Participants in Project Aiur will collaborate to build a blockchain “Knowledge Validation Engine” (KVE) which will allow users to submit research and have it verified against all other research in the world. Aiur aims to be part of bringing radical change to a highly lucrative, oligopolistic scientific publishing industry that has led to a dysfunctional incentive model.
There are a number of problems affecting the scientific community today which are hampering global progress. While there is an abundance of research being conducted and published, it is costly to access and difficult to manage. Research professionals are pressured to deliver, publish, and review on tight deadlines, creating perverse incentives to exaggerate facts and omit assumptions and constraints. With little to no accountability and reward for authors and reviewers, reproducibility suffers.
Project Aiur will join researchers, coders, and anyone interested in science into a community-governed ecosystem that works to address the numerous problems damaging the scientific research community. Participants will collaborate to build a “Knowledge Validation Engine” (KVE) based on artificial intelligence and machine learning technology. The engine, when ready, will allow users to submit a piece of research and have it verified against all other research in the world. Over time, the engine will host a vast collection of research that has been validated.
Anita Schjøll Brede, CEO and co-founder of Iris.ai said: “We envision a world where the right scientific knowledge is available at our fingertips, where all research is validated and reproducible, where unbiased scientific information flows freely, and where research already paid for with our tax money is free to us. With Project Aiur, we aim to give ownership of science back to scientists, universities, and the public. Validation of science is something that has to be an open process – no one company can be an arbiter. With Aiur, we’re setting up a self-sufficient, decentralised ecosystem that will disintermediate the process of validating scientific papers. Everyone who participates in Project Aiur will own the project, can use the tools we build together, and can vote in important decisions.”
Project Aiur is enabled by blockchain and utilizes the AIUR token. This functional token grants membership, voting rights, and access to the KVE and other tools developed on top of the community’s software. Anyone who contributes to the community through code, training data, and research earns AIUR tokens. The tokens can from the beginning be utilized for Iris.ai’s premium services.
Co-founders Anita Schjøll Brede and Victor Botev talked with Block Tribune about the project.
BLOCK TRIBUNE: Walk me through how this works for a researcher and a contributor.
ANITA SCHJØLL BREDE: So there’s a number of different ways you can be a contributor. There is from the start of the project as soon as the code and everything is launched; you can contribute by being AI trainer or by part of the team actually coding the platform. And of course that is an AI trainer takes a piece of research literature so a research paper and trains that by annotating the data. Selecting key words from the text. Selecting other contextual synonyms et cetera. And building up this data set. And that is how you can earn tokens in the beginning. The same goes for helping to code the knowledge, validation engine as well. Longer term as that part and as the knowledge validation engine is build out; then a researcher can choose to take their research paper that either is ready to be published or that they want to just validate as part of their writing process and give that to the knowledge validation engine to read and have that validated up against all the other research.
Currently we’re connected with the Iris.ai services to a database of 120 million open net access articles. That is where we’re starting.
BLOCK TRIBUNE: Okay. What is your validation criteria for what articles you will take and what you won’t?
ANITA SCHJØLL BREDE: The core of that technology is of course and AI technology. I’ll let Victor talk a little bit about the knowledge validation engine.
VICTOR BOTEV: When we start developing the knowledge validation engine, one of the things that we have planned ahead is that we will basically release a demands request for AI training to train a particular data set. Then everyone that participates during that campaign of training those requests will earn tokens. The articles will be pre-selected more or less automatically but data set has to be balanced. There’re certain rules so it’s not going to be completely random. It will have to cover all the different areas of science. That’s why we prefer to submit demand requests for the purposes of the knowledge validation engine. And not to have a completely randomly ongoing AI training in the project basically.
BLOCK TRIBUNE: So you’re drawing from an existing pool of scientific research?
VICTOR BOTEV: At least at the beginning, yes. In the future, the idea is that for the knowledge validation engine, we would like to also build a validated repository. Meaning that each of the articles that is actually used in the final report, and maybe I should say that the knowledge validation engine will take a piece of text; it could be a scientific article or a human build hypothesis and build a report on what are the main problems, what are the main solutions. What are the solutions that are needed for those problems and draw the papers that back up or that contradict that hypothesis. Basically on all of the branches of those papers annotate what is the reproducibility level of each and every article? Are there any facts that are missing? Are there facts that are not backed by other hypothesis in the literature? And produce this long report that people can navigate and see and basically identify issues or problems in the short text.
So, what we hope is that people will be also hosting articles in the knowledge validation engine and help us build the validated repository where all of the articles are with a score of [inaudible 00:04:40] threshold meaning following certain standards for reproducibility of research and for backing up by other validated articles. But we will retroactively execute that on the open access articles that we have as well when the knowledge validation engine is up and running.
BLOCK TRIBUNE: Who do you anticipate is going to be using this service?
ANITA SCHJØLL BREDE: It’s a mix. The majority of it is going to be people affiliated with either an academic institution so people working at a university or people working in industry in R&D departments. That can take a variety of forms. It can be someone working at a consortium working on a specific research project. It can be a department at a specific university or of course an R&D department at a specific company. They will have slightly different ways of accessing the tools from the services that is built by the community. It could be either querying the [inaudible 00:05:53] knowledge validation engine as one component into a tool that they’ve built themselves internally. For example say a large chemical manufacturer for example could do that or a university doing a more thorough systematic research landscape mapping would query the tool and just get the raw results. And of course there are also third party providers that want to build tools on top of what the knowledge validation engine that is built by the community can provide. So, they will be able to UX and the interface is on top of it.
