Accepting Goodhart’s Law in Academia - Tokyo 2024

okay next up on the agenda we have accepting good heart's law and Academia on how token rewards can improve upon biblo metric um this talk will be given by Patrick Joyce co-founder of research Hub and uh yeah pass it off to you Patrick awesome well uh yeah my name is Pat and I'm here today to talk about the role of good heart's law in Academia I'm one of the uh co-founders of research Hub and uh to get started I'm going to share a little bit about my background I'm a huge science nerd I was very lucky in undergrad I got to study at John's uh looking at tumor biology specifically the epigenetics of cancer stem cells I fell in love with academic science decided to start a PhD went to Boston University to study molecular biology and cell biology but quickly found that not all labs are quite as wonderful as a well-funded Hopkins lab there's a lot of issues in the way that we fund and publish academic science that causes a low quality of life for the average researcher and also a low quality of results so I decided to actually drop out of that PhD program and start medical school at Georgetown and also found out that clinical medicine is not a whole lot better so then uh decided to do something about it and create a project create token incentives to reward healthy research behaviors so just to kind of like zero Wi on sort of the state of Academia today here are three solid headlines that I think describe like some really big problems that we're dealing with in the world um there's the replication crisis here did a great job of highlighting this at the beginning of the a summit today um between 50 to 85% of all science can't be independently replicated this is a huge problem there's kind of the Advent of paper mills there's tens of millions of dollars every year being spent on like spam papers that don't serve any purpose except for bolstering one's resume and then finally we actually have like a kind of tremendous amount of scientific fraud that exists in the world if anyone here is familiar with the Dilemma around amid beta uh hypothesis for the pathology of Alzheimer's it actually was created out of a fraudulent uh like PhD thesis where someone did photoshop on figure in order to birth like a whole field of research where billions of dollars have been spent chasing something that doesn't work due the fraud um one of the results of this just in the last couple of years is someone who had risen to the top of science actually the president of Stanford was forced to resign because of their role in producing fraudulent research around this hypothesis so I think this slide does a good job of summing up what we're dealing with today um the Lan it in 2016 estimated that 85% of all research funding is wasted and when you think about about it it's not really that unreasonable if 50% of results can't be replicated that's 50% down the drain right away and then more than 50% of clinical trials actually aren't fully reported because they don't produce positive results that will refect uh well on the publishing academics career so when you look at this uh back in 2016 this was $170 billion annually that was wasted or about the GDP of quit so why does this happen I think it makes sense to zoom out for a second and kind of look at the economy of science as a whole and how Capital flows through the system so uh you start with wealthy patrons of science this can be public entities nonprofits or wealthy patrons um what they do is they create requests for proposals where scientists apply for Grants the problem is that it's really difficult to decide who to give your funding to um there's really it's hard to be objective and determine scientific quality just as like an example if you look at the Kern and the heliocentric model of the universe it kind of started out and people thought that it was a really bad theory he actually you know was persecuted pretty significantly this happens all the time in science often times contrarian ideas are kind of like rejected at the beginning but then over time they're found to be true so as a fun it's actually really hard to know who to give your money to the thing funders use now in order to help determine quality of scientists is something called bibl metrics or citation-based metrics the idea here is that you look at a scientist and you judge how productive they are how often do they publish papers also you look at their impact when they publish papers are they cited by other scientists here's a very common uh bibliometric that's used in making grant funding decisions it's called impact factor this is basically a measure of the quality of a journal it's the average number of citations uh produced by papers within a journal over a 2-year period so if you publish a nature which is a very high impact Factor journal and a funder sees that on your resume they'll think you're a pretty good scientist here's another very common bibl metric called H Index this is kind of like an average of the number of citations that you generate per publication over your career if you have an highh index you're also very likely to Perce or be perceived as a quality Scientist by funders if you look at Google Scholar here uh which is kind of like the number one search engine for scientific research they have citations front and center on average researchers profile so we have our friend Albert Einstein here um right at the top of the page you see the number of citations and the H index of this scientist so this is a big deal and scientists spend a lot of their career thinking about the sen or citations they're generating and how to maximize their own H index the problem is that scientists are really smart when they start to kind of optimize for these citation-based metrics they prioritize them over producing high quality science I'm sure lots of folks here amiliar with good heart's law when a measure becomes a Target it ceases to be a good measure the use of citation based metrics is to determine scientific quality so we're trying to measure quality but unfortunately it's become a target for behavior and so it's not really that good of a measure anymore because scientists are optimizing for citations so when you're optimizing for citations what kind of incentives this is create when you want to maximize your