Latest / Never Post / Is It Even Possible To Ban A.I. In The Classroom?

Transcript

Is It Even Possible To Ban A.I. In The Classroom? · Never Post

0:10
Mike Rugnetta: Friends, hello, and welcome to Never Post, a
0:13
podcast for and about the Internet. I'm your host, Mike
0:15
Rugnetta. This intro was written on Tuesday, 09/23/2025 at
0:20
11:56AM eastern, and we have a slightly late show for you this
0:25
week. Sorry about that. I got a toddler cold, and we're in the
0:28
middle of renegotiating AI Mike's contract, but we're here.
0:32
We made it. We did it. In our third ever show length segment,
0:37
contributing producer Tori Dominguez Peak returns to look
0:41
at artificial intelligence in the classroom, including one
0:44
instructor who has said no more and has banned its use entirely.
0:50
Tori tells Jason what that entails and tackles the old
0:54
Kennard is writing, thinking, and also bop spotter. But right
1:02
now, we're gonna take a quick break.
1:03
You're gonna listen to some ads unless you're on the member
1:05
feed. And when we return, we're gonna talk about a few of the
1:08
things that have happened since the last time you heard from us.
1:13
Hello? Is it five stories for you this week you're looking
1:16
for? YouTube is reinstating the channels of creators previously
1:22
suspended for violating COVID nineteen and election
1:24
disinformation guidelines an alphabet lawyer says are no
1:28
longer in force.
1:29
This according to a document obtained by Fox News and
1:32
prepared for the US house judiciary committee. Reflecting
1:35
on the company's commitment to free expression, Daniel f
1:38
Donovan, counsel for alphabet rights, YouTube will provide an
1:42
opportunity for all creators to rejoin the platform if the
1:45
company terminated their channels for repeated violations
1:48
of COVID nineteen and elections integrity policies that are no
1:52
longer in effect. YouTube takes seriously the importance of
1:55
protecting free expression, the document states elsewhere, and
1:59
access to a range of viewpoints. The document also explicitly
2:02
points out that YouTube has not, does not, and will not employ
2:06
any kind of fact checking or labeling mechanism in its
2:09
software. At time of writing, no list of the channels to be
2:12
potentially reinstated has been published, but I bet it's not
2:16
that hard to figure out who might be on it.
2:18
In completely unrelated news, Alex Jones recently appeared on
2:24
his show sporting a Hitler mustache about which he said, I
2:29
could tell you it had a wild effect on women.
2:31
Jason Oberholtzer: Ew.
2:33
Mike Rugnetta: The US Secret Service shut down a high powered
2:36
cellular network that they claimed posed a threat to tri
2:39
state area mobile communications this week. CBS News reports
2:43
that, quote, law enforcement discovered 300 SIM servers over
2:46
a 100,000 SIM cards, enabling encrypted anonymous
2:49
communication and capable of sending 30,000,000 text messages
2:53
per minute that could have, again, allegedly disabled cell
2:56
phone towers and launched a distributed denial of service
2:59
attack with the ability to block emergency communications like
3:02
EMS and police dispatch, end quote. Secret service claims the
3:06
operation was well funded and possibly under control of state
3:09
actors looking to cause trouble for UN week in New York.
3:13
Independent tech auditors and security analysts are not so
3:16
convinced. Well funded, yes, but capable of causing such
3:19
widespread havoc in New York City of all places, not so much.
3:23
There is nothing about this infrastructure that would be
3:25
hugely disruptive or damaging to mobile phone networks, writes t
3:29
profit, the self described telecom informer for Hacker
3:32
Magazine 2,600 on Blue Sky. BookTok has managed to shoot
3:41
Timothy Snyder's lean but weighty 2017 book on tyranny to
3:46
the top of indie bookshop sales lists over the last few months.
3:49
On Tyranny's bullet point style format and short chapters,
3:52
writes Laura Miller for Slate, make it easy to break into
3:55
nuggets of exhortation. A particular favorite is lesson
3:58
number one, do not obey in advance, urging individuals and
4:03
institutions not to appease authoritarian governments before
4:07
they are even asked to. Some fans on TikTok temporarily turn
4:10
over their accounts to On Tyranny, reading one chapter
4:14
aloud per video until they've narrated the whole thing.
