BU Dean Recommends Replacing Striking TAs with AIs (updated)
The actual dean of an actual college of arts and sciences at an actual university has actually recommended replacing striking instructors with artificial intelligence apps.
Last week, graduate student workers at Boston University (BU) went on strike to demand increased stipends and improved benefits.
Yesterday, Stan Sclaroff, Dean of Boston University’s College of Arts & Sciences, emailed recommendations to faculty on how to manage course discussion and lab sections while their teaching assistants are on strike. Among his suggestions: “Engage generative AI tools to give feedback or facilitate ‘discussion’ on readings or assignments” (via The Daily Beast and a source at Boston University).
Dean Sclaroff’s home discipline is computer science.
Readers are referred to to this earlier post.
UPDATE (3/29/24): Here’s the text of the email that Dean Sclaroff sent:
To: Arts & Sciences Faculty and Staff
From: Stan Sclaroff, Dean of Arts & Sciences
Subject: Course discussion sections and labs that are impacted by the BUGWU strike
Date: March 27, 2024
Dear Colleagues, I understand that many of you have questions about how to manage course discussion sections and labs that are impacted by the BUGWU strike. I understand as well that faculty and staff are working to creatively and thoughtfully ensure our students continue to achieve their learning.
Given the disciplinary and pedagogical breadth across the College, there are a wide-range of approaches that can be taken. We know that one size does not fit all. However, in the hope that they can be useful and serve as inspiration for brainstorming, we have provided below some general guidance and examples of approaches that could be utilized during this time.
In general, you may employ all teaching modalities, including holding live Zoom sessions, sharing asynchronous recordings, or combining discussion sections. We appreciate that this openness to remote teaching is a deviation from past University guidance; however, given the circumstances and the need to support our students’ continued learning, these options may be employed. Below, we have listed some creative ways in which, we have heard, some faculty are adapting their course formats and using technology to serve their students.
For discussion sections:
-
- Combine discussion sections (on Zoom, for instance) and record the meeting so that students who are unable to attend can watch them asynchronously; those participating asynchronously can be asked to submit written exercises or complete an additional written assignment to make it clear they watched the discussion
- Assign readings and ask students to write a reflection on these readings
- Consider alternative assignments such as viewing a video/film, visiting an exhibit, or attending a special lecture on campus or in the area on a subject relevant to the course and then writing a short response
- Use Blackboard’s discussion board feature to ask students to pose questions about assigned readings and then have them respond to each other’s questions
- Engage generative AI tools to give feedback or facilitate “discussion” on readings or assignments
For labs:
-
- Deliver some labs through alternative delivery modalities, such as holding online labs or asking student to complete a literature review on conceptual topics
- Reduce the number of in-person labs (total), for instance moving weekly labs to every other week (lab formats vary across CAS, so other adjustments may be more appropriate)
- If appropriate, conduct some labs as recorded demonstrations or blended in-person labs
BU’s Center for Teaching and Learning offers faculty consultation, which can be booked here; they are prioritizing appointments for faculty with questions on these topics.
The Geddes Language Center and CAS IT can also be useful resources for technology enabled approaches.
We encourage you to connect with your colleagues to share ideas and insights about various approaches. Ultimately, you are in the best position to know what will work for your course and students, and we encourage you to think creatively about how you can best conduct your teaching during these challenging times.
Finally, we are launching a process to support replacement coverage requests. If you identify other faculty, graduate students, or staff to cover teaching activities, including discussion sections and grading, these individuals will be paid for their time. A separate email has been shared with chairs, directors, and administrators that explains how to request payment for replacements. Please coordinate with your chair, director and/or administrator to facilitate this process.
As I said in my note Monday, thank you for your advocacy, hard work, and continued commitment to and care for all of our students. Please know that the Dean’s Office and in particular your associate deans of the faculty are available to help with any questions or concerns as they arise.
Sincerely, Stan
—
Stan Sclaroff
Dean of Arts & Sciences
Boston University
Pronouns: he/him/his
Actually, this give me a great idea: let us replace ADMINISTRATORS with AI. The output of AI surely can’t be worse, especially since administrators have been overtaken by corporate speak — and such speech is easy for computers to replicate. It was reproducible even before AI: https://www.atrixnet.com/bs-generator.html
So, let us “competently foster empowered process improvements” by replacing this joker of a Dean with AI. The cost savings will be enormous. Who’s with me?
Rental Agent Replaced By AI For More Empathetic, Human Experience
https://www.betootaadvocate.com/humans-of-betoota/rental-agent-replaced-by-ai-for-more-empathetic-human-experience/
You joke, but I worked in academic administration, and most of my job could (should?) have been automated a long time ago. But my level was much lower than this dean’s, of course.
