a clue either, so they happily paid money in exchange for reports that showed the Facebook campaigns "worked".

Now advertising industry is large, and by definition filled with companies that aren't a paragon of virtue and honesty. These companies specialize, providing building blocks and platforms for each other, and they compete internally. It's not like people building tools for lies and manipulation are suddenly honest when dealing with their in-industry customers and competitors. After all, convincing advertisers that your A/B testing package is worth the money requires... well, advertising.

So my personal view on the industry is that it's mostly self-reinforcing bullshit. Doesn't change the fact that it generates stupid amounts of money, though.

I used to work in Ad Operations (literally buying ad space and running campaigns) and can attest to the accuracy of this.

Clients were clueless: they had their metrics and they looked at them often, but from my interactions, deep understanding of those metrics and the realities behind them was lacking. The chain of technologies was patchwork and would rarely support all the required features from ad-serve back up to agency: click and view attribution was especially flaky and inconsistent. The adserving environment we worked in (in app) often had issues with view attribution, and we'd tell clients that, but we knew for a fact that some of our competitors didn't and clients would always ask us why our view attribution numbers were worse.

Combine that with more suspect behaviour from suppliers and competitors than you can poke a stick at (questionable traffic sources and campaigns that were probably outsourced from under you, suspect and plausibly forged numbers, etc) means that most of the metrics are plausibly poisoned with illegitimate data to a degree that is difficult if not impossible to nail down, which more or less makes lots of those metrics worthless.

> So my personal view on the industry is that it's mostly self-reinforcing bullshit. Doesn't change the fact that it generates stupid amounts of money, though.

Couldn't agree more.

> People will rebuy good products,

The big reason advertisers show the ads for products you already purchased is also to reinforce their brand. If you buy some stupid cable, you wont remember the name of the company that made it, but you will if they show you the add couple of times in a row and will likely to buy the things there again even if it not the same product.

Agree, that was in my 3rd paragraph above.

> Say you searched for a TV a month ago. Now you're seeing lots of ads about TVs. Stupid Google.

> But is it?

Yes, because Google knows who started searching for TV's again and who didn't.

Reply on news.ycombinator.com