Thread by @emvdn

Here at the Monash's Clayton campus today for the Data Science and Artificial Intelligence in the Humanities, Arts & Social Sciences symposium - going to hear about all the research going on in this area

In the mid-1980s at Monash University, there weren't any women in the politics department. @bestqualitycrab was told, "We don't encourage women to study Honours when all you'll do is leave to get your wombs working". She enrolled immediately.

Thomas Osborne says there are four types of academics: experts, legislators, interpreters, and mediators. @bestqualitycrab notes these and adds 'agitator', someone who wants to make a difference in the world. She sees herself as an academic agitator.

Humanities focuses on tactile, viscous, dynamic connections and insights, the way @bestqualitycrab thinks of it. She updated @gapingvoid's information-knowledge visualizations thusly.

.@bestqualitycrab: What does it mean to be lost at sea? How do we move from incoherence to perception? At one point, the weather was considered to be unfathomable and unknowable, the cause of deaths at sea. Now we see weather forecasts daily. Big data can be understood.

.@bestqualitycrab asked in her research: Can we use forecasting to discuss patriarchy? Does patriarchy have a Plimsoll line (a marking on the side of a ship to indicate its load limit)?

When @bestqualitycrab found women's participation rates in the film industry were declining, she realised this problem needs more than "add women and stir". She decided to study men in the film industry instead.

Statistical analysis of networks of creative teams in the film industry showed that 41% of male producers over 10 years didn't EVER work with a woman. "I look at that and I think, how do I change that?": @bestqualitycrab. Why don't we just... take these men out of the network?

Network analysis also helped @bestqualitycrab determine what men who don't work with women had in common. In the Australian film industry, it wasn't age or race, it was class. These are men who went to private schools.

What does it mean to give weight and shape to relationships? If we visualise networks, we might be able to see the contours of justice. Patriarchy looks like a network of red lines connecting red nodes where women are coded as blue. @bestqualitycrab.

Counting numbers doesn't give us the full picture, concludes @bestqualitycrab. We need to consider what counts.

We're now hearing from people who are traditionally outside the humanities/social sciences disciplines to hear about their research, and I'm interested even when I can't keep up! Klaus Ackerman is telling us about SoDa labs at the Monash Business School sodalabs.io/about.html

Religious customs like prayer times influence internet use, and tracking patterns like this by pinging IP addresses is one use of big data, says Ackerman

When a hurricane hits a certain region, pinging IP addresses can reveal which areas are back online and when, indicating patterns of recovery after a natural disaster

Simon Musgrave introduces Vector Space Models, or word embeddings, to us. This involves taking large samples of language and looking at which words co-occur in a certain window, like 12 words of text. Doing this produces a huge amount of linguistic data.

Using artificial intelligence, algorithms like word2vec or Glove can "reduce dimensionality", (which I don't quite get but apparently it makes the data more manageable)

We get a demonstration: Musgrave types in words with past tense into this window and gets some numbers that he tell us can shed light on irregular past tense

But this technology can also be used to map semantic change over time. If you look at what words related to "gay" - or even "broadcast" - these changed a lot from the 1970s to now

There are really interesting possibilities for these tools. A questioner asks if you could generate content with programs like this, and if it would be identifiably "synthetic". Musgrave says making generated text "convincing for the consumer" isn't there yet.

From a paper intriguingly titled 'The Geometry of Culture' using word embedding models: "“Boy” only has meaning in that it is positioned near “man” but closer to “child,” near “girl” but closer to “male.""

arxiv.org/ftp/arxiv/pape…

Word embedding models can also identify bias. For example, “doctor” is consistently found to be more “white” than “black,” and “scientist” more
“masculine” than “feminine.”

arxiv.org/ftp/arxiv/pape…

Asher Flynn is part of a team using artificial intelligence to research technology-facilitated sexual violence: pro-rape groups, harassment of rape victims, simulated sexual violence, online sexual harassment, image-based abuse

Flynn focuses on the complex phenomenon of image-based abuse: the non-consensual taking and/or sharing of sexual images. Her research group includes work with the Office of the eSafety Commissioner, and has a report mapping the scale of the issue research.monash.edu/en/publication…

Artificial intelligence is being used to *create* image-based abuse, like deepfakes: swapping faces in videos to make it look like someone is nude or starring in porn. AI generated porn looks a lot more convincing than Photoshopping someone's face into a nude pic.

But artificial intelligence can also *detect* manipulated images, and that's where future research is heading. A questioner adds that blockchain technology is another potential response, as it can flag image manipulations. But Flynn says blockchain can be prohibitively expensive

There's a lot of onus on the person featured in image-based abuse. It's usually up to them to detect and report how their image has been misused, which is yet another issue.

"Movies are better with a soundtrack, right?" In his talk 'Sonification of Tropical Storms', Mark Ballora has interpreted data about storms and maps to so we can *hear it*. Tremors, waah-waahs, and gong sounds indicate intensity.

Sound is processed more quickly and more viscerally than visuals. Higher pitches and rates of pattering might signify higher rainfalls more naturally. You can hear some storms here:
theconversation.com/turning-hurric…

"I'm hoping that sound will make this more emotionally easy to understand". Ballora wants it to be part of science pedagogy - when kids learn how to read graphs, they could learn how to listen to them too (and I love that double meaning of 'listen' as 'hear' and 'pay attention')

A big question for social media research for a long time has been how to access and collect data. Seth Erickson notes that Social Feed Manager gwu-libraries.github.io/sfm-ui/ is used to investigate things like how people on Twitter talked about a hurricane before/during/after it hit

You "dehydrate" the data by replacing usernames with numbers so you can release it publicly, then "rehydrate" it with another program later. This is done to comply with legal restrictions on data use, and with Twitter's terms of service

I asked Erickson what's to stop people from copy-pasting any given tweet into Google and finding who originally posted it, and he said "nothing". So there are some ethical discussions to be had here.

Social media data can reveal phenomena like linguistic alignment: people tending to use similar language when replying to each other. The perception of power is important - someone who perceives someone else as higher status is more likely to draw on their language.

Maria O'Sullivan: If you introduce artificial intelligence and algorithmic decision-making into processing refugee status decisions, you'll make things more complicated. How will a refugee know what exactly stopped their claim?

Maria O'Sullivan: AI can make things more objective and efficient. But are there some decisions that AI should never trespass into?

Anne McKenna: Private sector actions can have national security impacts. Fitness tracking app Strava gave away the location of secret US army bases when it released a global map of people going for runs
theguardian.com/world/2018/jan…

Anne McKenna: Just look at how many different laws in the US might regulate data. It's a complex landscape.