Evaluating AI
Exploring ways to support informed decision-making about AI
Hello after an inadvertent blog hiatus! Between travel and a poorly timed cold, I was away longer than anticipated. Nevertheless, I’m back and ready to dive into another part of my small series focused on the AI Competencies for Academic Library Workers. Continuing my contrary tradition of going out of order, we’ll be looking at competency area 3 today: Analysis and Evaluation.
Photo by Marco Kaufmann on Unsplash
Evaluating Output and Evaluating Tools
Being a librarian, when I first read “Analysis and Evaluation” I immediately thought of source evaluation, a topic that librarians teach frequently. Source evaluation involves looking critically at a piece of information and determining whether that information is factual, reliable, worth using for a research assignment, or worth sharing on social media, as the case may be. This sort of evaluation is a bedrock of information literacy skills.
But I was pleasantly surprised to see that this section of the competencies was actually talking about evaluating AI tools themselves in order to “critically assess their application.” In other words, asking questions such as “Does this AI tool work? Is it useful? Are there benefits or harms associated with using it?” Or is this AI even an AI or rather a group of humans pretending to be an AI, in the equivalent of a person dressing up in a robot costume and saying beep boop? True story actually! If you aren’t familiar with the wild tale of Builder.AI, please take a moment to familiarize yourself with the brazen (and to my mind, somewhat hilarious, albeit in an awful scammy way) scheme, which I mentioned briefly in my first ever post on this platform. Fond memories. Anyway, moving right along.
Capabilities, Ethics, and Implications of AI Technologies
While it is important to evaluate the output of an AI tool, that kind of approach can assume that we should be using a generative AI tool for the task at hand in the first place. The evaluation methods outlined here are valuable, in my opinion, because they emphasize asking critical questions about whether or not an AI tool is appropriate or optimal for a task in the first place.
The competencies outline a few lenses for analyzing AI technologies.
Consider technical capabilities of AI technologies, including accuracy, relevance, and robustness of their performance.
Consider ethical aspects of AI technologies, including transparency, explainability, biases and fairness.
To see the implications that use of AI technologies has for learning, development of critical thinking, and research skills.
The first consideration, technical capabilities, certainly makes sense and is something that people could achieve through observation. At the most basic level, the tool either works for what you are trying to do, or doesn’t. This area involves determining if AI is the best fit for a given task, or if it can even do what you are asking it to do. However, this can be difficult to determine at times amidst the relentless churn of AI hype, which often argues that AI can do anything and everything.
The second two areas here are trickier though. Considering the ethics of AI begs the question as to whether you should use AI at all, regardless of how well it performs. Is AI performance incidental in light of ethical concerns, or does the usefulness of a tool sway the decision-making process here? And considering the implications of AI technologies on things like critical thinking involves a degree of research and background knowledge. You need to informed about the trends, research, and conversation around areas like learning and critical thinking skills in order to utilize these areas as an aspect of the decision-making process about using AI. While the technical capabilities can be determined with trial and error, or observation, the ethical aspects and implications on learning require further research and study as well as reflection and complex decision-making.
Lateral Reading and Learning Communities as Ways Forward
I was really intrigued and glad to see the emphasis on decision-making around using or adopting AI in the first place included here, but I do wonder what sort of support librarians (and others) need to engage in this kind of work and this kind of informed decision-making. Here are a few ideas and thoughts that came to mind for me.
First, I went back to my initial read (or misread in this instance) of evaluation referring to the evaluation of AI output rather than AI tools. There are various techniques and acronyms to guide people through the source evaluation process, but one strategy that I find incredibly powerful is something called lateral reading. This technique involves reading horizontally, as it were. Essentially, rather than read something in detail from top to bottom, you open up new tabs and go outside of whatever it is you are reading, and whatever platform you are on, to see what others are saying, to check up on the author’s credentials, to fact check a claim, to see where a link goes, etc. I wonder if adopting lateral reading techniques to situate AI in context can help people make more informed decisions. Basically, I am interested in ways to “read” AI tools. Rather than just using the tool and seeing if it works, we can adopt lateral reading approaches and research what people are saying about a given tool, the media coverage it is receiving, who developed this tool, the ways a tool operates in the broader landscape of AI hype and AI media coverage, etc. The Competencies here place an emphasis on critical thinking and a deeper and more contextual understanding of AI. Given that librarians often teach lateral reading techniques as part of their work, I wonder if exploring lateral reading for tools rather than just content could be an opportunity to bring elements of the work that librarians already do into the ways they engage with AI tools.
Second, I’ve been quite interested in researching ways to utilize communities of practice and other learning community approaches to navigate complex and difficult topics like AI. Librarians and other educators today are frequently asked to grapple with incredibly complicated topics in their classrooms, with AI being one of these topics. If the Competencies asks that librarians consider ethics and issues around AI and learning when making decisions about using AI technologies, then how can librarians effectively obtain the knowledge needed to make thoughtful and informed decisions? Self-study is an option, of course.
But given how overwhelming and fast-moving AI developments can be, I think a more communal approach might be recommended here. Whether through training, workshops, learning communities, discussion groups, or communities of practice, I think we have an opportunity to lean into professional development and learning opportunities that empower librarians (and others of course – I think this applicable to many fields!) to make informed decisions about using AI, to engage in conversations about AI with stakeholders, and to advocate for AI usage policies that reflect the profession’s values and views, as well as current research and best practices around learning.
Decision-making around AI technologies is incredibly important but it doesn’t happen easily. Librarians, other educators, and anyone really, put in situations where they are having to make decisions about AI technologies need strategies, knowledge, resources, time to learn and experiment, and support. I am glad the Competencies highlighted the importance of evaluating AI tools and technologies and of making thoughtful decisions about using (or not) AI. And I’ll be interested to see what kinds of support are needed to equip librarians and others to actually make these kinds of critical decisions around AI technologies.
But in the meantime, I think we all deserve a break. I wish you all a very happy holiday season whatever and however you celebrate!