4:18
Speaking of TikTok, an alleged so called framework deal has
4:22
been penned regarding the sale of the Chinese owned platform to
4:25
domestic concerns. The US government and ByteDance have
4:29
brokered a forthcoming deal whereby Oracle, Silver Lake
4:32
Technology Management, and Andreessen Horowitz would
4:35
oversee TikTok's US operations. This group would have an 80%
4:39
ownership share, and a member of the board would be appointed by
4:42
the US government. President Trump has also suggested Fox
4:45
News Baron Rupert Murdoch will likely be involved somehow. The
4:49
US based owners would lease TikTok's infamous algorithm,
4:53
which Oracle would oversee and, quote, retrain.
4:57
Larry Ellison, CTO and founder of Oracle, has also recently
5:01
financed a number of large scale media mergers with his son,
5:04
David Ellison. Paramount Skydance controls CBS, Paramount
5:08
Pictures, and the streamer Paramount Plus. The Ellisons are
5:11
also allegedly eyeing a takeover of Warner Brothers Discovery,
5:14
which owns, among other things, CNN. And finally, get ready for
5:28
a really good sentence. You ready?
5:29
Ready for this good sentence? Here we go. Limewire, relaunched
5:34
as an NFT marketplace, has purchased the rights to the
5:37
infamous fire festival brand. The New York Times reports that
5:43
the purchase was made for 245,000 US dollars in an eBay
5:47
auction. It is unclear what Limewire fire will become.
5:51
The music downloader turned NFT peddler is apparently aiming for
5:54
something that, quote, expands beyond the digital realm and
5:57
taps into real world experiences, community, and
6:00
surprise, a thing which no doubt aligns well with the fire
6:04
festival brand. Ew. In show news this week, if you ordered a t
6:11
shirt, they are being printed next week. Once the print is
6:14
done, they will head to Neverpost HQ where they will be
6:16
packed and shipped one by one by hand with love. We will also
6:22
have very few stock designs available at the end of that
6:25
process.
6:26
I'm gonna let you know in the show news portion of future
6:29
episodes when and where you can snag those if you missed out.
6:32
But when I say very few, I really mean it. We're gonna
6:35
have, like, fewer than 10 stock shirts. And finally, holy cow,
6:40
we are a signal podcast award finalist in the technology
6:45
category. That is fun.
6:47
Heck yeah. If you could please go vote for us, we would love
6:51
that. We'll put a link in the show notes. We are up against
6:53
some really rad folks, including close all tabs, who you may
6:57
remember from our hentai segment and kill switch of whom we're
7:00
just generally fans. But please go vote for us.
7:03
We will love you forever. Signal awards, technology category,
7:06
there's a link in the show notes. Okay. That's the news I
7:09
have for you this week. In this episode, Tory talks with Jason
7:12
about AI in the classroom.
7:14
But first, BopSpotter is a project by Riley Walls, and it's
7:19
described this way. Somewhere in the Mission District of San
7:25
Francisco is a microphone pointed down at the street
7:29
below. It is using a Shazam manner. So in our interstitials
7:48
this week, Hans took it upon himself to recreate what he
7:54
imagined to be the sonic environment at the time of some
8:00
spotting. So what you are about to hear are not field
8:04
recordings, but carefully crafted audio collages
9:09
Jason Oberholtzer: Midnight, 12AM. 03:30AM. So I'm sitting
10:32
here at my desk today, watching the leaves slowly change color
10:36
when an email comes in, From friend of the show, Tori
10:39
Dominguez Peek, who submitted to us a year ago a piece you might
10:42
remember, wherein AI chatbot companies reached out to her
10:47
with the proposition of turning her deceased mother into an AI
10:50
chatbot. Well, Tori is back with another piece. I asked, what's
10:56
it about?
10:57
No one would let me know. Tori wanted to tell me herself. So
11:00
please, welcome back to Neverpost, Tori.
11:04
Tori Dominguez-Peak: Hey, Jason. Thanks so much for bringing me
11:06
on.
11:06
Jason Oberholtzer: I'm excited to learn what I'm about to
11:09
learn.
11:10
Tori Dominguez-Peak: So today, I have a tale for you about AI on
11:15
college campuses and Brazilian Portuguese and solving crimes.
11:24
Jason Oberholtzer: All three of my biggest interests. Let's get
11:29
started.
11:29
Tori Dominguez-Peak: Story in three acts, the Aspheric life.
