AI and Blockchain: An Introduction | Hacker News Reader

AI and Blockchain: An Introduction

mattturck.com

132 pointshunglee235 comments

I’m curious as to what incentive people would have to provide their data to a public network. If they don’t, the premise of the article sort of falls apart.

I personally think that you’ll end up disappointed if you’re investing in decentralization. I see blockchain being applied at companies like Maersk, but it’s not really decentralized because it’s mainly operated by Maersk and the only people they let on the nodes are their partners.

ML is it’s own thing, the thing about ML is that it’s not really new. We’ve been doing manual data analytics since the birth of data, and all ML really does is digitize the analytic department. Right now it’s too expensive, too hard to utilize and requires data sets that are too huge for it to work, but eventually, it’ll automate much of your analytics department, because one data scientist will be able to do the work of x old school analysts, making it more efficient in terms of money.

I think the reason only AI and Blockchain were ever put together was to sell snake oil, because why on earth would you use a terrible database to run ML? Even today we don’t run our analytics on live data, instead we construct separate data warehouses on database structures build for analytics.

Good luck with your investments though, I personally put my “risky” investments into a robotics company called Odico and tripled the value.

I think the truth about blockchain is that the fad is over. It’s been 9 years. At one point I could buy coffee with bitcoin, today I can’t even turn my remaining crypto into fiat.

I simply believe (as someone with an economics and GT background) that you are right on your first statement. The issue with these more 'utopic,' i.e. would be great but hasn't happened for reasons, is that the incentives to contribute to such systems are lacking on both societal and corporate levels. If data isn't just out there, that's because the owners thereof are better off keeping it for themselves. It just takes a lot of goodwill to counterweigh this lack of incentives.

> I’m curious as to what incentive people would have to provide their data to a public network.

When people talk about combining AI and Blockchain that's very often the exact problem they think Blockchain can tackle: programmatic transfer of value, so as to systematically compensate the data providers, in ways and quantities that'd be inconvenient to deal with otherwise. I won't bother finding a buyer for my car's windshield data if all that brings me is 50 cents a month. But if data from all my various devices can get me 100 bucks a month, and the set up and maintenance for the system is automated and trivial, I'm very likely to take the free money.

The second aspect is that you don't need to provide your data to the whole network, simply to those who compensate you. Not everything needs to be on the blockchain.

Thirdly, efforts are made into developing techniques that allow users to provide data for training, while keeping the identifying aspects, or even the whole data, private, either with simple k-anonymization or techniques that distribute the AI training part (Enigma does this iirc).

Don't get stuck on the "blockchain as a database" part. What people are excited about is a whole field of cryptography and trustless decentralized networks.

I get that, I just think the assumption that people are willing to sell their privacy data for petty cash is flawed. It’s not like traditional advertising haven’t tried similar approaches, and failed, largely because petty cash is a terrible motivator for most people.

I guess you could argue that it’s different with crypto because your data is safer, but facebook has a billion users, and the people who’ve been doing the whole #deletefacebook thing, just signed up for Instagram instead. People really don’t care.

Would you sell your data for $5 a month? Maybe $15? I wouldn’t, especially not if it took time to setup and maintain, and I have a gmail account.

Plus: the people willing to sell their data for $5/month are absolutely worthless as advertising targets.

On the other hand, the intersection of these "people who want to sell for $5" to the general population gives some worthy targets!

I think there are plenty of people who are sufficiently unconcerned about privacy data to sell it for $xx per month. I also think this set of people has literally no overlap with the set of people who care about whether the data broker is centralised or not or whether it's possible to obtain cryptographic proof of which entities have received the data

If I didn't have to worry about security, yes. So a system that sold my data for me while keeping it private, I would use. Same deal with AI assistants or Facebook-like social networks, I would use them if there was a way to check that they don't leak data. I believe cryptography will open these advances to huge untapped demographics.

You can already sell your data though, and almost nobody bothers to do so. I don’t think it’s the privacy bit that’s stopping people, at least not at large, I think it’s time. $5 is a drop in the ocean to me, I’d just never bother giving you my data for that little, even if my data were safe.

