At York University management tells us that Microsoft Copilot can "[enhance] your productivity with our advanced AI tools—YU AURA and Microsoft Copilot. These services streamline tasks, improve efficiency, and support your everyday activities." (source). Policy documents exhort the work potential for Copilot and Copilot Premium, with helpful tips about guardrails and data types:
It's hard to not get the impression that management is firmly of the opinion that the Copilot AI tools are the bees knees when it comes to productivity tools. Turns out, though, that Microsoft disagrees.
"Copilot is for entertainment purposes only"
Microsoft, itself, is unequivocal in its terms of use: "Copilot is for entertainment only". (source). No ifs. No ands. No buts.
Let that sink in a moment. Microsoft is not saying "for work, too" -- Microsoft is saying that it's for fun and games ONLY.
And, of course, this isn't just some random professor saying it. Many mainstream media outlets, including Tom's Hardware, PCMag and The Register have confirmed Microsoft's admission:
LLMs are just glorified predictors -- for the easy problems
As remarked upon in The Register's story about Microsoft Azure's stability issues, the rush to replace people with LLMs is revealing negative impacts at both primary and secondary levels. What layoff-happy managers don't realize is that LLMs aren't magical thinking machines. They're just "very good at reproducing patterns" and so even seemingly amazing tasks they perform like finding errors in code or data is by "observing deviations from their probabilistic expectations".
When a student discovers that a chatbot is able to write an essay or Python coding homework solution it's not because the chatbot is smart -- it's because the chatbot assimilated millions and millions of lines of code and text written by humans and posted online. When an office worker tells the Zoom AI assistant to take notes it's only able to do so because it has already been trained on curated meeting minutes. Likewise for all the pie charts and white papers that managers share with each other. These are all mundane, routine tasks. There's nothing novel about them. They're low-hanging fruit, intellectually.
But by routinely outsourcing our cognition to the chatbots we're forgetting that we need to hone those same skills to work on the harder, more challenging -- and rewarding -- tasks that chatbots can't do. The can't do them because they can't be trained on the hardest, most novel problems... because the data simply isn't there for them to be able to suck up into their models.
Universities are supposed to be bastions for thinking -- at all levels. We need to keep the chatbots at bay because we need our students to practice thinking, starting with easy chatbot-able tasks and then working their way up to the problems and concepts that aren't chatbot-able. But it's not just about the students. If university staff and faculty routinely offload their cognition to chatbots then that will permeate down to the student experience, too. We need to set the example, because in higher education an uncritical adoption of AI is unacceptable.
Remediation of digital asbestos
There is no doubt that lots of employees and students are using AI tools here at York. Whether it's AI assistants for recording Zoom meetings or chatbots to answer homework questions and write essays, we're hooked.
I've seen AI referred to as "digital asbestos". Useful, but dangerous. And incredibly hard to remediate.
It's going to take a massive effort to fix this, but it will start with trying to get people to read Microsoft's warning.
Updates
April 5: new section (LLM predictors) added about MS Azure, instability and the difficulties that LLMs have with respect to novelty. Link to Dr. Olivia Guest et al's work and reference to digital asbestos added.
James Andrew Smith is a Professional Engineer and Associate Professor in the Electrical Engineering and Computer Science Department of York University’s Lassonde School, with degrees in Electrical and Mechanical Engineering from the University of Alberta and McGill University. Previously a program director in biomedical engineering, his research background spans robotics, locomotion, human birth, music and engineering education. While on sabbatical in 2018-19 with his wife and kids he lived in Strasbourg, France and he taught at the INSA Strasbourg and Hochschule Karlsruhe and wrote about his personal and professional perspectives. James is a proponent of using social media to advocate for justice, equity, diversity and inclusion as well as evidence-based applications of research in the public sphere. You can find him on Twitter. You can find him on BlueSky. Originally from Québec City, he now lives in Toronto, Canada.