11:33
So Jason, like you said, fall is in full swing.
11:37
Jason Oberholtzer: Absolutely.
11:37
Tori Dominguez-Peak: I have I have purchased pumpkin spice
11:39
lattes. The leaves are crunching. Sure. Hans has been
11:43
cooking beans.
11:44
Jason Oberholtzer: Hans has been cooking beans. I'm up to five or
11:46
six layers every time I leave the house.
11:48
Tori Dominguez-Peak: And with fall happening comes a new
11:51
semester on college campuses everywhere across The US. And so
11:56
kind of with that comes the renewed conversation that people
12:00
have been having about AI in education, and is it cheating,
12:05
and like all of the things. Sure. So earlier this year, the
12:10
New Yorker ran this piece with the title, everyone is cheating
12:14
through college. And the whole crux of it was just talking
12:18
about like how commonplace it is for students to use ChatGPT or
12:22
to use it to like help with assignments or even going as a
12:26
part of like, write my term paper for me.
12:28
Jason Oberholtzer: Mhmm.
12:29
Tori Dominguez-Peak: Blurring the line between getting it to
12:31
help you and then like, what has become plagiarism. And like, the
12:35
most stunning part about that article to me that I still think
12:39
about is they did a very small survey. It was like a thousand
12:42
college students. But 90% of them had said they had used
12:46
ChatGPT to help with homework assignments.
12:49
Jason Oberholtzer: And you found this surprising?
12:51
Tori Dominguez-Peak: Yeah. Mean, just the the number was wild. I
12:54
knew it would be over 50. When I saw 90, I was like, oh, we're
12:57
cooked. Okay.
13:00
But I just couldn't stop thinking about like, what
13:04
happens when we are letting a piece of technology kind of do
13:08
the thinking for us or do the talking for us
13:11
Jason Oberholtzer: Sure.
13:12
Tori Dominguez-Peak: On that scale. And so I decided to talk
13:16
to someone.
13:17
Megan Fritts: When I'm working on a new paper, I'll be, you
13:20
know, typing up some section and realize I don't know how to
13:23
phrase it. And that tells me, oh, okay, I need to go figure
13:27
out what I actually think here. Because if I can't write about
13:30
it, that indicates a lack of understanding there. So I think
13:33
what we're missing when we stop writing ourselves is the ability
13:36
to check ourselves for misunderstandings.
13:39
Tori Dominguez-Peak: So that's a professor I interviewed. She's
13:41
professor Megan Fritz. She teaches philosophy at University
13:45
of Arkansas at Little Rock.
13:47
Jason Oberholtzer: I find it really interesting here that
13:48
she's using misunderstandings as a framing. As if like we are
13:52
interrogating our own brain when we set down to write. Yeah. That
13:56
feels pretty right to me. Tori, can I ask you a question
14:00
quickly?
14:01
Tori Dominguez-Peak: Sure.
14:02
Jason Oberholtzer: Did you cheat in college?
14:05
Tori Dominguez-Peak: No. I cheated in high school though.
14:06
Jason Oberholtzer: Did you cheat in ways that you think
14:09
fundamentally changed your understanding or inhibited your
14:13
understanding of what you were doing? Or did they just help you
14:15
get a better grade?
14:19
Tori Dominguez-Peak: Okay. Let me just lay out the one scenario
14:20
that I cheated and I could help you could help me here.
14:23
Jason Oberholtzer: Mhmm.
14:23
Tori Dominguez-Peak: So I was failing chemistry. And so one
14:27
thing I noticed was that kids who took a long time to take the
14:29
test, it was like third period. And then you had fourth period
14:34
and then there was lunch. And so I was like, oh, if I just take
14:37
forever on this test, I can finish it later and study for it
14:41
during fourth period.
14:43
Jason Oberholtzer: Yeah.
14:44
Tori Dominguez-Peak: And so I was just like, oh, it's just
14:46
taking me forever. I have a headache. Like, I guess I have
14:49
to come back for lunch. And like, come back during lunch and
14:51
finish up this test. Bell rings, I go to fourth period.
14:56
It's like my study hall period. I'm like study I'm like going
14:59
over the stoichiometry formulas. I'm hitting the books. And then
15:04
I lunch happens. I go back to the chem room.
15:07
I actually reanswered some stuff because I was like, I have it in
15:10
my brain fresh now.