productivity or number of papers that you're publishing you kind of have to be a little wary of your competition rather than sharing with everyone across the world who's working in the same field as you you kind of guard your data you guard your hypotheses you work in closed silos because you want to make sure that you're the one who ends up getting the publication in order to boost your own resume in order to get a grant in the future when you're trying to maximize impact um you also kind of change your behavior away from just producing high quality research for instance if you're studying something you have a hypothesis you doing experiment and you get a negative result most people don't share those because if you get a negative result it's not that exciting you're probably not going to get a lot of citations it reduces your average citations per paper and it actually makes you less likely to get funding in the future so negative results just aren't shared kind of like at the beginning I mentioned 50% of clinical or clinical trials aren't even shared harking is hypothesizing after results are known so halfway through a project you might realize that your results aren't that interesting and you'll kind of pivot the project in order to take advantage of the data that you do have and have a better result the problem is that this uh creates lowquality science that is hard to replicate and contributes to the replication crisis and then kind of at the extreme with the amid beta hypothesis some scientists and actually a frighteningly large amount of scientists will partake in fraud like they'll simply just fake results in order to get more citations in order to move along in their careers and like you can't even really blame people when you look at the United States less than 2% of first year phds end up becoming research professors so there's this hyper competitive job market where to even have a chance at having success within Academia you need to optimize your results in order to try and get the best uh citation based metrics you can because the person next to you probably is and they're going to get the next Grant because they are so I just want to take a second and look at this quote from Peter higs he's a theoretical physicist who uh was kind of the name snc of the higs bokon um so Peter higs he would publish like once every 3 to 5 years and they would be very like impactful Publications that would really like change his field the problem is that when you publish once every 3 to five years it's not very productive and so in today's world he actually doesn't think that he would have been regarded as productive enough in order to receive grants and this is a problem so how do we fix it I think the key and really at research Hub what we think is the key is that we need to acknowledge that good heart's law is sort of like a law of physics when it comes to academic publishing just the the activity of trying to measure quality is inherently going to fail no matter what measure you use it's eventually going to become a target for behavior and it's going to be gained by scientists and lose its efficacy as a measure so instead we need to accept that good Hearts law exists and use it to our advantage and use these measures as targets for Behavior so this is kind of like a cliche that was in the blockchain industry in 2017 is that blockchain is a hammer looking for a nail and when you look at the problem of wasted research funding It's actually an incentive problem and blockchain and tokenization token reward structures are probably the best tools that exist when it comes to actually like creating novel targets for behavior um by rewarding people with financial incentives so you can look at a variety of different blockchain projects obviously Bitcoin being the most famous where you can like elicit some pretty weird behaviors out of people I just uh checked it earlier today and right now the entire Bitcoin blockchain among all the miners they're collectively using about the electricity of Poland so block rewards are able to like incentivize people to spend a tremendous amount of money and energy just hosting instances of this database in order to receive Bitcoin as a reward so it's pretty effective as a tool in order to create targets for Behavior so at research Hub we're interested can we take this sort of idea of using token rewards to incentify specific behaviors and apply it to scientific research this is just kind of like an overview of our product today we have like a Reddit for science style Forum where anyone can uh share existing Publications they can publish natively to our Forum when you share Publications on Research Hub they can be uploaded and receive tokens based on what our community perceives as valuable so we have this ability to sort sort of create token incentives for specific behaviors on Research Hub uh a couple weeks ago our community got together and we were like hey let's let's play around with this maybe we can uh use these tokens in order to incentivize something interesting within the scientific community so we decided to create token rewards for people sharing open peer review if you look at this curve here we've had the ability to peer review on Research hub for about two years just as a feature um at the beginning of may we started to place $150 bounties four peer reviews so after the first Spike um maybe in the last like six or seven data points is when we uh started to like specifically reward peer reviews with $150 worth of research coin and as you can see it's kind of working we're right now averaging about 140 peer reviews per week uh using this reward structure so here's just a little bit more insight on these results right now we're uh if you extrapolate it across 12 months going to have 6,000 p reviews shared on Research Hub normally a peer review in a traditional scientific journal takes between like 3 to 9 months in order to get like a a peer review and get it published on Research Hub it takes about 16 days so using these token rewards in order to incentivize quick and open peer review it is working and it's actually an improvement on the current infrastructure for peer review just like to show this uh in more detail if you look at science which is probably like the most prestigious Journal that exists they average around 