AI and education – notes from reading group

Will Higher Ed Keep AI in Check? (notes from Chrysanthi Tseloudi)

In this article, Frederick Singer argues that whether the future of AI in education is a positive or a dystopian one depends not on a decision to use or not use AI, but on retaining control over how it is used.

The author starts by mentioning a few examples of how AI is/ may be used outside of education – both risky and useful. They then move on to AI’s use in educational contexts, with examples including using an AI chat bot for students’ queries regarding enrolment and financial issues, as well as AI powered video transcription that can help accessibility. The area they identify has the most potential for both risk and benefit is AI helping educators address individual students’ needs and indicating when intervention is needed; there are concerns about data privacy and achieving the opposite results, if educators lose control.

The final example they mention is using AI in the admissions process, to sidestep human biases and help identify promising applicants, but without automatically rejecting students that are not identified as promising by the AI tool.

I think this is something to be cautious about. Using AI for assessment – whether for admission, to mark activities, progress, etc – certainly has potential, but AI is not free of human biases. In fact, there have been several examples where they are full of them. The article Rise of the racist robots – how AI is learning all our worst impulses and Cathy O’Neil’s Ted talk The era of blind faith in big data must end report that AI algorithms can be racist and sexist, because they rely on datasets that contain biases already; e.g. a dataset of successful people is essentially a dataset of past human opinions of who can be successful, and human opinions are biased –  if e.g. only a specific group of people have been culturally allowed to be successful, a person that doesn’t belong to that category will not be seen by AI as equally (or more) promising as those who do belong to it. AI algorithms can be obscure, it is not necessarily obvious what they are picking up on to make their judgements and so it’s important to be vigilant and for the scientists who make them to implement ways to counteract potential discriminations arising from it.

It’s not hard to see how this could apply in educational contexts. For example, algorithms that use datasets from disciplines that are currently more male dominated might rank women as less likely to succeed, and algorithms that have been trained with data that consists overwhelmingly of students of a specific nationality and very few internationals might mark international students’ work lower. There are probably ways to prevent this, but awareness of potential bias is needed for this to be done. All this considered, educators seeing AI as a tool that is free of bias would be rather worrying. Understanding the potential issues there is key in retaining control.

How Artificial Intelligence Can Change Higher Education (notes from Michael Marcinkowski)

For this meeting, I read ‘How Artificial Intelligence Can Change Higher Education,’ a profile of Sebastian Thrun. The article detailed Thrun’s involvement with the popularization of massive open online courses and the founding of his company, Udacity. Developed out of Thrun’s background working at Google in the field of artificial intelligence, Udacity looks to approach the question of education as a matter of scale: how can digital systems be used to vast numbers of people all over the world. For Thrun, the challenge for education is how it can be possible to develop student mastery of a subject through online interactions, while at the same time widening the pathways for participation in higher education.

The article, unfortunately, focused most on the parallels between Thrun’s work in education and in his involvement with the development of autonomous vehicles, highlighting the potential that artificial intelligence technologies have for both, while avoiding any discussion of the particulars of how this transformational vision might be achieved.

Nevertheless, the article still opened up some interesting concerns around questions of scale and how best of approach the question of how education might function at a scale larger than as traditionally conceived. At the heart of this question is the role that autonomous systems might have in helping to manage this kind of large scale educational system. That is, at what point and for what tasks is it appropriate to take human educators out of the loop or to place them in further remove from the student. In particular, areas such as the monitoring of student well-being and one-on-one tutoring came out as areas ripe for both innovation and controversy.

While it was disappointing that the article largely avoided the actual issues of the uses of artificial intelligence in education, it did offer an unplanned for lesson about AI in education. Like in the hype surrounding self-driving cars, the promises for a new educational paradigm that were put forward in this 2012 article still seem far off. While the mythos of the Silicon Valley innovator might cast Thrun as a rebel who is singularly able to see the true path forward for education, most of his propositions for education, when they were not pie-in-sky fantasies, repeated well worn opinions present throughout the history of education.

Suggested reading

JISC Horizon Report on wellbeing and mental health – notes from reading group

Suzi read the first section of the JISC Horizon Report mental health and wellbeing section. This talked about the increasing demands on mental health services and discussed some possible causes including worries about money and future prospects, diet, use of social media, and reduced stigma around talking about mental health.

Many institutions are increasing their efforts around student wellbeing. The report mentioned a new task force looking at the transition to university and support in first year: Education Transitions Network.

Four technologies are mentioned as currently being used within HE:

  • Learning analytics to identify students in need of being checked in on
  • Apps and online mood diaries, online counselling
  • Peer support (overseen by counsellors) via Big White Wall
  • Chatbots

The report didn’t have a great amount of detail on how these are used. Using learning analytics to see who’s not engaging with online content seems like the simplest application and is doable in many systems but even this would require care. Revealing that you are keeping students under surveillance in a way they might not expect may cause them to lose trust in the institution and retreat further (or game the system to avoid interventions). Then again, maybe it’s just helping us know the sort of things a student might expect us to know. Universities can be quite disjointed – in a way that may not seem natural or helpful to students. Analytics could provide much needed synaptic connections.

It also struck me that using technology to support wellbeing (and even mental health) is in some ways similar to teaching: what you’re trying to achieve is not simple to define and open to debate.

Johannes read the blog post Learning Analytics as a tool for supporting student wellbeing and watched a presentation by Samantha Ahern. Samantha Ahern is a Data Scientist at the UCL and does research concerning the implications of Learning Analytics on student wellbeing.