BLOCK TRIBUNE: What are the tools they’re using today and how will you differentiate from those?
ANITA SCHJØLL BREDE: This kind of comes back to the company that is the initiator of Project AIUR, Iris.ai; we’ve been running the company that company for three years. We’ve built these tools that semi-automate the early stage of a research process. Those tools and the future tools then now build in where the knowledge validation engine and project AIUR; we’re really replacing really, really old school manual processes. Google scholar, PubMed’s own search query engine is what is being used today. It is incredibly inefficient and requires a lot of time. If we talk industry, we talk two to four weeks of manual labor that can be cut down to just a day or two of work. If we talk academia, we’re talking six months worth of human labor that can be cut down to a month even a couple of weeks or shorter.
BLOCK TRIBUNE: Okay. What do you anticipate will be the effect of this project?
ANITA SCHJØLL BREDE: Long term, the ability to have a knowledge validation engine where you can test any hypothesis up against all of the validated knowledge in the world. It’s going to be vital for, and this is going to sound fluffy, but to the success of the human 0. Not just with us, I mean obviously we see ourselves just as a initiator of this. But having the scientific world coming together like this is vital. The research world has struggled immensely for the past 20 years being more and more siloed with the big publishing houses sitting on top and holding on by [inaudible 00:08:10] content making sure that universities in the world that cannot afford to access research. There’s literally researchers who cannot read research because they don’t have access to it.
By opening up that and having the entire research community coming together to build these tools and to share their knowledge; that is vital. If we want to solve problems that we face like climate change like the survival of our species. I’m not saying that we are the one and only we’re going to be able to manage that but we do need to come together as an academic community to make sure that science and research survive as a discipline.
BLOCK TRIBUNE: Are there any other applications for this tool you’re building?
ANITA SCHJØLL BREDE: Absolutely. The ability to take knowledge and validate it is valid for a number of different things which is why as Iris.ai, the commercial entity that is starting this project; we focus specifically, as I said, on the research process in the early stage. But the ability to take a hypothesis and validate that up against academic research is invaluable in a number of cases. In the patent landscape and innovation in that sense is of course one of the areas that most easily come to mind. There is also the legal space of course as well. Like the ability to take legal claims and validate that up against research on the topic is vital. There’s a number of different areas that the core technology can be applicable to. As I said this ecosystem will be entirely open so other parties can come in and build tools on top of that.
BLOCK TRIBUNE: Do you have any revenue streams beyond the ICO?
ANITA SCHJØLL BREDE: Yes we do. Iris.ai, the company, we are generating revenue. We are selling to large R&D departments in industry. Project AIUR is a not for profit side project that we are doing. Seventy-five percent of the money raised in the ICO will go to the community itself. Part of that will be released through escrow mechanism as certain milestones are reached that could be released to us as service provider to the community or it could be released to others who are helping to build the tools. Iris.ai, the commercial entity we will keep … We are [inaudible 00:10:32] funded and we will keep selling to R&D departments and universities as we have been doing already. Project AIUR is born out of a need that the industry has. We’re not going to change the industry by selling tools to big R&D departments.
The way to change the industry is to build this movement from the grass root up. That part of the project is a non profit part.
BLOCK TRIBUNE: Have there been any breakthroughs or astonishing discoveries in the early stages of this tool?
ANITA SCHJØLL BREDE: The Iris.ai tool? Yes. We presented a couple of research papers all ready. Actually, Victor, maybe you want to talk briefly about the Wisdom paper and the Scithon paper.
VICTOR BOTEV: We have developed a couple of very efficient algorithms let’s say for in calculating documents similarities which allows us to basically use the AI mechanisms to quickly get you the documents that are most relevant to another document just based on the content. We also have developed an algorithm to be able to evaluate human annotations which allows us to actually award people for annotating text and kind of making sure that they’re not just putting random words when they annotate key words or stuff like that. So, we think that gives ability to make those rewards plausible. Also we have developed a framer to evaluate the tools for scientific discovery. More like a framework through evaluating different teams. Operating with different tools and using indicators to basically assess their [inaudible 00:12:36].
These are three areas in which we have put a lot of effort and we have published already papers in two of the areas and the third one is coming this year.
BLOCK TRIBUNE: That was my question. What are the breakthroughs that you’ve had? The papers that you’ve published?
ANITA SCHJØLL BREDE: So, two of the three have been published. The third one on the Scithon evaluation framework is from a commercial perspective, the most exciting one. We did a demo paper of it that has been published and released already. Now we’re presenting a more in depth full text paper. The break throughs are there. It has been accepted into the conference. It has just not been gone through the publishing process yet. It has been peer reviewed as well already. [crosstalk 00:13:24] the break through there [crosstalk 00:13:26]. We are able to prove that the tools that we have already built consistently out perform the existing tools of the day. So teams using our tools consistently out perform teams using Google scholar or PubMed and the likes. [crosstalk 00:13:42]
BLOCK TRIBUNE: What was the topic of the paper, though?
ANITA SCHJØLL BREDE: It is an evaluation framework.
VICTOR BOTEV: Maybe just one last thing about the breakthroughs, if it’s possible. I just want to say that one of the things with the Wisdom paper which is a document similarity mechanism which is very efficient and we show that it performance wise it can compute much more like more than 30 percent more than the state of the art metrics existing already. Comparisons between documents within one second. The second one is the Scithon framework and the third one is the algorithm for evaluating human annotations which we think is a very interesting algorithm because it does not … The machine doesn’t really guess the key words. It use other indicators to make sure that the key words cover [inaudible 00:14:58] text.
Source: Read Full Article