15:11
Jason Oberholtzer: Beautiful.
15:12
Tori Dominguez-Peak: And I took it. Is that cheating? It kind of
15:14
is.
15:15
Jason Oberholtzer: I well, yeah. I mean, by the letter of the
15:16
law, but it's gamesmanship is what I think it is. Like, you
15:19
still
15:20
Tori Dominguez-Peak: the player.
15:20
Jason Oberholtzer: Like Yeah. You still walked in there with
15:23
the requisite knowledge or the understanding of how to find the
15:26
knowledge, and you applied your brain to the problems at hand
15:29
and Yeah. Got a better grade to them. Under the framework that
15:33
professor Fritz is setting out here, that seems to be a
15:37
different kind of malfeasance in the classroom. And one that I
15:41
perhaps look more fondly on.
15:43
I cheated constantly.
15:44
Tori Dominguez-Peak: Okay.
15:44
Jason Oberholtzer: Either buy more time because I had not
15:47
prepared myself sufficiently on time, or route myself around
15:52
rote memorization which I considered to be an impediment
15:55
to learning and not a benchmark by which you measured learning.
15:58
And honestly, resented having to regurgitate things that one
16:01
could find in a book onto a page later. So I count neither of
16:04
those things as cheating. But like the thinking that I had to
16:07
do with that information still happened in my head and hit the
16:09
page.
16:10
Tori Dominguez-Peak: That's the thing. Like, I was still
16:13
studying.
16:14
Jason Oberholtzer: Yeah.
16:14
Tori Dominguez-Peak: I just convinced my chem teacher that I
16:16
had a headache when I didn't. Perfect. Right? Yeah. But I was
16:20
not plugging formulas in the chattypety and being like, what
16:23
are the answers to these questions?
16:26
Which I think is kind of different.
16:29
Jason Oberholtzer: Sure.
16:30
Tori Dominguez-Peak: And professor Fritz wrote this
16:31
article for the Chronicle of Higher Education, and I will
16:35
read you out loud an excerpt because I think it says
16:38
something like really poignant. It says, we're not simply
16:41
frustrated by just trying to police AI use or the labor of
16:44
having to write up students for academic dishonesty or the way
16:48
that reading student work has become a rather nihilistic task.
16:51
Our frustration is not merely that we don't care what AI has
16:55
to say and therefore get bored grading papers. It is that we
16:59
actively miss reading the thoughts of our human students.
17:05
Jason Oberholtzer: That is so dispiriting. Wow. Famously easy
17:10
job gets easier.
17:11
Tori Dominguez-Peak: It's kind of a bummer. And she is like
17:15
hitting something here. Like, when you write down something on
17:18
a page, you are transmuting your thoughts onto a page. And when
17:21
you turn it in, your instructor is reading your thoughts. Like,
17:24
we know there is a relationship between writing and thinking.
17:28
Jason Oberholtzer: Right. Exactly. And I like that she's
17:31
extending it to like a relationship between people on
17:33
either side of that activity. You will have a relationship
17:35
with your thoughts to the writing and the people reading
17:37
your writing have a relationship to those thoughts and therefore
17:40
you. And I think that is probably one of the great joys
17:43
of teaching is to be in relationship with those people
17:46
via their thoughts.
17:48
Tori Dominguez-Peak: Yeah. And I I kind of got to this sort of
17:51
thing like writing and thinking and like how they're related.
17:54
And I asked her, hey, is writing basically the same thing as
17:57
thinking or are they kind of intertwined in some way? And she
18:02
was like, yeah, they definitely are related. And that we write
18:05
it to document our thoughts, but we also write to come up with
18:08
thoughts.
18:09
It's kind of this really unique like both and relationship.
18:13
Jason Oberholtzer: Yeah. You definitely you write to find
18:15
when you have to stop writing because you don't know what's
18:18
there, and then you have to go think about it.
18:21
Tori Dominguez-Peak: Yeah. I mean, whenever I write a script
18:23
for Neverpost, I will write like a paragraph and then walk away,
18:27
and then come back like two days later. And then, you know, it's
18:29
just it's a slow process. But it's because I'm in relation
18:33
with my brain and trying to figure it out. And then I'm also
18:38
doing something that is also very thinking heavy, which is
18:42
I'm learning a second language.
