Disparate mistreatment: different error rates for two groups (eg. men and women) if you minimise error rates overall data.
The semantics of the errors also matter, even if the error rates are the same for both groups.
They found an algorithm for predicting likelihood of criminal re-offending disproportionately favoured white people in terms of false negatives, and biased against black people for false positives.
Algorithms learn correlations between features, but what actually are the features and what does this mean in the real world? This is a really nice look beyond the numbers to who is affected by algorithmic decisions... and how to correct for it!
And the speaker was excellent, explained things really well.
(Muhammad Bilal Zafar, Isabel Valera, Manuel Gomez Rodriguez and Krishna P. Gummadi)
It's about keeping our jobs and keeping publishers in business. It's not about advancing the state of the Web. Boring.
Keeping your job is more important than your research having an impact. Okay fine, you can have that opinion. But get out of my way.
I struggle to relate to pushback against Linked Research ideas because of money/job/norms problems. I get that these are important issues for some people, but I'm so TIRED of people using this to blockade even discussing it. It's like they don't want the status quo to change; it's comfortable. They know one way to keep their job or get funding, and won't contemplate developing others. Use the goverment funding assignment as an excuse. Easy out, no point in bothering, the government will never change. Reinforce this for the next cohort of student. Keep perpetuating the problems for the next generation. This isn't what I want from academia, and I won't play.
Prof Denemark crystallises some of the complex political contexts in which people use the Web and social media globally. When we study social media it's easy to forget the real people in real places with real events occurring around them.
In reply to: https://rhiaro.co.uk/2017/04/at-wow2017
We had ten years of the Web before we had Google.. We take for granted the future of the Web, but it's really only been around for a tiny fraction of time... Because we put content on the Web is why it grows. If we stop, it dies.
Technologists can't study the web alone. We need a whole interdisciplinary relay to try to understand the Web and make sure it stays good for humanity. Right now I'm worried it's on the cusp of *not* being good for humanity.
Today we're talking about the Web Observatory, and then about decentralisation of the world. They're connected by data. We want Web Observatories around the world... lots of telescopes, links to datasets eg. webobservatory.soton.ac.uk.
Twenty years ago, we started Southampton eprints. People thought we were mad. The rest of the world caught up... but I did not forsee a world where I'd have to pay thousands of dollars to publish an open access publication. That's not we envisioned. To me, a Web Science approach to studying this issue is absolutely overdue.
This year, the Linked Data on the Web workshop at WWW2017 held an open discussion session about academic publishing. In particular, about what is getting in the way of us getting more research outputs into the LOD Cloud. The discussion was led by Sarven Capadisli, the loudest voice behind the Linked Research initiative, who did his best to remain a neutral facilitator, and let others in the room argue about the things he is normally arguing about.
There was a general consensus that we should have research from the Linked Data on the Web community in the form of Linked Data, and on the Web. Sounds like kind of a no-brainer, but you'd be surprised. Or maybe you wouldn't. The group, around 30 people, discussed the reasons why we, as a community, are lagging.
One of the reasons is tooling. Nobody wants to hand-author HTML, especially not with RDFa in. Nobody really wants to turn their research articles into datasets in any other syntax either. There are (apparently) still a large portion of researchers even in CS who use MSWord to generate their final PDFs so they need something that works at least as well as that to generate Web-friendly submissions.
There is also a matter of incentives, on several different fronts. Most researchers are a-researchin' and publishing their work to advance their career, and to get funding so they can do new cool stuff. So how are individual researchers incentivised to put in the extra effort it takes to generate HTML? On another level, how do we incentivise researchers to improve the state of tooling and resources? The first can be tackled along multiple fronts, one of which is petitioning publishers to demand HTML (just like today they demand LaTeX). What are the incentives for publishers to do this? There were a few ideas thrown around, including how they can improve their SEO, access, discoverability, and creating more pleasurable reading experiences than PDFs can deliver.
But there's a large faction ill-at-ease with depending so much on publishers to drive this change, even if it didn't look like it would be an agonisingly slow process. Some of us, myself included, would like to shift the whole scholarly communications process to be more self-sufficient and less dependant on centralised third parties. We do not want to do this at the expense of quality of work of course! (Which some people immediately assume is the case). Beyond publishing, we want to open up the review process, so it's both more transparent, and so researchers get the credit they deserve for this work. Conversations can continue well beyond the submission process if the reviews are open and public. But again, we need to work on the tooling and incentives to enable this.
Whilst I agree that we are woefully underdeveloped on the tooling front, I object to the weight this was given in the context of getting just Linked Data researchers to adapt to the Web. Personally I think if Linked Data researchers are going to cry about being required to submit their contributions as Linked Data, I will have trouble taking them seriously. Writing HTML is not a high bar. If you're comfortable with LaTeX, you can switch off a few neurons and write HTML instead. Or you can use Pandoc. As Jens Lehmann succinctly put it, the LDOW workshop is about advancing the state of Linked Data on the Web. If this community is not prepared to drive the state of this forward, even (especially?) if it includes working outside of the current system and taking some risks, who is going to?
So overall there was a vague feeling of consensus that we need to do something to take better advantage of Web technologies, and that LDOW as an established and respected venue is a good grounds for an experiment. I don't know if LDOW will manage to require HTML submissions next year, but the organising committee seem like they'll be inclined to strongly incentivise it. Stay tuned (watch the public-lod@w3.org mailing list. And speak up, if you care. Notes from the session are here.
We'll be continuing this discussion in the Web Observatories workshop this afternoon, around 4pm.
PS. the Enabling Decentralised Scholarly Communication workshop at ESWC this year requires Web-based submissions and is all about building and connecting the tooling to advance the state of academic publishing on the Web. Deadline is 17th of April. See you there.
rhiaro arrived at an event for 4 days, 7 hours, 30 minutes,