Why apply Fred Ehrsam’s concept of blockchain-based machine learning marketplaces to healthcare?

Medium Post by Ivan Galanin and Jonny Dubowsky for YuMatch:

In a March 13th MEDIUM blog, Fred Ehrsam laid out a vision for blockchain-based machine learning marketplaces. He noted that technology allowing data and models to yield utility without being disclosed, enabled building meta-models that were more effective than silo’d AI approaches. Ehrsam foresaw this approach being implemented in finance, where the advantage of clear-cut outcomes yield immediate gains which can be allocated amongst contributors.

YuMatch is an enabling framework that applies Ehrsam’s concept to healthcare. This Medium post is the first in a series of blog posts that will introduce the YuMatch framework, while providing context into the use of blockchain protocols, and cryptographic primitives that we have found useful in fulfilling our vision of a medical knowledge commons.

Blockchain AI Marketplaces for Healthcare Wisdom

The rationale for blockchain AI marketplaces for healthcare wisdom is unusually strong. Many have pointed out how blockchain technology can resolve the privacy and security issues related to the exchange of patient data, and held out the concept of financial incentives for patients to supply health records or genetic data. The corollary becomes a proprietary token-based healthcare data marketplace, sometimes with a prescribed AI component. More than a dozen such marketplaces have been proposed and are being built.

We propose a different approach based on interlocking principles:

1. Individuals need maximum freedom to unlock their data resources. Individuals, not electronic health record systems, are the primary determinant of the breadth (e.g., how many different forms of data are collected) and longitude (i.e., for how long) of the health data that is available. The optimal data supply is achieved by setting the cost of sharing as low as possible. This means not prescribing any one standard for how data is collected, formatted or shared. It also means that the individual has the option of not doing any data processing. He or she should be able to throw their data in with the same ease as they attach files or subscribe to Twitter channels and have someone else ready this data for use.
2. Enable knowledge workers to fill the gaps in the absence of universal standards. In a microcosm of the coding industry, Flatiron Health employs hundreds of nurses to process patient records. Treating coders as a cost center sets a ceiling on the market incentives for improving the coding process. Flipping the paradigm and giving knowledge workers a stake in the outcome of their work product creates a much larger market for data tagging and processing tools and will lead to exponential improvement in this domain.
3. Use the many pieces that already exist. Rather than reinvent a blockchain equivalent for every wheel in the healthcare system, our approach seeks to incorporate the pieces that already exist. Almost every piece needed to make the distributed healthcare marketplace work exists in a Web 2.0 form. Some of the pieces are open-source (e.g., Quantified Self as a community that curates individual data, Open mHealth as a source of data tagging code) and readily incorporable. Most of the pieces are proprietary (e.g., WEGO Health for curation, Apixio to facilitate patient data coding) and will be incorporated when the incentives for joining and the cost of being left out become apparent.
4. Give everyone a stake. Open communities of contributors own and govern the ecosystems. We anticipate that contributors will comprise:

1. 1. Data contributors, either individuals or institutions.
   2. Knowledge workers, who process data into more usable forms.
   3. Technology contributors, who provide enabling code, smart contract templates, plug-ins and other open source software artifacts to facilitate participation by technical and non-technical people alike.
   4. Computer Network Support/Infastructure aka Miners. Extending the network model of Bitcoin, Ethereum, et al, Token Curated Registries (TCR’s) will depend on miners, contributors, who make processing power available, run the blockchain protocols, and in the future, we will likely see a knowledge worker mining pool emerge to handle coordination of optimization dynamics across multiple blockchains. Jonny Dubowsky will publish a mult-part series on hybrid blockchain integrations for TCR’s in the coming week, as part of YuMatch’s ongoing series.
   5. AI/Machine Learning model contributors, who develop algorithmic models with the ecosystem resources. AI models will be shared and licensed through YuMatch’s marketplace integration feature, which will connect stakeholders to all available data sharing marketplaces on the traditional web, as well as on Web3.0 and blockchain platforms.
   6. Curators, who evaluate all the components of the ecosystem. Curators will create turn-key templates and plug-in libraries that keep the complexity of these new and powerful protocols “behind the curtain”, for users who just want something that works, but more advanced users will find the fine-grained preference and customization choices they demand, to combine and share resources and best practices, while maintaining control over their privacy and security.
2. Plan for pluralities and evolution. Ecosystems will draw different types of data (e.g., sequencing, diet records, microbiome results, etc) and have different requirements for data processing, computing, model building, etc. The requirements for a single ecosystem will change over time. For example, initially it may be more important to attract data and process it; later it may be more important to attract curation and model building. New ecosystems can branch off of existing ones to attract different resources and alternatively combine to experience synergies. For example, an ecosystem focused on microbiome data may combine with an ecosystem focused on nutrition in order to create novel models. Smart contracts binding users and rewarding contributors will carry over into new ecosystems to prevent freeloading.
3. Maximally exploit blockchain resources. Recently, the number of Ethereum wallets exceeded Bitcoin wallets, serving as a reminder that sources of innovation cannot be predicted. The platform seeds we are sowing are designed to work across all major blockchains and port to new resources as they become available.
4. Free ecosystems from proprietary tokens and time-bomb token economics. In the absence of markets that freely set value, proprietary tokens are forms of compulsion, whose primary purpose is to lock in users. Effective ecosystems will need to modulate incentives as they grow and would be held back by economic models set a priori.
5. Democratize AI. The complexities of blockchain technology and AI are such that arguably there are fewer than one thousand technologists in the world that master the intersection of both domains. Consider how much utility would be constrained if there were only one thousand people capable of using a spreadsheet. YuMatch promotes the democratization of AI on both software and hardware fronts; creating software on-ramps and templates to make algorithm development accessible to game coders on the one hand, and by planning for the next wave of low-cost edge computing on the other.
6. Curate every element. Ehrsam advises that the best marketplaces will have clear-cut outcomes for AI models. The best way to deliver such outcomes for healthcare ecosystems is through curation. Curators will earn opportunities to evaluate the efficacy of models through N of 1 or other trials. The results of the curation will establish the utility for products and services emerging from the ecosystems (including the potential to validate a wave of new digital therapeutics). While this is the most compelling case for curation, all aspects can and should be subjected to curation, including the performance of knowledge-workers, the suitability of enabling code, etc. YuMatch as an ecosystem enabling project is going through the process of curation as we speak.
7. Though the main elements of the YuMatch are functional, specific additional capabilities are being funded through development grant proposals to several leading blockchain technology companies, submitted in collaboration with Sense Collective and My Personal Therapeutics. As such, companies interested in exploiting the platform will receive trusted 3rd party assessment of its suitability for different blockchain environments.