Hacker News Reader

Comment by mrweiner | Hacker News Reader

A company called Vectorspace Biosciences/Vectorspace AI’s revenue and technology model is based largely on crypto/blockchain. They’re focused on helping to facilitate scientific discovery through what’s essentially AI data analysis, so knowing when and why data changed is critical. Blockchain and their token are used as part of their dataset provenance and security pipeline to help keep track of dataset modifications and history. It also acts as a licensing mechanism, so that tiered API access is granted based on how much of the token is held in the user’s wallet.

Disclaimer: I’m an investor.

View on HN

This sounds like a persiflage. But I have the feeling it's not?

How do they use blockchain? Can you tell it without jargon?

Sorry, didn’t mean to make it jargony.

The licensing API key part is fairly straightforward. A customer holds X amount of the token in their crypto wallet, and then the wallet address is used as an API key. Different service levels, etc, can be granted based on the amount of token held in the wallet. And API access, customer datasets, etc, can be associated with the wallet-as-an-API-key.

By provenance, I just mean that somebody who’s leveraging a dataset can see the exact history of where the data came from, when it was manipulated, etc. I believe the way it works is that each time data in a dataset is manipulated, a corresponding hash of the dataset is generated and stored on-chain. So if a you had a dataset representing the relationship between some proteins, and a study was published that provided additional data regarding one of those proteins and potential relationships, then the dataset might be updated to reflect the new information, and a hash of the dataset would be created to represent the updated dataset at that point in time. Like a git commit hash. Then they do some proprietary stuff that associates the dataset with the customer wallet address noted above, and that is stored on the blockchain in a transaction using their token.

Does that make sense?

Thanks for the explanation.

When you say "see the exact history of where the data came from", what does that mean?

When I publish the string "Dogs have two ears" on a blockchain, that notarizes that I published this string at a certain time or earlier. But how does it tell where the data comes from?

If I understand your question, it’s just a part of their data pipeline. So a new paper is published, they ingest it, and then their language modeling/analysis/etc see that, and your dataset is updated. Presumably what caused the change would be discoverable via the hash.

I should be able to give a better answer, here, but I’m not sure that I can. Not sure that this helps, but this is the non-technical overview currently on their site: https://vectorspacebio.science/technology/

I think all that lingo just makes things more muddy here.

The "dataset" is a string, right? A sequence of characters.

Like "Dogs have two ears".

But what does it mean when you say "know where the data came from"?

If someone publishes that string on a blockchain, you know they published it. Because they signed it. And you know when. Because the blockchain timesteamped it. Fine. But "where it came from"? What does that mean?

So let’s say 1000 scientific papers were published, and we want to see how two proteins relate to one another based on those papers. When I say underlying data or “where the data came from,” I’m referring to the papers.

Those 1000 papers would be run through the data analysis pipeline to generate a “dataset” representing the relationships. For instance, a CSV file of the relationship information.

That dataset is what is hashed, and that hash is the “string” that is published on-chain.

It would be like if you had a bunch of papers about how many ears animals have, whether they have fur, where they’re from, etc, and you want to find out how those things relate to one-another. The info about ears and fur is the “data that changed” and the “dataset” is essentially vectors for how the information relates to each other. Then that dataset is hashed, etc, and that hash is what’s stored on chain. Then when another paper is added, the relationships csv changes, and that is re-hashed and published.

It’s hard to describe this without some jargon, just because of what we are discussing. I left another comment before with a diagram that might or might not help.

I think you are cought up in trying to make things sound grandiose.

Why "1000 scientific papers"? Why not 500? 200? 17? 3? 2? How many are needed to make your point? Why "scientific"? Does it only work for "scientific" papers? Why "papers"? Does it not work for strings in general?

Same for "proteins", "underlying data", "data analysis pipeline" and all the other concepts you introduce.

If you comb through all of this and remove everything not necessary to explain the underlying idea, maybe you get to something and maybe not. I don't know.

Imagine we understood blockchains already and someone wants to explain the payment functionality of Bitcoin with a garden of wonderful terms like you do. "Thousands of virtual banks connected to millions of owners. Payments flowing through this massive global network instantly and nearly free, guaranteed by scientific cryptographic proof". It would tell us nothing.

