SAIL: Open Analytics
Hi all,
A number of years ago, a group of us published a concept paper on Open Learning Analytics. The intent was to create a platform for analytics work where various tool sets and resources could be shared and the joyful network effects could be realized as we engaged in analytics research collaboratively. A modularized and integrated infrastructure, comparable python or R libraries, would enable researchers and practitioners to plan, deploy, and share analytics tools, scripts, and data. The concept was not fully operationalized. Recently, however, Microsoft has launched Open Education Analytics which is the closest instantiation of the idea of OLA that I've come across. (disclaimer, I serve on the advisory board with this MS project).
If AI continues its rapid encroachment on "all things related to human cognition", educators are faced with important questions: what do we do better than AI? What remains unique to human cognition as AI advances? One area of interest is in the domain of "who we are" not "what we know". Who we are focuses on the core of our being rather than emphasizing knowledge that we've acquired. I came across this article recently about a study that "used AI to determine the shortest path to happiness". An explainer video, short on technical detail, provides a broad overview of Self Organizing Maps - an unsupervised machine learning technique - that is central to their model. I then completed their self-assessment tool that, after spending about 10 minutes answering various questions, returned some key recommendations about having a happier life. I'll keep you updated. The use of AI in mental health and self-help is no doubt going to be a growing trend, but one that is fraught with enormous concerns. Last week I noted that Microsoft stopped supporting emotion detection software. Psychological assessment instruments present an even more complex challenge. And raises questions if AI could replace therapists. With growing concerns about mental health in higher education, it's reasonable to expect AI to play a greater role.
A few topics of interest this week:
The Gov't of Canada has developed a data literacy resource which includes a range of webinars and self-directed tutorials.
I mentioned synthetic data last week and wondered where the educational data sets are in this area. Virtual worlds (simulations) may be one area to generate data to train education/learning models
Access to data is the big challenge for AI advancement in education. Authentic data sets (whether real, hybrid, or synthetic) align with what individual learners actually do in specific settings. OpenAI used 70k hours of Youtube videos to teach a model Minecraft. As more of the learning process digitizes, more data will be available for developing complex models. The key challenge then arises as to which companies will have a large enough scale of data to build complex models of human cognition. Right now, I'm seeing: Instructure, Coursera, Google, and Microsoft (the K-12 space would have Byju) as a small number of companies that have large enough learner datasets to create truly innovative models that would dramatically impact teaching and learning. This raises important questions about data access and researcher involvement.
And just because climate change impacts us all: AI is Essential for Solving the Climate Crisis
Democratic AI - Given human bias, suspect motivations, should AI govern us? This paper doesn't go that far, but instead suggests that "human-in-the-loop" models can provide democratic models preferred by humans over ones designed by humans.
Coursera has launched an intro specialization for Machine Learning.
GRAILE has been tasked with developing AI literacy in knowledge processes (learning, sensemaking, decision making) for the Australian Defense Sector. We'll share more soon, but if your organization is interested in AI literacy training, let us know.
I hope y'all have an amazing weekend!