Wednesday, March 21, 2018

The blockchain is not efficient

Nature mag reports:
Dexter Hadley thinks that artificial intelligence (AI) could do a far better job at detecting breast cancer than doctors do — if screening algorithms could be trained on millions of mammograms. The problem is getting access to such massive quantities of data. Because of privacy laws in many countries, sensitive medical information remains largely off-limits to researchers and technology companies.

So Hadley, a physician and computational biologist at the University of California, San Francisco, is trying a radical solution. He and his colleagues are building a system that allows people to share their medical data with researchers easily and securely — and retain control over it. Their method, which is based on the blockchain technology that underlies the cryptocurrency Bitcoin, will soon be put to the test. By May, Hadley and his colleagues will launch a study to train their AI algorithm to detect cancer using mammograms that they hope to obtain from between three million and five million US women.

The team joins a growing number of academic scientists and start-ups who are using blockchain to make sharing medical scans, hospital records and genetic data more attractive — and more efficient. Some projects will even pay people to use their information. The ultimate goal of many teams is to train AI algorithms on the data they solicit using the blockchain systems.
No the blockchain is not efficient, and does not offer any advantage to a project like this.

The blockchain is surely the least efficient algorithm ever widely deployed. Today it consumes energy equivalent to the usage of a small country, to maintain what would otherwise be a fairly trivial database.

It appears that someone got some grant money by adding some fashionable buzzwords: AI, blockchain, women's health.

The blockchain does not offer any confidentiality, or give patients any control over their data. This is all a big scam. It is amazing that a leading science journal could be so gullible.


  1. Stupidest bubble ever. You can easily have a gold-backed currency but Bitcoin people claim they want decentralization. Why? They don't trust Brinks with a prepaid card? It turns out that the blockchain makes all of your transactions public. Since the criminals found out it isn't even private, there is no more reason to use services like Bitcoin. People in inflationary countries like Venezuela can’t even use it because the transactions fees have been so high. It also gives no scarcity because you can fork it all you want.

    Here is the overall hype: "In this world every agreement, every process, every task, and every payment would have a digital record and signature that could be identified, validated, stored, and shared. Intermediaries like lawyers, brokers, and bankers might no longer be necessary. Individuals, organizations, machines, and algorithms would freely transact and interact with one another with little friction. This is the immense potential of blockchain."

    The problem is that a centralized system can work much cheaper to do the same exact things. The blockchain only has the ‘benefit’ of decentralization. There is no other advantage. The problem in our larger economy is that of regulation & monopoly, which is especially true of banking. The fundamental cost of the service is trivial. Hell, if the benefit was so great, the government could run it as a public service.

    This is similar to Uber. It's an app you can make in a weekend using public APIs for GPS and maps but it takes 20% of fares! It should cost a one-time sign-up fee and 20 bucks a year!

  2. Interesting. In law school, they taught us a blanket rule: If "your" data (on a PC) differs from the bank's data (on some big iron), the bank wins. Blockchain solves that one.

    I'm glad Roger qualified one of his statements with the wording "projects like this". I'm almost finished writing a new patent application, and I figured I'd better explore the use of blockchain due to best mode requirement. If the limiting factor for my algorithmic application is real-time write speed, with reads being performed only lazily, post hoc offline, and auditability is critical, and ownership of the writes to the blockchain might be both valuable and less contestable because of the blockchain, I'm convinced I need to include it as an option within a dependent claim.

  3. The patent best mode requirement is obsolete. Altho still on the book, it has been neutered by patent reform.

  4. What can I say? I'm an incorrigible formalist. Neutering by "reform" sounds like something I need to mount a bulwark against.