Comment by CodiePetersen | Hacker News Reader

Comment by CodiePetersen | Hacker News Reader

Parent Full thread

I never said generic data was the sole item that acheived them their victories. I claimed data was helpful and lead to victories and then clarified data helpful in targeting certain individuals. You claim there is no evidence that data was useful and go so far as to say Trump didn't even have a real voter database. Even though running as a republican he had access to the RNCs and Cambridge Analytica did indeed provide support.

Targeted advertising can and does work. Google has shown this with an experiment called the redirect method in which they used ads to target and deradicalized potential isis members. Another individual used the same method to target suicidal individuals and was able to run a campaign to get them to call a hotline. He had I believe a 38% conversion rate where as a standard ad conversion rate is around 2% to 4%. Again this is in the public domain.

Data, targeted or otherwise, is like any other tool. It is not going to be some magical device that gives victory. In the end you have to use it wisely and correctly infer a pattern that can be exploited. Any relevant information is better than having no information. And based on the many papers and efforts to use data to influence the population there is more than enough evidence to suggest such data gives you an edge. It may no be a large margin but it could be enough to make a difference.

If we consider amazon for example, they don't need 40% conversion rate. If they can just gain even a 1% extra conversion rate, that represents 100s of millions of extra revenue. Especially at scale, data is indeed useful.

Yes, I agree targeted ads work for some things, like people who are wanting to find the thing the advert is for. But the claims being made here are very different - that ads can "capture" people politically even if they didn't set out to be converted and this was "pivotal". I'm very skeptical of this claim and the only studies I'm aware of refute it.

To be clear, I'm not arguing that voter databases are useless: they're useful to get your own base out to vote and not to waste time canvassing hard-core supporters of the opposing party. This isn't the sort of data people are hyperventilating about when they talk about Trump, Brexit or Cambridge Analytica, is it?

But even if data was useful in this way, why would that be bad? If data is used to most effectively focus effort on convincing people, they're still convinced by the arguments at the end of the day, and feel they made the correct decision. What's wrong with that?

Google has shown this with an experiment called the redirect method in which they used ads to target and deradicalized potential isis members.

I recall this claim. It's implausible on its face.

The problem is they have no idea if the people who saw the ads were actually potential ISIS members. This isn't a precisely defined term and they can't test that theory against reality - it's not like normal AdWords where you can measure against sales to figure out if the ad viewer was really interested in making the purchase. The assumption that they targeted "potential ISIS members" was based on using (some admittedly clever) keyword choices and then assuming the only people who search for those things are people who want to become a jihadi - a massive and untested leap of logic.

They say over 320,000 people clicked the ads in only 8 weeks. That's a staggering number of supposedly potential terrorists. In 2014, when ISIS was growing in strength, they had only about 14,000 fighters according to US intelligence. So they're claiming they "redirected" a population 20x the size of all of ISIS in only eight weeks. If there were really so many people trying to join ISIS at that time their army would have been vastly bigger.

Much more likely, the vast majority of people who clicked the ads had no intention of ever becoming a jihadi and their campaign achieved nothing.

The suicide ad is much more believable - it's well known that many people who attempt suicide don't really want to die, they just want help.

I think what is wrong with using such data, as I have stated, is that it is used at scale from single entities. If you look at the point of free speech what was it? The point of free speech was that many ideas could be expressed and there would be less of a chance that bad ideas take over. If you look at anti monopoly laws, what was the purpose. The idea was again that single entities had less bad influence over the economy. These are obviously simplified reasons, but in the end, the effects are actually that simple.

I believe we have run up to scenario where we must again reevaluate our laws to protect society from bad/harmful trends from large singular actors. They may not intend it to be harmful, but the fact is, it could be.

"[..] a massive and untested leap of logic." Google advertising is that massive and very well tested leap in logic. As for potential ISIS members, if you recall it, then you know that they were not only targeting just fighters they were also targeting ISIS brides and sympathizers. Any counter intelligence expert, or you can just read the manual on counter intelligence readily available to the public [https://publicintelligence.net/us-army-counterintelligence/], will tell you that terrorist or guerrilla support networks are just as important and damaging as the actual fighters, hearts and minds they say. They provide intel, monetary support, propaganda support, non combatant support like supply runs, etc. So you can be a member without being a fighter and still be vital.

Also when you talk about conversion rate, you're not talking about who clicked the ad. You are talking about who "bought" or "interacted" with a product after clicking. So you are gauging real activity from people who are clicking and looking at your stuff. 500k minutes of video watched. On average about 2 minutes I believe. That's a huge amount of interaction. You can barely get an invested mobile game player to watch 15 seconds for rewarded things they can actually use.

You claim untested and unproven with no evidence a lot, but the results are right here: [https://redirectmethod.org/pilot/#results]. They showed remarkably different results. Twice as good actually, in the targeted approach. So I don't know what to tell you. There is evidence, there are experiments, and there are real tangible results and effects on the public. The only question in my mind is at what scale do we allow that.

Reply on news.ycombinator.com