🪴Knowledge Assets and the Infrastructure Needed for Them to Prosper

Imagine if our knowledge could generate revenues for us...

In approaching the problem of decentralised AI services, we made three key assumptions:

1. A decentralised AI system must be technologically competitive - the models and apps provided must match the level of the industry-leading performers;

2. A decentralised AI system must be user-friendly - OpenAI revolutionised the market with its simple interface and users now expect at least this level of useability;

3. A decentralised AI system must be allow participants to profit from their contributions - a much harder criterion to meet, but one that must be met if the system is to enjoy any measure of longevity.

Simply put, to persuade users of existing centralised systems to switch and to acquire new users, a decentralised AI system cannot hope to thrive based upon the mere fact of its being decentralised. It must also perform at least as well as the centralised equivalent and also incentivise the participation of contributors (since these are not salaried staff, as in the case of the large AI companies).

Furthermore, these three components are highly integrated. In order to become a core distribution platform for the latest and best models it is necessary to provide sufficient economic incentive for the designers to make them available via our platform (particularly given the high compute cost involved), but this incentive will not materialise unless the platform as a whole is both useable and - back to point one - incorporates state of the art technology.

In other words, no matter how technologically advanced your decentralised AI solution, if it is inaccessible or provides no financial incentives for participation, it will eventually fail.

To solve this, we set out to solve three key problems:

1. The security problem: if there is no easy way to allow selective access to an asset to only those who have paid - as tends to be the case with information - it is very difficult to turn it into a mass market product. Either it must be entirely private or entirely open-access. Traditional publishers have tried many methods to safeguard their revenue streams and prevent piracy since the advent of the internet, and are still fighting a losing rearguard action. For individuals without their legal and economic clout, the same battle is almost impossible. This results in:

2. The money problem: If it is impossible to implement selective access to assets, it is thus impossible to monetise them. The absence of channels by which creators can easily monetise their work reduces or eliminates its economic value - a product has no price if there is no way to get it to market. If, when a product finally makes it to market, there is every chance that it will be widely pirated, the incentives to work on it drop even further. This, in turn, creates:

3. The connectivity problem. If no money is to be made from providing access to an asset, no one will make any effort to enable this access. Consequently, there are currently 500,000 models on Hugging Face, but the vast majority are simply inert weights, unused on account of the difficulty of running them. Currently there is no accessible marketplace allowing creators specialising in a particular type of AI knowledge asset - whether models, data or apps - to connect and sell their work to others. We believe that this will change when KIP's token-gating measures allow asset owners to implement selective paid access, thereby giving them an incentive to improve access to their work.

Our solution is a corresponding three-layer stack:

1. The ownership layer: each asset functions in web2, but is "wrapped" in an ERC-3525 semi-fungible token, which serves as proof of ownership but also as an access-gate mechanism. This, in turn, lets us build:

2. The settlement layer: this allows user interactions with apps to be reconciled in a secure and transparent manner on-chain. These credits let them interact with AI apps, and - via these apps - models and datasets. Because each interaction is recorded on-chain, it is possible to calculate exactly how much of the revenue from each app should go to each of the creators whose work contributed to it, and to redistribute this revenue in the form of drawing rights upon the pool. This then implies the development of:

3. The application layer: if it is possible to easily earn revenues from assets, creators are incentivised to create easy access pathways to their work via the KIP standard APIs, as well as to build front-end code and promote their products. Apps, models and datasets are intgrated and interact with one another in web2 giving creators and users complete freedom of choice when it comes to the models, data and app formats they use, while simultaneously recording the interactions via a set of transparent and anonymous web3 contracts. In other words, it is possible to know exactly how many people have used a given app, and which AI models and data were used to produce it.