The reason I willingly give my data to google by using gmail is that they give me an easy to use cloud mail with decent spamfilters. Or in other words, they are saving me time.

Maybe you could offer me something other than money? Like financial statistics on how I use my money? Sure, but those apps are already here, and people are willingly giving away their data without privacy in those.

So you see, it’s not that I don’t think you can’t convince people to give up their data, I absolutely do think that you can, I just don’t think petty cash and privacy are good motivators.

Maybe you have a bias because of your favorable financial situation (which is normal in here). For most people in the world, a reliable $10/month to cover their Netflix subscription or buy them a meal would be something they'd be very happy about.

Also I think the reason you willingly give data to Google for Gmail is that you don't have a choice. It's take-it-all, the spam filters and the data mining, or leave it and use ProtonMail, no in-between.

For most people in the world, a reliable $10/month to cover their Netflix subscription or buy them a meal would be something they'd be very happy about.

Your personal data has a cash value only because it can be used to influence future purchasing decisions. It's a simple equation, $x worth of personal data has a y% chance of generating sales worth >$x. It therefore stands to reason that the market value of the personal data of a person who doesn't have as much purchasing power will be less, and may not even cover the cost of stealing and then processing it.

You’re right, I’m rich enough that $5 a month is very little to me. Those $5 a month were something I grasped out of thin air though. If you look at Facebook advertising revenue, they are making less than $3 a year per account.

Would you also sell your data for $3 a year?

> untapped demographics

What kind really? Here's the thing many people touting blockchain forget things like cryptographically secured messengers are not the "must-have" of an average Joe. Is it easy to use and can I talk to my friends? These are the only two things matter. Blockchain, for better or worse, has made things even more cumbersome to use. You have people relying on good-old physically secured hard wallets or paper to secure their keys. This is not a sign of progress.

And frankly about selling your data for money. I think people will find it easier to have some utility like free email or apps in exchange than money through blockchain.

Data about the world ("There is a granite formation here") & personal data ("I am interested in chairs") are two fundamentally different ways to look at this.

So far we've built companies off personal data. But it can really only be used for one thing -- ads. Hence the many adtech startups of dubious social good.

If decentralized big data were to take off, I'd bet all my chips it will be because of data about the world, which enables fundamentally different, more interesting, and more productive use cases. And is bounded by utility instead of personal spending.

Currently you have interested parties either self-generating data, or specialized companies compiling targeted data sets (usually on commission). What you don't have is diffuse, global datasets (e.g. geotagged temperature readings from every cell phone on the planet, right now).

That starts to get interesting.

The "hard" technical problem is the driver of homomorphic encryption. In order to truly own data, I can't allow copies to be made. But in order to make that data valuable, I have to allow arbitrary computation to be performed on it.

«I think the truth about blockchain is that the fad is over. It’s been 9 years. At one point I could buy coffee with bitcoin, today I can’t even turn my remaining crypto into fiat.»

It's pretty clear that being able to transact and store digital money on your device itself (without any 3rd party having control of it) has huge practical applications. As I like to say, "buying coffee" is probably the least interesting thing you can do with Bitcoin.

PS: today it has never been easier to turn your crypto into fiat; there are hundreds of exchanges around the world. Not sure why you'd say otherwise.

Also, you can buy coffee with bitcoin again paying almost 0 fees using lightning network.

I don't think you understand what's really going on. Read more here - https://medium.com/@FEhrsam/blockchain-based-machine-learnin...

I've rarely, if ever, seen such incredibly high readings from my bullshit-o-meter. But yeah, if you believe that "the best data" and "the best algorithms" merely need to come together (and will do so "through competition") to make useful, competent AI (with an actual "I" in there) a reality, I guess that's pretty cool stuff.

vape

Why are we sticking these two concepts together, are they not two fundamentally different things?

What is the synergy by talking about them together other than they are the flavor of the month buzz words?

Good God people, the second or third slide in the article has "AI(ML,data science, big data)"

All of which are literally just applied statistics.

Agreed. Comparing the 'paradigm' of AI and blockchain makes no frigging sense at all. It's actually frightening to see how little VCs have an understanding of the technical details, with blockchain very much still being in the hype zone where reason is not desirable.