18:44
I spoke more Spanish as a kid growing up in a Latin American
18:49
household. And then I lost it as it became a teenager. And then
18:53
I'm trying to get back into it as an adult. Mhmm. And so like,
18:56
when I speak Spanish in my adult learner's Spanish class, I'm
19:00
like having a thought in English and then translating it in my
19:04
head and then saying it out loud to them in Spanish, right, to my
19:07
instructor.
19:08
Sure. And then she says something back to me in Spanish,
19:11
and then I'm trying to translate it into English in my head. And
19:14
it's just this very like mechanical relationship, and
19:18
it's not easy. And I feel a little bit like a baby alerting
19:21
to speak for the first time.
19:23
Jason Oberholtzer: Yeah. Okay. So if I'm if I'm hearing this
19:26
right then, it's like you're sort of seeing the the the gap
19:30
between the thought that you're having and your ability to
19:33
articulate it in this seconds now. I guess, re seconds, third,
19:38
second, again, language. Probably especially because at
19:42
one point, it was not there.
19:43
And you're feeling this, like, this break in the chain between
19:46
you having a thought and being able to articulate that. Yeah.
19:49
And that feels sort of similar to what you think is happening
19:53
with the insertion of these AI tools into the way people are
19:58
writing these days.
19:59
Tori Dominguez-Peak: Yeah. And professor Fritz kind of brought
20:01
this up that AI is kind of this middleman between thought and
20:04
language that's never been there before.
20:06
Megan Fritts: The difference between being a native speaker
20:08
of a language and being someone who's learned a language, I
20:11
think is is the perfect example of what we are risking, when we
20:15
use generative AI for our writing and, speaking, that we
20:21
risk going from this kind of native speaker status to a a a
20:24
situation where we have to if we want to have these skills at
20:27
all, we have to reteach it to ourselves in a in a really
20:30
artificial way.
20:31
Jason Oberholtzer: So is the concern there almost like what
20:34
happens in its absence? Or like it's like what happens if you're
20:37
without your Spanish English dictionary as it were Yeah. That
20:41
you'd actually don't have control over the language.
20:43
Tori Dominguez-Peak: Yeah. It it feels like if we outsource for
20:47
example, you didn't do the reading for your college class,
20:52
but you have to write paper about it and you're just like,
20:53
shitty, this paper about this thing. You didn't engage with
20:57
the text. Yeah. You didn't transmit the text into writing.
21:01
And so like, you're probably not even remember what that class
21:03
was about. Yeah. You're losing some type of critical like brain
21:07
step that helps you metabolize information. Does that make
21:11
sense? Like, I think writing helps you metabolize
21:14
information.
21:15
And like, if I could just be candid, I think writing is kind
21:20
of mentally painful for me.
21:21
Jason Oberholtzer: Oh, yeah. That's the whole thing about
21:23
writing.
21:23
Tori Dominguez-Peak: Like It
21:24
Jason Oberholtzer: hurts and it's bad.
21:25
Tori Dominguez-Peak: It hurts and it's bad, and that's the
21:27
whole point. And I will write something and I'll walk away and
21:30
I come I will come back three days later, And I'm just like,
21:33
oh, this psychologically hurts. But then you get into a groove
21:35
and then it feels good and you've kind of metabolized your
21:39
thoughts and it comes out and it feels great. And there's just
21:42
such an emotional experience in that. And so when you're just
21:45
like, hey chat, GBT, write this podcast episode, summarize the
21:51
notes from this interview I have with the source, or write the
21:53
interview questions.
21:55
Jason Oberholtzer: Yeah.
21:55
Tori Dominguez-Peak: You're kind of losing some threads.
21:59
Jason Oberholtzer: Yeah. You know, not to be the person who's
22:01
constantly defending cheating here. But when you brought up
22:05
writing the report in the book you have not read, that is
22:08
something I believe a It's lot of us have a classic move. You
22:11
have to try it at some point. And when you do that, you
22:14
marshal enough information about the book, you skim it, you look
22:17
up some notes, you try to find as much as you can to walk in
22:20
there.
22:21
But the thing you're doing when you walk in there is like using
22:26
your brain. It is learning. It is performing. It is something
22:30
that requires you to undergo a process that will help you be a
22:33
better thinker and communicator in the future because you are
22:37
actually doing a task. And to me, what feels scary about this
22:42
is that it is removing the mental load of doing the task.