I’d actually be curious to hear a serious discussion of this as a cost-saving approach to upper administration.
I doubt that most people would be able to tell if 75% of upper university administration were replaced by AI (which I say fully believing that there are some good people who work in upper admin, who are really trying to support the faculty and students).
How about someone runs a test? Obviously, most university administrators wouldn’t go for it, I bet there’s a state senator somewhere out there who would be curious about what would happen if we secretly replaced a couple Vice Provosts with AI for a week.
Related discussion when massive open online courses (MOOCs) were all the rage:
(Emphasis added.)
https://www.mindingthecampus.org/2013/06/13/forget_moocslets_use_mooa/
Heck yeah, this is a great idea!
This discussion has me wondering: Are there any recent real world examples of faculty somehow successfully getting a university to reduce the number of its admins, in light of the modern uni’s obvious administrative bloat? Perhaps by the faculty themselves working their way up to admin and then advocating for other admin to be eliminated?
Has faculty ever “taken back” a uni?
At my university, the administration has been shrinking right along with the faculty. For example, we used to have four associate deans, and now we have two.
I think there are other non-dean administrative bloats, like our “Center for Evaluation”, that nobody really understands. But I’ve been grateful the dean’s office has been responsibly downsizing as our faculty’s been shrinking.
Maybe Hampshire College?
The Unmaking of a College
“This is a film about young people on a rescue mission—for a school and an educational philosophy. The Unmaking of a College delves into the 2019 crisis at Hampshire College when students led a 75-day sit-in – the longest in American college history – at the new president’s office to thwart her underhanded attempt to give up the independence of one of the most experimenting colleges in the United States. From the start, the students were filming events to hold the administration accountable and empower their movement. Alum Goldstein weaves their powerful documentation with interviews with students, professors, whistleblowers, and alumni including filmmaker Ken Burns. A raucous ode to democracy in action, this film evokes the courage required to stand up to power at a time when many liberal arts colleges are failing. Can the storied tradition of liberal arts education—and the nation it has created—be saved? This is a story of young people moved to action by these questions, how they were changed forever by their choices, and how they are becoming our future leaders.”
You can support the strikers here
https://bugwu.org/
and here: https://secure.everyaction.com/jVXiTXlemEmq6k4s9BHWaA2
And stay tuned for ways to help philosophy grad students specifically! (They need the profession’s support.)
I’m a proud BU alum, but this is outrageous.
Boston University was just in the news yesterday as being one of the most expensive schools, at over $90,000 a year in tuition and costs.
With AI instead of people, that would slash the price tag to a fraction of that, right? Right? (We already know the answer.)
https://www.cnn.com/2024/03/27/business/college-tuition-new-england-ninety-thousand/index.html
This dean needs to get fired, least of all for failing to understand the limits of a technology from his home discipline.
You know he’s going to get promoted to university president somewhere, instead.
Here’s my over/under on when this guy gets canned or when that announcement is made: May 15. Which side do you want, Fritz? 🍺
Idk if you still want that action now that we’ve seen the actual email:
The Dean does not “recommend[] replacing striking TAs with AI”, but rather “listed some creative ways in which, we have heard, some faculty are adapting their course formats and using technology to serve their students.”
One of those is to “engage generative AI tools.” He doesn’t even *advocate* for it, so much as apparently mentioning that he “heard” some faculty are trying it.
I’ll take the over and think he’s still in this job past May 15.
I’ll still take the action, despite now being less confident.
But it can very reasonably be interpreted that the dean is endorsing those options; why would he suggest something he wouldn’t want to see happen? E.g., maybe he heard that some faculty are just saying “eff it” and not covering TA duties—would he offer that as a non-endorsed option?
Even if the dean isn’t saying that TAs can be replaced with AI, he’s still saying that some faculty work (discussion sections) can be handled by AI, which is still a silly/premature thing to say, esp. by someone who’s supposed to be familiar with computer science.
We’ll see!
Just a small note to make clear that tons of faculty are not covering TA work not because we’re saying eff it but because we support the strike and it is strike breaking to do so.
That was kind of my thought…
Isn’t the point of a strike to get people to realise that these jobs are important, and therefore, should be compensated properly?
If there are no negative impacts to education, then it seems to defeat the purpose of the strike. Of course, it is horrible to make the students suffer as a result, but there is an easy solution: pay people properly!
just to underline one important point from Patrick: as of right now, you simply cannot “Engage generative AI tools to give feedback. ” no ai can do this straight out of the box. it takes work, lots of work. you have to go watch, like, 8 videos on prompt engineering, learn about fine-tuning. setup a paying account, prep a dataset …. and a dean from cs must know all this.
