How AI changed organ donation in the US
qz.com
qz.com
Another horrible example was a patient with fulminant liver failure secondary to Hep C. Most people in Australia contract Hep C through IV Drugs or sexual activity, but during the 80s people contracted it through contaminated blood. This patient had been at the top of the transplant list 3 times but because he developed spontaneous Bacterial Peritonitis each time he was ineligible and dropped back to the bottom. I believe he died not long after I took his history. I’m so pleased we have free treatment for Hep C patients in Australia now.
Organ donation is a heartbreaking issue and the sample conversations of the God Committee allocating access to the first dialysis treatment is revealing (even though they’re discussing access to treatment and not transplant)
For those wondering, as I didn’t see it mentioned in the article, dialysis treatment costs approx $20-30k AUD a year (at least in Australia). Average survival is 10 years from initiation of treatment.
A kidney transplant costs $30-40k. In Australia there are access schemes whereby patients who may not otherwise qualify (ie obesity) have been given gastric sleeve procedures by the healthcare system in order to be eligible for a transplant, on the economic basis that $200k in dialysis is more than the $50k of an elective surgery and then transplant. Usually in these situations the donor is a family member.
The use of matching algorithms as described in this article is so valuable in terms of lives saved, but ultimately no substitute for what we need: the ability to grow new organs.
http://freakonomics.com/podcast/make-me-a-match-a-new-freako...
But to call it an triumph of AI is stealing credit from the economists. It was an economist who came up with the idea and the field of study, market design, is fascinating, especially to those who are into startups and disruption.
But the article does point out that we do a lot of goalpost-shifting when it comes to defining "AI", so "algorithm solving what we thought was a very human task" plausibly falls under that umbrella. I think that's the rhetorical point of using that word.
Or maybe they just figured "AI" drives clicks.
Here are some good slides.
http://www.cs.cmu.edu/~arielpro/15896s16/slides/896s16-14.pd...
The "credit" should go to the research community that worked on this. which is a mix of economists and computer scientists..........
but you're right, as is currently deployed, a better title would be "how mathematical optimization changed organ donation". or "how letting computer scientists attack a problem changed organ donation".
Learning is not a criterion for AI/ not AI. The kind of system described (though only very vaguely) in the article is a rule-based system, possibly an expert system. These hand-crafted systems were the norm in AI up until the '90s or so and are still in widespread use in various domains (e.g. in finances, or air travel bookings etc).
In fact, machine learning took off largely as an effort to avoid having to hand-craft rulebases and elicit expert knowledge from domain experts. You can see this very clearly in decision-tree learners: these basically learn the rules that a human would be creating by hand.
It's also possible that the system described in the article is using some kind of resource allocation algorithm, like a linear programming algorithm, or some such. Those are actually optimisation algorithms, so they are easier to recognise as AI by today's standards (i.e. by the standards of the last 5 years or so of press releases by Google :).
In any case- the majority of AI, historically, has nothing to do with (machine) learning and instead focuses on reasoning in complex domains.
https://www.gurobi.com/resources/getting-started/mip-basics
this is wrong at least for UNOS (united network for organ sharing)-which is the main system in the article-the current computer program used to find matches is an integer linear program. I think it is an interesting question if some kind of rule based system could replicate their ability to find the matches. I tend to think not.
https://pdfs.semanticscholar.org/0837/80a157a7f8ebfa44dce0bc...
page 16 "To our knowledge, there is no solver that would scale to the nationwide steady-state size— including the CMU solver used by UNOS. This solver is based on the work of Abraham et al. (2007), with enhancements and generalizations by Dickerson and Sandholm, and uses integer linear programming (IP) with one decision variable for each cycle no longer than L (in practice, L = 3) and constraints that state that accepted cycles are vertex disjoint. With specialized branch-andprice IP solving techniques, Abraham et al. (2007) were able to solve the (3-cycle, no chains, deterministic) problem at the projected steady-state nationwide scale of 10,000 patients"
edit: wait wait this comment could be out of date. The google scholar for this paper says 2018 but the paper says 2013??
One night in 2000, tired of delivering the heartbreaking news to patients and their loved ones that no suitable kidney could be found, a US nephrologist named Michael Rees lugged home several crates of files and spent the next few hours scrutinizing blood, antibody, and tissue data, and comparing patient charts. The work was mentally grueling. Eventually, he realized he had no viable matches—but also, that if the pool were bigger, pairs could be made. Working with his father Alan Rees, a computer scientist, Michael Rees created a simple computer program that did the work of pairing up donors and recipients, introducing AI to the matching process.
but on the topic of experts check out this paper by the same research group. This is much more of an 'AI' paper that does include expert domain knowledge and learning
"FutureMatch: Combining Human Value Judgments and Machine Learning to Match in Dynamic Environments" from AAI2015
Chill! Your refs are up-to-date :)
There's two versions of the paper, one published in the Proceedings of the ACM Conference on Electronic Commerce, in 2013, the other published in 2018, in, er, Informs PubsOnline (??) in 2018. The latter appears to be an online journal- sorry if I messed up the name, I don't know the journal.
“This series explores the promise of AI to personalize, democratize, and advance medicine—and the dangers of letting machines make decisions.” You have to work with what you have, I guess.
Like I say in another comment, most of AI is not about machine learning, but rather about reasoning in complex domains, very often using hand-crafted procedures.
For example, most of the AI systems of the '80s and '90s, which were in fact hailed as great successes of AI, were of the expert system variety: in short, a large, hand-crafted database of production rules elicited from domain experts, coupled with an inference procedure.
For instance, an early expert system was MYCIN, used to diagnose bacterial infections. It was one of the earliest AI systems to perform at least as good, and even better than, experts:
[1] http://www.cs.cmu.edu/~arielpro/15896s16/slides/896s16-14.pd...
[2] https://www.cs.cmu.edu/~sandholm/futurematch.aaai15%20with%2...