Hacker News Reader

Comment by godelski | Hacker News Reader

I'm honestly surprised we aren't seeing these big tech companies have an arms race with respect to homomorphic encryption. Since you can perform computations on the encrypted data itself this appears to be the best of both worlds: anonymous and still able to pull big data. Probably won't be able to serve as personalized ads, but from what I'm aware, the big data is far more important than the direct targeting.

I know Meta is betting big on the Metaverse, but it's also wild to me that they similarly don't bet big here and keep their ad infrastructure.

https://en.wikipedia.org/wiki/Homomorphic_encryption

https://news.ycombinator.com/item?id=32987155

View on HN

If it the concept that counts, not just that particular tech, some do "race" to get big data without compromising privacy, even when they don't have to (because no competitors are racing on the privacy dimension).

For a survey, check out what firms are doing on "differential privacy" not just "homomorphic encryption". As a for instance, Apple -- since the OP is about them -- spent significant extra effort to make Maps anonymized.

Concept: https://www.apple.com/privacy/docs/Differential_Privacy_Over...

In Apple Maps: https://www.idownloadblog.com/2019/03/13/apple-maps-navigati...

Apple maps was based on OpenStreetMap originally, which actually tracked users less than the Apple Maps[0] app. Guess all that special work was for nothing.

[0] https://www.apple.com/privacy/labels/

Ads targeting is actually a huge motivation for ads platforms, IIRC; customers buying ads want to know if they work.

Ad targeting is much broader than that though. Even with encrypted data you can match keys to profiles. You just don't know who that profile belongs to and you don't even need to know the contents of that profile. The whole point here is that you can treat the system like a black box but still work with it in a useful way.

Brave has proven that targeting is not always the best answer. They did an analysis on their own platform, their CTR is very very good and they do not compromise privacy in any way and still deliver ads (opt-in).

I assume brave shows mostly crypto adds? If so, the targeting is baked in

Yes, mostly web3 projects but there are other advertisers as well. Like Twitch for example.

> They did an analysis on their own platform

That's always how good studies start.

Studies? This is CTR we're talking about. It doesn't have to be a research paper to show what's the average CTR, advertisers know it beforehand, they just made the info public.

https://brave.com/brave-ads/casestudies/

Main reason QUIC was created iirc.

Apart from the fact that they have no incentive to, there are technical downsides, even assuming you could get a useful homomorphic scheme to work in practice.

For example, encrypted data can't be compressed. Columnar big data systems rely heavily on that to be performant.

> Apart from the fact that they have no incentive to

They have huge incentives. Apple is positioning themselves as the privacy king. Meta has suffered huge losses because of their privacy abuses. Neither of these suggest no incentives. I'd argue that they suggest large incentives.

The technical downsides are a fair critique though. But this also is where competition excels. Our machines are getting faster. Other algorithms have also gotten extremely faster and it would be naive to assume that homormorphic algorithms similarly don't. This is why I say an arms race.

Apple has done plenty of major screwups that each put a big doubt on whether “Apple loves your privacy” is something more than PR BS, but for some reason, they all got memoryholed.

An iphone user is _probably_ tracked by less ad agencies than a stock android user, but that’s it.

> They have huge incentives. Apple is positioning themselves as the privacy king

I am willing to bet that it is a deal breaker only for very small minority of users. I would argue that search engine data is of much bigger privacy concern than OS, as compared to search history no OS action comes close in disclosing personal action, and google/bing almost has universal market presence even for Apple users.

> Meta has suffered huge losses because of their privacy abuses.

They had their revenue reduced, true. But it is not loss. They wouldn't be in any better position if they didn't used extra data when it was available.

I wonder if there's new feature or improvement Apple could introduce that would actually increase their sales. It seems like at the moment people buy the new iPhone because 1/ they need/want a new phone, 2/ they want an iPhone, and 3/ they want the new iPhone.

Apple know what they need to do to be successful, and that's continue to release updates to the iPhone that in a few small ways make it slightly better than the previous one.

Not to mention being able to keep processing data, while your competition gets squeezed by tightening regulations around the world.

The other thing is you could likely squeeze your competition out of the market. Countries are waking up to the implications of data harvesting and that it isn't only being used by themselves but their adversaries. So if a big company, like Apple, got it working, it would be much easier to lobby legislation that would harm (or kill) your competitors. (While I think we would all be better off if this happened, I do recognize that this same power can be abused and cause a significant disruption in the ecosystem. But I think it would also be extremely difficult to impossible to gain such an edge that others couldn't quickly adapt. Even if they drag their feet)

It's unclear when homomorphic encryption will be ready for prime time, especially at big tech scale. It's easier to give up the data under assurances of how exactly that data will be used. Secure enclaves (e.g. AWS nitro enclaves) can be used to offer cryptographic attestations as to what the code is doing with the user data. Practically speaking, an average user won't inspect the code, of course. But it does allow for other companies to audit and vouch for the usage.

as the a fan of homomorphic encryption I very much hear what you are saying (two examples I thought up https://news.ycombinator.com/item?id=32993163). But at the end of the day, there's no way to prove that Bob's Homomorphic Encryption SaaS company isn't "backing up" all your data into a regular database as well. And that's assuming Bob never gets hacked, too.

At the end of the day, even with HE, we have to trust Bob to be trustworthy, and to have designed and implemented it totally correctly (or else data will leak in the clear).

If it was Bob's small company then having a 3rd-party external auditor review the system would go a long way to gaining the public's trust, but unfortunately Meta has no such luxury. They could say they can't see your data until they were blue in the face, and they could even be telling the trust, but people don't currently believe they haven't sold your data (or the specific nuance there), so I'm not surprised they haven't made a large investment in HE. HE is also not quite there yet. It's ridiculously slow for anything but the most simple operations and Meta's data needs are far from simple.

Yeah you make a really good point (especially with those two examples, which the latter might not even require data collection at all). But I do think that an arms race in this direction would change the public view. You'd sure have a lot fewer nerds like us arguing about privacy. Though we'd probably be arguing more for open source to lessen the trust issue.

I don't think we'll ever have fully trust-less systems. But I do think there's a big difference between saying "trust us, we don't look at the data" vs "we encrypted the data and use this method, trust us that we aren't decrypting it". The former method is worse because there's a lot of people that have clear access to the data. The latter is better because there are more speed bumps and the argument is more sound. A lot of trust is built from demonstrating good faith efforts.

Reply on news.ycombinator.com