You do not know Dean Sclaroff: he is the best dean BU CAS has had in 30 years. Humane, supportive of faculty, UG, and grad students, and also respectful of the striking TAs as much as he can behind the scenes. The title of this piece of news is a typical case of spinning and misinterpretation, as the message by Althoff below clarifies. Many of the comments in this thread are un-informed and superficial, as is the short intro to the letter of the Dean by Daily Nous. The letter is very respectful of the learning needs of BU’s undergrads, and does not say a word against the striking TAs.
Good people can make bad choices. This was a bad choice.
this guy is way out of touch. the first faculty to get replaced are going to be from cs, not from phil. and its gonna be tt fac in cs, not ta’s in the humanities.
at my school, tt cs fac get higher salaries because “we have to compete with private industry.” well, no. now we dont. you can learn to code from the ai. it will code for you.
2 cs assist profs = ~8 phil TAs.
As a matter of fairness and fidelity to the truth, I should (as a BU faculty member in the same college as Dean Sclaroff) point out that in a large, college-wide meeting I attended yesterday afternoon, the dean took back this imprudent statement. I am very supportive of the grad student union and their strike, but I do not think our dean should be fired.
The Daily Beast report said there was an email. Do you have a copy to share? Was the email retracted, too—and, if so, on what grounds? Curious to know what their understanding is on why this is such a bad idea, but the fact that it was even proposed at all suggests they don’t understand.
It’s seriously disturbing that a tenured computer scientist would think this was remotely a good idea given AI/LLMs well-documented limitations. And it’s no longer just a BU issue: if this were to happen there, it would set a precedent for others to follow globally, or at least reduce those barriers.
The recommendation was objectionable if only because it undermines solidarity with the strikers, but even so, I didn’t quite take him to say TAs should be *replaced*. He seemed to be echoing what *faculty members* had been doing and, I assume, suggesting ways of mitigating the impact of the strike on learning by utilizing those tools. I haven’t seen the email but the article doesn’t quote any part that makes me think he believes those tools can replace or would be sustainable substitutes for the labor of TAs. Again, even those suggestions may be objectionable, and I would have found them misplaced myself, but more context would help before we call for an imprudent dean to be fired.
Fair enough, and that’s why I’d like to see the entire email myself.
But an imprudent dean is also someone who may have lost the trust of their faculty and students, and trust in leadership is essential for deans and other university leaders.
I don’t know what “taking back” a comment means here; if it was uttered, it’s not like everyone else can pretend they didn’t hear it and question the dean’s judgment. The damage (loss of trust) may have already been done, and that can’t be so easily taken back.
Imagine if he had made a racial slur; can those be simply taken back without consequences?
To take it back simply means he retracts his statement and/or admits he was wrong. That doesn’t mean no damage was done. But it’s a common expression (and speech act). (This may be pure PR and he may have irreversibly eroded faculty and grad students trust, but that doesn’t make ‘I take it back’ meaningless.)
I think this issue is complicated in part because in NO university, but especially no private university, in the US, should students and faculty trust a dean to begin with. The dean is essentially the go between between the purely corporate/financial/image interests of the upper administration of a university and the faculty. But they are much more constrained by the upper administration (the people who have power over them/employ them!) than by the faculty (who typically have little power). So it’s a bit odd, I think, to talk about a dean losing the trust of the faculty and students. I don’t care one way or another whether this dean gets fired, but everyone should be clear that this person is just representing the interests of the upper administration in the email they sent, and from the perspective of the upper administration, the only misstep they made is one of image and not one of actual content. Anyone who would expect anything much better from a dean at a private university is wrongly placing trust in that dean to begin with, and we’d all be better off if we’d understand that even when people go into the job with good intentions and principles, if they want to keep the job, the main thing they have to do is act in the interests of their overlords. Academics would be much better off if we had a more nuanced understanding of the fact that in most cases (not just in private universities, but also in public ones that depend heavily on private $), the people paying us–and those who represent their interests, like our deans–do not have the same interests as us (except insofar as we are interested in being paid). And with respect to students, they would also be better off if they understood that administrations of private universities see them basically purely as customers, and insofar as they want to satisfy their needs it is only instrumental to their own financial interests.
(And, yes, they can do better or worse, be more or less supportive of faculty or students, etc.–but the default should be to have very little trust in your dean, and when they do or say things that seem to go against faculty or student interests, I’m not sure that’s an obvious reason to trust them less–because, in some cases, they are just making something explicit that is otherwise implicit, which is often then used for manipulation or exploitation given the fact that naive faculty and students do trust them.
