5 ms·

I had a similar reaction, since it's not like these matching algorithms do any "learning" (as far as I know). And you are right that the credit here goes largel

by majos 8y ago

I had a similar reaction, since it's not like these matching algorithms do any "learning" (as far as I know). And you are right that the credit here goes largely to economists. 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.

peoplewrong 8y ago
I agree this isn't really "AI" but this project required a lot of algorithm tuning and development.... Here are some good slides. http://www.cs.cmu.edu/~arielpro/15896s16/slides/896s16-14.pdf 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..........
currymj 8y ago
there's a nonzero amount of genuine AI/machine learning going on in the research end of things right now in this field. 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".
YeGoblynQueenne 8y ago
>> I had a similar reaction, since it's not like these matching algorithms do any "learning" (as far as I know). 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.
- YeGoblynQueenne 8y ago
  Oh, right. I see in peoplewrong's link that the system actually uses MIP, which is to say, Mixed Integer Programming, a type of Linear Progamming: https://www.gurobi.com/resources/getting-started/mip-basics https://www.gurobi.com/resources/getting-started/mip-basics https://en.wikipedia.org/wiki/Integer_programming https://en.wikipedia.org/wiki/Integer_programming https://en.wikipedia.org/wiki/Linear_programming https://en.wikipedia.org/wiki/Linear_programming
- peoplewrong 8y ago
  > 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). 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/0837/80a157a7f8ebfa44dce0bc1e3e5035a5cdee.pdf 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??
  - YeGoblynQueenne 8y ago
    Cheers. Just as you were posting this I read your comment and noticed, in the link, the reference to MIP. However, it does sound like early matching systems where simply hand-crafted rule-bases encoding experts' decision-making process, so expert systems: 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.

by majos 8y ago

peoplewrong 8y ago

I agree this isn't really "AI" but this project required a lot of algorithm tuning and development.... Here are some good slides. http://www.cs.cmu.edu/~arielpro/15896s16/slides/896s16-14.pdf 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..........

currymj 8y ago

there's a nonzero amount of genuine AI/machine learning going on in the research end of things right now in this field. 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".

YeGoblynQueenne 8y ago

>> I had a similar reaction, since it's not like these matching algorithms do any "learning" (as far as I know). 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.

YeGoblynQueenne 8y ago
Oh, right. I see in peoplewrong's link that the system actually uses MIP, which is to say, Mixed Integer Programming, a type of Linear Progamming: https://www.gurobi.com/resources/getting-started/mip-basics https://www.gurobi.com/resources/getting-started/mip-basics https://en.wikipedia.org/wiki/Integer_programming https://en.wikipedia.org/wiki/Integer_programming https://en.wikipedia.org/wiki/Linear_programming https://en.wikipedia.org/wiki/Linear_programming
peoplewrong 8y ago
> 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). 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/0837/80a157a7f8ebfa44dce0bc1e3e5035a5cdee.pdf 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??
- YeGoblynQueenne 8y ago
  Cheers. Just as you were posting this I read your comment and noticed, in the link, the reference to MIP. However, it does sound like early matching systems where simply hand-crafted rule-bases encoding experts' decision-making process, so expert systems: 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.