5,000 peer reviews per year so if we just maintain our current level as of last week we're kind of at a similar level to science and some of the biggest journals that exist which is obviously like pretty cool considering we only started this maybe like uh six weeks to eight weeks ago so at research Hub we have this ability to set targets for human behavior using token rewards which is super exciting and so what should we try and incentivize one thing that we're worried about is like as a company as a group of people we have about five people on our team is it possible for just our team to set incentives that will end up like increasing like the quality of research within the world we don't think so so we have a dow the research Hub Foundation all of our users people have been earning tokens they get together and they actually discuss like hey what do we perceive is valuable research and what do we want to incentivize what do we want to create token rewards in order to create targets for Behavior to try and change how scientists operate so this is what we've aligned on and we've voted on as a community we want to incentivize folks to create Open Access research so like any publication rather than being shared in nature where only people with subscriptions are able to read it we want to reward scientists for sharing Open Access Publications that anyone in the world can read regardless of where they go to school or what type of subscriptions they have we want to incentivize pre-registered research so pre-reg registrations are kind of like a blueprint for your research scientists will share their plan for a research project before they actually do it as kind of like a mechanism to hold them accountable rather than maybe pivoting if you're not that excited about your results Midway through a project you have to stick to what you described in a pre-registration and that has been shown in an evidence-based fashion to increase the replicability of research projects and then of course we think it's important to have open data if you share science and you have open data alongside it it allows anyone in the world to audit your project and then like see see if it's actually true what you're describing as a result of your your science so this was voted on uh by the research Hub Foundation which is our Dow and we're creating a reward structure in order to implement this so Laing exactly seven days from now we have a V2 of research hubs reward structure where the goal is to reward scientists for sharing Open Access pre-registered open data science so basically our plan is to iterate on the current incentive structure in Academia we look at citations and then kind of augment them through the lens of pre-registered Open Access and open data research so if you published in nature and you have a million citations on one paper you're going to have a lot of research coin waiting for you but you're not going to be able to claim it unless you share an Open Access version of that paper so uh it's actually legal for you to share a pre peerreview version of your closed access manuscripts so if you have say like $10,000 worth of research coin waiting for you but it's in a closed access paper you have to share a pre peer review version in order to claim that $10,000 um these rewards are three or 3x if it has open data accompanying it and then 2x again if the project was pre-registered so our goal here is similar to how we were able to actually increase the number of open peer reviews on Research Hub to try and incentivize scientists to share Open Access open data and pre-registered Science by using token rewards so we think a lot of the problems that exist in science are due to this like poor incentive structure that's a result of good heart's law there are funders who are trying to measure quality rather than trying to set targets for Behavior so at research Hub we're trying to set targets for behavior and we believe that this is the best way forward in order to incentivize high quality research so like a little bit hyperbole here but we think like if the world decides to accept good Hearts law and create targets for uh Behavior rather than trying to measure quality we're going to have a lot more efficient science so here's some QR codes if you'd like to get involved and I think we have maybe like two or three minutes so if anybody has any questions like happy to answer them yeah thank you Patrick uh so if anybody has any questions feel free to raise your hand now I'll come bring you the mic hey Patrick thanks for that um what is the the biggest challenge that you're facing and and getting those papers on boarded from traditional Academia right now so it's kind of like a sales issue um which we've started to sort of figure out so like not a lot of people know about research hub so we have to kind of have our community reach out and tell people we exist uh just in the last maybe month or two we've started to kind of create an internal like uh inside sales function within our community where uh we're basically trying to recruit folks to do peer reviews so we'll have a preprint that's shared a peer review Bounty will be placed on it and then we'll shoot like LinkedIn DMS and cold emails and we actually get like a pretty high kind of uh conversion rate it's about like 30% of folks that we reached out to will actually come back into research Hub and complete a peer review which is shockingly high for cold Outreach if anybody's tried before um what is the the sort of General perception of token rewards obviously the turn the rate is quite High 30% but tokens in particular do people turn their nose up at this it depends so kind of like older like successful folks they oftentimes don't really need the money and so they're a little bit more hesitant but we still had some professors in Department who are willing to do peer reviews um younger folks love it so the research of foundation was largely comprised of like postto and uh early stage grad students um but yeah paying people for something they're not currently being paid for is actually pretty successful so it's again like kind of just a function of how much Outreach we do awesome okay well that's all the time we have for questions can we give one more round of applause for Patrick please thanks everybody thank you Patrick [Applause]