The joys of hierarchy ought never be underestimated.
The update makes it very clear that the AI stuff was just one small suggestion among many others. More importantly, the dean is supporting faculty who want to replace striking TAs with human strikebreakers. That seems, at least to me, to be a more controversial move, if one wanted to object to the dean, than the relatively anodyne bullet point about AI tools.
I agree this is the really important issue. The CAS Dean is following the Provost’s lead in insisting that faculty do the work that the graduate student workers would otherwise be doing or find some way of employing scabs. The messaging is that if we don’t do this we don’t care (enough) about our undergraduates (we have received emails saying as much), but in suggesting some of the very subpar options mentioned in this email, at the same time as making the negotiations go very slowly, the administration is actually providing evidence that it’s the administration that doesn’t really care (enough) about the undergraduates.
It is an all too common move of (union busting) administrators to try to pit faculty and / or graduate students against undergraduates — and it is an extremely cynical move on their part: “Help us exploit the graduate students because otherwise we won’t be able to exploit the undergraduates to the tune of 90k a year…” I renew my proposition that this turkey of an administrator ought to be replaced with AI.
I think the concern here is a bit overblown – even if the striking TAs were replaced by bots; if the bots did not perform well enough, they would surely be replaced by human TAs again (simple supply and demand). On the other hand, if the bots did perform just as well as the TAs, then we should rethink whether it’s so outrageous to automate the TAs. (In philosophy, the part that can’t be automated is the serious research — simple discussions about basic facts about, say, Mill and Kant’s ethics to a tutorial group of 1st year undergrads , it seems like an open question whether an LLM bot would be able to handle this. As I say, if not, then they’ll get replaced with humans, if so, then save money and time and automate the discussions (and let the TAs pursue something more worthwhile, like their own research).
This take really undersells what it is to be a good teacher and to guide students through philosophy – yes, even when it comes to 1st year undergrads, which is by far the largest population of students philosophy teachers encounter. For many of us as well teaching these students is no less worthwhile than pursuing our research.
If I could give this a billion thumbs up I would.
“For many of us as well teaching these students is no less worthwhile than pursuing our research”. I agree 100% that many feel this way. The problem is that (for reasons noted in my comment below), given that it’s increasingly likely that bots will be able to do exactly as well as humans with first year tutorials (at least on certain material LLMs can be extensively trained on, e.g., utilistarianism, categorical imperative, etc.), people who like or derive meaning from doing those tutorials will (for better or worse) realistically be freed up to do other things. Or they should at least be prepared for this eventuality. I said ‘research’ above (as a salient thing people would be freed up to do) – but that might be too optimistic; associate deans, etc. will probably see this gap freed up in the workload and reallocate admin across this space.
it has never been the case that the introduction of automation leads to less work. it leads to more, different work. replacing typewriters by word-processors did not reduce the secretary’s work hours, it gave them additional responsibilities.
any time savings gained by ai use will surely be recaptured by admin.
good news, you dont have to grade papers anymore! so now you can teach 3 more sections for the same pay.
Tell that to Eli Whitney (he is, by the way, the inventor of the cotton gin)
Surely if anything proves Mario’s point, it is the cotton gin.
Martin,
I’m curious about how well you understand how LLM bot work. Assuming that you are knowledgeable about their basics (which is all I know), how do you imagine such bots to even approximate a human TA?
Since the LLMs have no actual intelligence or understanding of the material and simply produce algorithmic text trained on web sources as they pull web info from largely non-academic sources such as Wikipedia, it seems to me that you have a real problem with accuracy of what the LLM produces.
Hi Ian, fair question. So, I do agree an LLM bot (say, ChatGPT4) would have accuracy issues with quite a few kinds of philosophy discussions. The reason I said it’s an open question about first-year discussions is that when it comes to questions about what Kant said or what Mill said about ethics (basic first year material), they are very reliable – simply because they have been trained on tons of material about this. I wouldn’t trust an LLM to have a TA-style discussion on content on which it’s not been trained on an enormous amount of data. But (and this is why I’d say it’s an open question) if the goal of leading a TA discussion is just to accurately guide a discussion about utilitarianism vs deontology in ethics, there really is not a reliability issue there. As for how the LLM woudl facilitate the tutorial – it woudl need to be present in class and so (I should have said this initially) the way forward woudl be via the new OpenAI speech-to-speech robot, which can also notice when students hands are raised, collect papers from desks, etc. Not kidding at all – watch this video https://www.youtube.com/watch?v=Sq1QZB5baNw
Maybe your experience is different from mine, but so-called hallucination remains an issue with LLMs and I suspect it’s one that won’t go away. Certainly an AI chatbot may successfully tell you what Kant said (say by quoting a translation of one of the categorical imperatives) but the issue of training and algorithmic text remain an issue for two reasons.