What we need is:

With a blockchain, we can create an ordered list of messages like "I give coin 17 to Bob. Signed: Alice"

That explains everything we need to know to understand how payments are possible on top of a blockchain.

That's the type of statement I would dig for if you want to know if there is actually something useful about a technology.

Because this whole thread is about how a specific company is using blockchain in the context of their business. The business context is relevant. This is a biosciences company and their tools are meant to be used in the context of scientific discovery.

I included a more basic answer in my last response that followed your concept of dogs with ears. Why wasn’t that abstraction simple enough to avoid this criticism?

This is about as simple as I can make this.

1. Collect source data 2. Run data analysis 3. Processed dataset is generated 4. Processed dataset is hashed 5. Hash is combined with the API wallet address and rehashed 6. That hash is stored on chain and with a transaction involving their token 7. Data from step 1 changes, repeat

Why 1000 papers? Because…I picked 1000? Sure, choose 10. Whatever. Does that really affect whether you can understand what I’m saying?

Why data analysis pipeline? Because the graphic that I sent you has its literal first step as “language modeling pipeline” and I was trying to avoid getting flamed for mentioning language modeling. https://ibb.co/5rBnTsv Substitute “analyze data” if you want to. But it’s silly that I can’t use the word “pipeline” on a technology forum.

Why underlying data? Because I need to differentiate between the raw, pre-processed data and the post-processed data, the latter of which is actually being hashed. I figure this was important since the whole point of this thread was how blockchain is being used.

Why proteins? Because it’s a biosciences company with a product called The Protein-Protein Interaction Network (PPIN) API, and that’s a common example they use when explaining what they do. From https://vectorspacebio.science/cmdb, A REST-based API which can be used to generate a multi-level graph network from a correlation matrix dataset. The graph network represents context-dependent known and hidden relationships between proteins, pathways, drug compounds and molecular sequences.

Did I have to choose proteins? I guess not. They have products to be used in financial markets as well that use the same technology. Technically this can be implemented in any domain. I just picked a particular one.

Why scientific papers and not just papers? Same response as above.

Not sure how much further we are going to get with this conversation, to be honest. But I’m trying to operate in good faith.

    1. Collect source data
    2. Run data analysis
    3. Processed dataset is generated
    4. Processed dataset is hashed
    5. Hash is combined with the API wallet address and rehashed
    6. That hash is stored on chain and with a transaction involving their token   
    7. Data from step 1 changes, repeat

It is getting more structured. I like that.

The way I read this:

    - A company created a dataset and updates it from time to time.
    - For each update, they put a hash of the dataset on a blockchain.

Ok. And what is the use case? What would be different if they did not publish the hashes on a blockchain?

> A company created a dataset and updates it from time to time.

I suppose update frequency doesn’t matter from the perspective of just conveying how it works, but it could be updated frequently. For instance, any time a paper is published, or if the customer provided a data stream, etc.

> Ok. And what is the use case? What would be different if they did not publish the hashes on a blockchain?

The idea of blockchain is “don’t trust, verify,” right? Imagine you’re a researcher and you’re using this company’s services. Having the hashes on-chain allows you to verify the integrity of the data over time. Having it on-chain removes attack vectors associated with e.g. a standard database. Similarly, all of Vectorspace’s tooling for various services benefit from the same assurances that the data is sound. So like I said initially: data provenance.

"Verify" in which way?

What would be the simplest example of "using this company’s services"? Maybe an API call like this?

    GET /animals/dogs/ears/count

And that returns "2"?

Now you know that the company claims that dogs have 2 ears.

How does the fact that they published a hash on a blockchain allow you to "verify" that statement?

At best, you can say "The company claims that dogs have 2 ears and they did so 3 months ago already", right?

But the hash does not help you verify the claim on how many ears a dog has.

Also, here’s where some of my reference comes from, if a diagram is helpful. Downloaded from telegram and uploaded here: https://ibb.co/5rBnTsv

Reply on news.ycombinator.com