Big data isn't just applied statistics, big data is just data.

... and often not especially big, just too big for Excel

IT makes sense to mix AI and blockchain on very specific cases. But dropping names as hype is bad. Probably nothing obvious about AI and blockchain.

This is a very interesting insight into how AI and blockchain are seen by VCs. I agree with IoT and AI being the next big things, but I fail to see how the blockchain fits into the picture.

The author linked the "blockchain is the worst possible database" idea, but this is a consequence of the design requirements, so making a "better performing" blockchain database won't solve the problem.

The ideas about a decentralized market for AI training data are interesting as well, but we have a hard time securing our local databases. If followed through, this will end up with malicious actors and buying up our personal data from these markets in bulk, and to blackmail us.

Maybe the tech will be developed in the next decades, allowing a remote AI to be trained on my data without exposing it to third parties, but I'm sceptical about the technical feasibility, and this will require significant up-front resources to get there.

There is some interesting work done with federated learning https://ai.googleblog.com/2017/04/federated-learning-collabo...

Ok, so 99.5% of the time anything about Blockchain and AI is going to be nonsense. (Aside - this is an interesting application of Bayes law!)

One thing that wasn't, and was actual machine learning and was actual blockchain was the Algorithmia competition[1]. It's pretty interesting how it works, and wasn't an entirely stupid idea (faint praise I know, but most ideas in both AI and blockchain are).

[1] https://blog.algorithmia.com/trustless-machine-learning-cont...

Slightly off-topic, but I was rather hoping for something on the distributed training of neural networks via the blockchain: make the "work" in "proof-of-work" involve backpropagation. Like SETI@home [1] but for training neural networks.

Here's a StackExchange question about this:

https://cs.stackexchange.com/questions/86364/distributing-ne...

Which links to this startup:

https://www.deepbrainchain.org/

A quick skim of their white paper [2] leaves me skeptical.

[1] https://setiathome.berkeley.edu/

[2] https://dlc0uvc36o5o6.cloudfront.net/assets/whitepapers/Deep...

I've been contemplating this idea for a few years now. If you have fully digital currencies that are completely unregulated and can be used programmatically, it seems obvious that the missing puzzle piece to a valuable product is a standardized API economy that can transact via this currency. But I've been talking about this up and down the Blockchain space but I'm still waiting for a proper solution to this.

There is nothing I can put in front of my Rest API that it starts earning me money, no marketplace to advertise that my API accepts WhateverCoin for the calls..

Once that is done, you wouldn't even need AI mostly, it would be sufficient to even just speed up the normal API economy.

https://ost.com/ could of some interest to you... As it will allow companies to create their own branded token and offer everything needed to manage it through a REST API, a wallet... They are currently on alpha phase III focusing on the wallet https://medium.com/ostdotcom/announcing-ost-kit-alpha-phase-...

Personalized AI agents seem promising. They could data-mine user choices and make suggestions for new issues to vote on, along with predicted user vote. If the mining is done before submission to the database the user has a chance to review the implications of the vote before committing.

This could for example suggest research papers for scientists to read, and not suffer from tunnel vision.

People are always sleeping on the originators of the AI + Blockchain movement: Synapse AI (https://synapse.ai/) -- they (along with many others like Goog, Seti, MS, etc) even spoke at the Decentralized AI Summit in Feb. (https://decentralized-ai.com/)

This team even launched their own wallet application like two days ago. They're killing it.

Check out the yellowpaper, it has some really good stuff in it: https://synapse.ai/yellowpaper

... blockchain as the coded constitution for a future AI world: http://blog.stephenwolfram.com/2016/10/computational-law-sym...

is this numerai ? because basically they are pretty much using homomorphic encryption to create AI-compatible data and store it on blockchain.

Their data isn't on the blockchain, is it? I think their ERC20 token (NMR) is logically orthogonal to their homomorphically-encrypted data; they use NMR to incentivize you to do what they want with those datasets, but there's no more fundamental relation between the two.

Every time I think I’ve seen peak HN someone takes it to the next level

Reply on news.ycombinator.com