First, as far as I know the companies creating these chatbots don’t quite yet know what is going on “under the hood” so to speak, which means that when the LLM “chooses” to be accurate and when it “decides” to just generate in a way that is divorced from reality aren’t yet predictable. So even if we suppose that the data an LLM is trained on is correct (always problem with the internet), we can’t predict or prevent it from hallucinating responses and sources.
Second, LLM training is based on volume and not, yet anyway, quality. So let’s say there are multiple readings of Nietzsche, for example, on the English language web. Just to be programmatic, let’s say that there are four major competing readings: 1) that of Nietzsche’s sister Elisabeth, 2) that of Walter Kaufmann, 3) that of Brian Leiter, and that of 4) “Bronze Age Pervert.” How would the LLM be able to distinguish in any useful way between these other than to say (and this would be at best) “there are four major readings, which are” (and then accurately represent them)? Since it doesn’t actually know or understand any of this–nor is it capable of forming a view of the strengths and weaknesses of these interpretations–it’d seem like a kind of malfeasance to use it as a TA-replacement.
On AI replacement of TAs:
One ‘nice’ thing about TA’s work, and the work of most academics in a teaching capacity, is that a majority of it is not recorded and, is not recorded specifically as text. Lectures, discussions, meetings with students, office hours, etc are conducted verbally and are, presumably, not recorded let alone recorded in a text based way. Consequently, the work output of people in position’s like this can’t be ingested by AI which can then be used in lieu of them.
This differs from many other forms of work. In many corporate settings, much of the day to day activity is conducted over Slack/Teams/Whatever else such that a ‘fine-tuned’ LLM could be trained on the ‘data-set’ that is the, e.g., Slack, contributions of a senior person that has been with the company over many years. For example, a principle software developer that moves from team to team solving difficult engineering problems.
Instead of paying that developer 200k/yr to solve engineering problems over Slack/Zoom, just train a LLM on their text based data and it might do nearly as well as the human.
BU has plentiful (if under-appreciated) TA labor resources, so AI replacement sounds pretty fanciful there at the moment. There is a lesson here about the future of academic philosophy, however. As regional (i.e., non-elite and non-R1/R2) institutions continue to close down their philosophy departments in order to reduce the number of tenure lines for both budgetary and governance reasons, these institutions will meet the teaching needs of philosophy service courses not only with adjuncts as they presently do, but eventually with AI-taught courses. Currently it happens to be cheaper to pay adjuncts to teach philosophy while AI has not been developed/applied for that purpose to any great extent, but this will change. (I said “governance reasons” above: tenured faculty can stand up to admins, adjuncts are vulnerable but still capable of resistance, while course-teaching AI simply does what it’s told.)
Some of the discussion here centers on whether or not LLMs could replace the functions of TAs given their limitations. I think the limitations of LLMs are beside the point. Let’s suppose that an LLM could recreate everything I do in weekly discussion sections–every handout I make, every guided discussion, every piece of writing advice, etc. And let’s suppose that the LLM would be “present in class” and is able to replicate all my physical interactions. It is an ideal LLM. I find it doubtful that students would have the same experience with the LLM compared to the live TA. From my experience, both as a TA and as an undergrad, social interactions play an important role in philosophy discussions. Interactions with an AI are fundamentally different than interactions with a person. For example, I may share why I find Kant fascinating to read, and this may inspire my students because my interests resonate with them. An AI can say the exact same words, but our social empathy processes the words differently when they come from a human.
I majored in philosophy because I had a TA who inspired me by being friendly, engaging to talk to and by sharing my interests and motivating me to take advanced philosophy courses. I looked up to my TA because I thought that I could be like them and have their expertise if I tried hard. Even if an LLM could recapitulate those exact same 1-on-1 discussions with my TA, I don’t think it would have the same effect. I cannot look up to an intelligent LLM.
That being said, a class fully ran by LLMs would have much more value than, e.g. asynchronous fully recorded philosophy classes. And I bet there are TAs/Professors that provide a worse experience than an ideal LLM. But the point of a college level philosophy course would not be fully served by an LLM.
Actual philosophy professors in my actual department actually used AI apps (PackBack) to do grading for them and give feedback to students on discussion boards. I found out because I got on a call with a rep to ask whether student data is collected by this private company (FERPA?), and she told me which professors in my department were already using the service.