In her presentation, she outlined the current problem of the HE Sector with student wellbeing and provided some alarming numbers about the increase of reported mental disorders of young adults (16 –24 years old). According to the NHS survey on mental health in the UK, around 8% of male and 9% of female participants had diagnosed mental health issues in the year 1992. This numbers increased to more than 19% of males and even 26% of females in 2014. Interestingly, females are much more likely to report mental health issues than males, who, unfortunately, are the ones doing most harm to themselves.

In her opinion, the HE institutions have a great responsibility to act when it comes to tackling mental health problems. However, not all activities actually support students. She argues, that too many university policies put the onus to act on the student. But the ones that would need help the most, often do not report their problems. Therefore, the universities should take a much more active role and some rethinking needs to take place.

Her main argument is, that although learning analytics is still in its beginnings and it might sound like a scary and complicated topic, it is worth doing research in this field, as it has the capabilities to really improve student wellbeing when it is done correctly.

It was very interesting to read and listen to her arguments, although it was meant to be as an introduction to learning analytics and did not provide any solutions to the issues.

Roger read “AI in Education – Automatic Essay Scoring”, referenced on page 27 of the JISC Horizons report. Is AI ready to give “professors a break” as suggested in a 2013 article from the New York Times referring to work by EdX on development of software which will automatically assign a grade (not feedback) to essays. If so then surely this would improve staff wellbeing?

Fortunately for the Mail Online, who responded to the same edX news in outraged fashion (“College students pulling all-nighters to carefully craft their essays may soon be denied the dignity of having a human being actually grade their work”) it doesn’t seem that this is likely any time soon.

Recent work from 2 Stanford researchers built on previous results from a competition to develop an automatic essay scoring tool, increasing the alignment of the software with human scorers from 81% in the previous competition to 94.5%.  This to me immediately begged the question – but how consistent are human scorers? The article did at least acknowledge this saying “assessment variation between human graders is not something that has been deeply scientifically explored and is more than likely to differ greatly between individuals.”

Apparently the edX system is improving as more schools and Universities get involved so they have more data to work with, but on their website they state it is not currently available as a service.  The article acknowledges the scepticism in some quarters, in particular the work of Les Perelman, and concludes that there is still “a long way to go”.

Chrysanthi read Learning analytics: help or hindrance in the quest for better student mental wellbeing?, which discusses the data learners may want to see about themselves and what should happen if the data suggests they are falling behind.

Learning analytics can detect signs that may indicate that a student is facing mental health issues and/ or may drop out. When using learning analytics to detect these signs, the following issues should be considered:

  • Gather student’s data ethically and focus on the appropriate metrics to see if a student is falling behind and what factors may be contributing to this.
  • Give students a choice about the data they want to see about themselves and their format, especially when there are comparisons with their cohort involved.
  • Support students at risk, bearing in mind they may prefer to be supported by other students or at least members of staff they know.
  • Talk to students about how to better use their data and how to best support them.

Chrysanthi also read the “What does the future hold” section in JISC Horizon Report Mental Health and Wellbeing, which attempts to predict how wellbeing may be handled in the next few years:

  • Within 2 years, students will have a better understanding of their mental health, more agency, increased expectations for university support and will be more likely to disclose their mental health conditions, as they become destigmatised. Institutions will support them by easing transitions to university and providing flexible, bite-sized courses that students can take breaks from. The importance of staff mental health will also be recognised. New apps will attempt to offer mental wellbeing support.
  • In 3-5 years, institutions will manage and facilitate students supporting each other. Students’ and staff wellbeing will be considered in policy and system design, while analytics will be used to warn about circumstances changing. We may see companion robots supporting students’ needs.
  • In 5 years, analytics may include data from the beginning of students’ learning journey all the way to university to better predict risks.

The Horizon group then gives suggestions to help with the wellbeing challenge, including providing guidance, offer education on learning, personal and life skills to students, and regularly consulting the student voice. Next steps will include exploring the possibility of a wellbeing data trust to enable organisations to share sensitive student data with the aim of helping students, of a wellbeing bundle of resources, apps, etc and more work on analytics, their use to help students and staff and the ethical issues involved.

Naomi read ‘Do Online Mental Health Services Improve Help-Seeking for Young People? A Systematic Review’.

This article from 2014 talks about young people using online services to look for help and information surrounding mental health. The review investigates the effectiveness of online services but does state that a lot more research needs to be done within this area. The article flits between the idea of seeking help and self-help and talks about the benefits of both. It mentions how young people now feel they should problem solve for themselves, so providing an online space for them to access useful information is a great way for them to seek help.

The review mentions how ‘only 35% of young people experiencing mental health problems seek professional face to face help’.  This statistic adds to the fact that online services are needed to provide help and assistance to those in need. It does add that young people do have improved mental health literacy and are better at recognising that they or someone they know may need help. With face to face professional help becoming increasingly harder to access more are turning to online information. It has to be said however that online help has no follow up, and young people can often be given information online, with no way to continue gaining assistance.

One interesting part of the article talked about structured and unstructured online treatment programmes. Although effective at reducing depression and anxiety, structured programmes had poor uptakes and high drop outs with no way for help to be maintained. Unstructured programmes are more useful in the sense that the user could select links that appear useful and disregard to information that seems irrelevant.

This article wasn’t student focused and only covered data collected from younger people, but the ideas behind the review are poignant in a higher education background.

Suggested reading

Jisc Horizon Report mental health and wellbeing section

Or investigate / try out one or more of the online services listed here (or any other – these just seem like helpful lists):

Or related articles