In one sense Generative AI for Academics is almost finished. In another sense it is spiralling out of control because the only bits left are the immensely tricky sections on ethics and politics which I have been determinedly avoiding. The remaining bits of the manuscript are a matter of filling in a few blanks and tidying up now. But I realise that my avoidance of the ethics chapter (largely I suspect because at least one of my internal voices suggests the whole thing is just a disaster environmentally and politically, raising the question of why I am writing a book encouraging people to use it) is leading the rest of the manuscript to balloon.
I’m adding stuffing in elsewhere to avoid tackling the bits I’m ambivalent about, creating more need for tidying up and taking me further away from finishing the book. This is starting to remind me of how I made such a mess with the second edition of Social Media for Academics which is an experience I was determined to learn from. This blog post is an act of public accountability: this week I will work out how to turn the ethics chapter into something coherent and stop fiddling with the rest of the book!
Perhaps inevitably given the subject matter I just turned to Claude AI for assistance. I provided it with a detailed explanation of this chapter and the broader book, alongside (a) 6000 words of disjointed notes on ethics (b) the complete drafts of the previous two chapters. I asked it to provide section headings and bullet points which capture the topic of my notes, organised in a way which reflects their internal logic and which builds upon the arguments of the previous two chapters. This is pretty damn good and exactly what I needed to make this section tractable, after feeling like I was drowning in my own notes and unable to get traction on writing this chapter:
Introduction
- Bridge from prior discussion of principles and process view of scholarship to focus on ethics
- Emphasize these issues arise from your own exploration of using generative AI reflectively
Section 1: Environmental Impacts
- Energy usage contradicting climate goals of higher education
- Individual use contributes to broader harms of large models like GPT-3
- Explain environmental costs when introducing new tools to colleagues
Section 2: Data Ethics and Privacy
- Lack of consent in web scraped training data
- Potential to perpetuate inaccuracies and biases
- Surveillance capitalism business models behind many tools
- Transparency enables informed decisions about tradeoffs
Section 3: Algorithmic Bias and Representation
- Risk of amplifying exclusion and marginalization
- Hard to audit biases without transparency from creators
- Issues extend beyond text to images, audio and video
- Surface and mitigate biases through sustained dialogue
Section 4: Cultivating Responsible Use
- Avoid overstating capabilities to students and public
- Consider unintended consequences and second order effects
- Foster collegial cultures supporting reflective use
- Integrate ethical prompts into your own generative dialogues
Conclusion
- Ethics of care and responsibility vital in era of generative AI
- Process view supports using tools to amplify values, not substitute judgment
- Intentional cultivation of reflective practices key to mitigating risks
Mark Carrigan 