University Suspends Students for AI Tool It Gave Them $10k Prize to Make | Hacker News Reader

University Suspends Students for AI Tool It Gave Them $10k Prize to Make

404media.co

68 pointsdanso43 comments

You can get a tutor to check your homework and highlight your errors. Sometimes they just tell you answers in the process of teaching. AI can be a tutor if that is what you ask for.

Homework is for using every tool at your disposal to learn. Tests are for proving you actually learned. Grading homework for credit and not giving feedback until after the test is a pattern I have seen too many times in higher ed, and it is a sorry excuse for education.

> In each of these instances, it is becoming clear that schools do not know what the rules about AI should be and that they are often being made up on the fly.

This seems a little uncharitable to me considering society has no idea what the rules around AI should be and is therefore making them on the fly too. Why do we expected school administrators to be so far in front of everyone else on these issues?

Universities and the professors there, often act as judge, jury and executioner on their fiefdoms. It's not like most students are wealthy enough to afford to sue the university for failing to objectively apply the rules they themselves set.

This student was, from TFA.

But this student is definitely not most.

No, most students don’t win entrepreneurial competitions and then get suspended. I guess I’m missing the point.

We expect school administrators to be ahead of others on this issue because of the power imbalance with their application of the policies to students. We're hearing about it because it happened to an anomalous student who was able to fight back. The initial reason remains even though we only read about it through the anomaly.

I.e. the point of this thread isn't around the specific case of the anomalous student, rather the issue raised around AI enforcement in schools and what it means for the typical student. And yes, average IT students will still get involved with situations around AI and their school - they just aren't going to make the news with a court case.

Each of us should be consistent in our actions and charitable in our interpretations, even if society at large is chaotic, inconsistent and sometimes quick to judge harshly.

That applies to universities as much as to your employer and to me personally.

University/college administrators are deeply terrified about AI. They are fully aware of the exclusive role that respected institutions play as gatekeepers of accreditation. They know that the entire value of a diploma or degree from their school is completely founded upon the public perception of its legitimacy. AI threatens to totally erase that legitimacy by making it trivial for anyone to cheat on their assignments, thus preventing the school from properly assessing student learning.

Some of this danger can be mitigated by switching assessment exclusively over to in-person, proctored, closed book exams. However, this does not work for many courses in the humanities that are intended to teach critical academic writing. A 3 hour closed book exam is totally inadequate for assessing the student's ability to conduct research and write a long, critical essay. Similar issues exist for many other fields, particularly as you go deeper into specialization in a field where an examination covering basic facts and skills cannot suffice.

Users should have access to their own data. Whether it’s a university coursework app or anything else.

These students should be applauded. They are making studying easier. How are universities, of all places, so inept at making basic common sense decisions?

It was connected to a platform which hosted course reading material.

That's not the students data.

The students were intended to have access to the course data. They did not exceed their authorized access.

https://archive.is/niWzb

If you can receive a passing mark for AI generated content, that's principally an indictment of the evaluator.

Sure, but that’s not what was happening.

How does Eightball.AI work?

Is it using OpenAI in the backend so OpenAI gets all the data for free?

There's been a myria news regarding AI-related incidents in the academic settings. One constant in most of these stories appear to be that most educators have no interest in educating themselves about AI.

What bearing does this have on the article?

My comment was directly influenced by the article.

In the sense that they both contain the word "AI?" Otherwise, I'm not really seeing the connection to the article. I.e.,

> One constant in most of these stories appear to be that most educators have no interest in educating themselves about AI.

This article does not discuss or show this.

It explicitly says that the lawsuit filing showed a distinct lack of understanding by the University of what LLMs are/how they work.

Does it say the lawsuit was filed by most educators, or anything about their interest in educating themselves? This represents a minority of educators and says more about administrators than educators. And says nothing about the broad claim GP is making about their interests.

There’s still hope! We recently published a paper where we interviewed instructors about their concerns of AI tools in the classroom and then we deployed one.

See here: https://austinhenley.com/blog/codeaid.html

Am I understanding that they trained their model on the real course materials, including specific assignments, exercises, and even the actual tests, without the institution's permission?

