Jotting down a few notes from day 1 at the #LAK18 conference in Sydney before I forget them. I’ve got the usual grand plans for writing blog posts but as my WordPress drafts indicate, best laid plans etc…*
I did my pitch for the 2 day hackathon – slides below. There were a couple of other pitches focused around technical architectures for privacy, and algorithmic transparency and so we clustered together around a table to talk about privacy and data literacy and stuff. Tore Hoel explained some of the work he is involved with in thinking about an ISO standard for privacy. Our group was pretty big so after lunch we split into 2 – privacy architectures and algorithmic transparency. I went with the algorithmic transparency crowd.
A better write up will follow, but a few immediate things we produced:
(Edit: Adam Cooper produced a better write up)
Algorithmic Transparency – Definition
All steps from source data to final prediction or descriptive. Includes transformation steps and the definition of derived quantities.
Algorithm may be hierarchical, embedding, contain inner units which are considered ‘black boxes’ i.e. there are different levels of description – which level depends on the audience.
Dimensions of Algorithmic Transparency
- As the legal basis for consent e.g. GDPR
- Institutional QA acceptance etc.
- Data Literacy – broader educational aims – helping students better understand what the magic is
- Trust building (between institution, students, suppliers etc.)
- To act ethically; transparency enables ethical action (also research ethics)
- To ensure informed, rational and effective decision making and interventions
- Procurement Teams
- Pastoral Care / Tutors
- IT people / learning technologists
- Ethics Committees
This led us on to a bit of matrix mapping between dimensions and stakeholders – who needs transparency and for what purpose, which in turn led us on to a discussion about the kinds of information that need to be available, and how there will be several descriptions of what any algorithm does e.g. a plain language one for “end users”, a legal description, a technical description. These should be layered so one can dig deeper (read more) and could even be contained in a metadata schema associated with any algorithm (this relates to the discussions on privacy architectures – in particular to recording and tracking consent).
For tomorrow we talked about how we could push this on into some examples. We might even be able to use it as a concrete example of an activity within the data literacy playground idea – showing different explanations of what an algorithm does.
A few good links that I was pointed to:
Break time I had a chance to catch up with 2 colleagues from the University of Strathclyde (oh the irony) and had a quick chat about the Enhancement Themes work in the Scottish sector and their plans for piloting OnTask. We need to chat more…
In the evening I had a great conversation with Kirsty Kitto about how LAK brings together researchers and practitioners. We talked about professionalising the role of the learning technologist and how the distinction between IT and learning technology was increasingly a false divide. We share a concern that the conversations and inputs at other conferences about learning analytics are dominated entirely by the predictive / retention focused stuff that can be bought today – probably because that’s what’s possible to operationalise.
Also really lovely to see several colleagues who made the move to Australia after finishing up their PhDs at Edinburgh.
*Also I reckon I have a better excuse than most this time. After 2 days of travel delay, cancelling and rebooking of my entire outbound journey and hotel arrangements, I just did 6.5 hours flight to Doha, 7 hours layover, 14 hours flight to Sydney, arriving 06:00 in the morning. Hit the conference venue at 08:00, did a full days work and then went out for beers and food.