I and many others have been using food markets as an analogy for digital data marketplaces. It’s decent if you want to convey to the laymen that you’re planning to buy and sell data. However, as soon as you start detailing the desirable properties and guarantees of such marketplaces, the analogy falls apart.
In the past, I enumerated some desiderata for data marketplaces, and explained that they might not make sense, doubly so and why we might want them anyway. If you haven’t yet, reading through those articles first would set you up nicely for the rest of this post.
But to the point.
More than one reason
When you buy physical goods, several conditions hold:
- you exchange your currency,
- for an item from the vendor,
- the item and the ownership of it are bundled.
Sure, there are exceptions to this, e.g. when you get a mortgage to buy a property. Nevertheless, most transactions match this template — in particular, if you consider the food market analogy.
Don’t get me wrong, there are data brokers that operate like food markets but as I explained in the past — they often don’t respect the privacy of the data subjects, are susceptible to piracy, and the transactions are irrevocable.
On the other hand, a modern data marketplaces would have several niceties:
- when you sell the data, you don’t actually hand it over but licence it instead — you rent it in a way;
- the buyer is not allowed to work with the data as they please. Instead, they’re permitted to perform only a set of operations, e.g. compute averages, or train a machine learning model;
- all transactions must meet some minimum volume threshold — you can’t buy one image. Consequently, you can’t view your data in high resolution.
These might seem quite random if you didn’t spend much time thinking about data economy. Effectively, we’re trying to ensure that the privacy of the data (and its subject) are protected, and that the buyer cannot resell/share the data with other parties. A food market doesn’t convey these additional criteria implicitly, so we have to do it explicitly. At which point, the food market stops being a food market.
You might say “No reasonable marketplace could meet these criteria!” — and yet, we have great precedent for it.
Weird modern markets
As it turns out, we already have a bunch of quasi-marketplaces that satisfy some of these requirements.
Application stores. For instance, Steam, Google Play, PlayStation store, to name but a few, do not sell you the apps that you download. They provide you with a perpetual yet revocable licence to consume the content that you paid for. However, if the developer decides to pull the app, the billing changes from a one-off purchase to a subscription, or you cheat in a game and they ban you — you will lose the access. The one that personally annoys me is when movies and TV shows disappear from one streaming service and appear on some other.
These aren’t once in a blue moon situations — they occur all the time.
Financial markets. When you buy stocks, or ETFs, or other instruments, you’re telling your broker to put a number/uuid next to your name and handle the operations on your behalf. As far as I know, there doesn’t exist a mechanism that allows you to buy stock without interfacing with a licenced intermediary. Also, most of the time, you can’t just take your stock and move it to another brokerage without paying a fee.
Then, there’s the volume — minimum order volume, minimum transaction volume, minimum order size — put a lower bar on how much you need to buy. To top it off, most financial instruments have special mechanisms associated with them — e.g. a CFD (which was at the centre of the crude oil craze during COVID).
Non-fungible tokens. In essence, non-fungible tokens (NFTs) are identifiers stored on the blockchain that attest the ownership and authenticity of a piece of content. They were at the forefront of the blockchain hype, and thus got a lot of bad rap, rightfully so (e.g. the Bored Apes crap).
There exist many related technologies that leverage distributed ledgers to create special, cryptographic objects that restrict what kind of operations you can apply to the data. Private data objects are a good example of this if you want to read up more. Standalone trusted execution environments would also work in some limited cases.
Mental models help
Our three original conditions don’t seem so far-fetched anymore. In one form or another, they’re already present in modern markets and technology (for better or worse — looking at you disappearing movies).
But why are food markets a bad analogy; can’t we just explain these additional considerations? Well, no — it’s about the mental models.
When we equip a particular mental model or use an analogy, we assume a particular perspective, and take on a set of constraints and associations. It’s a convenient shorthand that allows us to reason quickly about otherwise complicated topics. Unfortunately, it requires everyone not only to understand but also agree on the obvious. Otherwise, false analogies can be deceptive; even if we don’t mean ill will.
At the end of the day, I haven’t figured out what the best analogy is yet. To me, financial markets are better because they come with a fuzzy notion of ownership and intermediaries already built in. There’s less to explain and the jump is easier. This works only if someone has some understanding of how they work. If I have to explain financial markets and instruments, and then digital marketplaces, it’s a failure and a waste of time.
On the other hand, literally everyone knows how food markets work. Across the entire world. Young and old. Regardless of their education. Go figure.