I don't see a problem with rescinding the prize and punishing the kids for dishonesty.

It's not a free-for-all no matter how much OpenAI wants it to be.

Yeah, it sounds like they were upfront about connecting it to the system in the original competition, but never actually received permission to do so after that. They just assumed they had permission.

The "it can be used to cheat" thing is just dumb and is totally on the university for selecting them in the first place if that was a problem.

> They just assumed they had permission.

Being given money by the system's owner to do it does sound a bit like permission to me...

i imagine that the people giving the money out and the people who then claim not to have given permission might be different persons. still, a lawsuit on that premise will probably result in a nice settlement.

We could extrapolate that the diplomas given out by said university have no value because some other department could decide after the fact to cancel them :)

the degrees are conferred by the highest authority at the university, the board of regents/trustees. the only reason they would rescind a degree would be evidence of academic fraud/plagiarism in the course of earning the credits for the degree.

That's the theory, yes. But the institution has already shown it can arbitrarily change its mind...

To which “dishonesty” do you refer?

The dishonesty of building exactly what you said you would build and being forthright about how you accomplished it such that a technicality of how you accessed the data gets you suspended.

The article quotes them as planning to train their model on course materials automatically. But it is not clear if that actually happened. Individual students were always free to upload course materials.

I'm not sure what is dishonest about that.

Edit: I'm not sure if the article has been updated from when I first skimmed it, or I just missed it, but this seems directly addressed now:

> The students were also accused of “disseminating course material” by allowing students to use their own, individualized Canvas API tokens to connect Eightball directly so they did not have to upload PDFs to the tool themselves. “The upgrade did not change what students could upload to Eightball or what learning materials Eightball could produce upon request,” the suspended students’ lawyers note in the lawsuit.

It is not really clear to me if it had privileged access to answers or if it saw the same material a student would see reading through the syllabus + assigned readings.

It's the latter. It sounds like they got suspended because they told their users (i.e students) to give them their canvas API tokens so that they could fetch course materials on their behalf.

Previously, students had to download the PDFs and manually upload them. If students gave eightball their API tokens, eightball could fetch their course materials on their behalf.

Perhaps I'm missing something, but that's how it reads to me.

From TFA: <<university’s IT department was angry that the company allowed students to connect their own Canvas API tokens to the app>>

Just a wild guess, but maybe the IT department gets charged an additional fee from Canvas based on the usage of API tokens?

Conflicts like these are always about money, one way or another.

Conflicts in bureaucracies like universities are often petty power struggles rather than anything financial. Especially between IT departments and anyone else.

> give them their canvas API tokens

in pretty much every single system's terms, providing someone else your tokens or other forms of sharing of credentials is expressly forbidden. this is why I am flabbergasted things like Plaid work. I will never provide a 3rd party my credentials to anything but damn sure not something like a bank or credit card.

There are many systems that allow generating a API token to be used with other apps. The article clearly indicates that this school learning system, “Canvas”, had such a functionality until it was removed during this saga, apparently by hiding it with CSS or something since there was an “inspect element” workaround to still access it.

Plaid uses OAuth for many of their connections and doesn’t violate ToS in that case. There usually are TOS related to not sharing your password and disclaiming liability if you do.

3rd party access is the entire point of API tokens. Use cases like this are why Canvas has them.

I think it is pretty clear they did not have privileged access. Either because they never connected to the courseware service, or because they did so with students' credentials.

Edit: I'm not sure if I just missed it earlier or the article has been updated, but this issue is directly addressed now:

Student could upload their own material, which I think is fine as long as you use it personally. The intention was to connect to the school system with all course materials, but that never took place.

No, I don't think you're understanding correctly. The students first pitched an idea explicitly based on users uploading "specific assignments, exercises, and even the actual tests" to their tool, and the institution approved this idea and gave them $10,000 to build it.

The change that prompted the suspension seems to be about the technical details of how users upload those files - using the Canvas API instead of manually downloading and uploading. Either way, there is no dishonesty alleged by anyone, even the school Honor Council.

Also, the tool did not use OpenAI or any other third-party AI service, so whatever opinions you have about OpenAI are not relevant.

Reply on news.ycombinator.com