22:46
Mhmm. Not so much cheating like the information, which is like
22:49
the veneer around which we all do the process of learning, but
22:52
it's removing the actual mental task, which is the point of
22:54
sitting down and being a part of a university or a class or
22:58
whatever the case may be. So I feel like this has to feel
23:02
different for teachers. Like, they've walked into campuses
23:04
every fall for millennia and been like, alright, everyone's
23:07
cheating. How do I make sure that I know that they are smart
23:09
enough to continue down the road after they continue cheating?
23:12
Like, I I that's probably unlikely that people believe
23:15
they have a complete fail proof method to stop all cheating
23:19
forever. Do you think that professor Fritz or other
23:21
professors are feeling like this is a different kind of stop
23:25
cheating move they need to make?
23:28
Tori Dominguez-Peak: Yeah. So professor Fritz has kind of gone
23:34
nuclear. Okay. She instated a policy in her classroom that is
23:40
just like, I am banning all AI from my classroom. Okay.
23:43
It includes ChatGPT. It even includes like Grammarly, which I
23:47
use Grammarly to like make sure that my emails aren't misspelled
23:50
or whatever.
23:51
Mike Rugnetta: Oh, interesting. Not even
23:52
Tori Dominguez-Peak: too many exclamation points. She's like,
23:53
nope, not even that. Because Grammarly can suggest rewrites.
23:57
Jason Oberholtzer: Yeah.
23:57
Tori Dominguez-Peak: And that in itself is kind of generative AI.
23:59
Jason Oberholtzer: Yeah.
24:00
Tori Dominguez-Peak: And so I asked her like, okay, so how are
24:03
you enforcing this?
24:04
Jason Oberholtzer: Sure.
24:04
Tori Dominguez-Peak: It just seems like a lot of work to
24:06
enforce this. And she has a very interesting way of going about
24:11
this. So at some point earlier in the semester, she has them
24:16
write these short essay type assignments in class. They are
24:19
handwritten, and they turn it into her right there, hard copy.
24:24
And so it's zero chance of AI use.
24:27
She had you write with a pen and paper. Yeah. And she keeps these
24:30
essays as kind of evidence of like, this is how these people
24:33
write. Oh. This is what your voice, your narrative voice is
24:37
like.
24:38
And so then later on in the semester, you turn in something
24:41
electronically and it's got AI written stuff, she's gonna be
24:45
like, this doesn't sound like you.
24:47
Jason Oberholtzer: Okay. Does she do the comparison process
24:51
herself or does she let AI do the comparisons?
24:54
Tori Dominguez-Peak: So she kinda does both. Interesting.
24:57
She does use eight different AI detection programs that she runs
25:00
it through, which is Wow. It's a lot.
25:02
Jason Oberholtzer: Does the school pay for these?
25:04
Tori Dominguez-Peak: I don't know. That's great question.
25:06
Don't seem cheap, do they?
25:08
Jason Oberholtzer: No. I would imagine not.
25:11
Tori Dominguez-Peak: And so that she runs it through all of
25:12
those. She also just looks at it herself and is like, yeah, I
25:18
could tell.
25:18
Jason Oberholtzer: Yeah. Of course. Right? Like, teachers
25:20
have been seeing this, like, forever.
25:22
Tori Dominguez-Peak: But I did ask, like, can you just tell off
25:25
the bat of a student's writing as AI generated just by looking
25:28
at their paper? And she said, yes. And that there's usually a
25:34
couple of clues.
25:35
Jason Oberholtzer: Okay. Is this where you're coming from my Em
25:37
dashes? Yeah. Alright. I'll listen to it at least.
25:41
Tori Dominguez-Peak: The first clue is she calls them 50¢
25:44
words. Weird, like formal words that the typical 18 year old
25:49
undergrad would just not be using. Okay.
25:52
Megan Fritts: An example of that is like the word want, w o n t,
25:56
where so you might use it in a sentence like, I want to take a
26:01
walk in the morning. So it's like talking like an
26:03
inclination. That's a 50¢ word that I I I would say most of my
26:09
students probably aren't just casually using in their
26:12
reflection writings.
26:14
Jason Oberholtzer: Now you're coming from my vocabulary.
26:16
First, my em dashes.
26:17
Tori Dominguez-Peak: Are you want to use wants?
26:19
Jason Oberholtzer: I mean, of course, I am. But like, where in
26:22
an academic setting would I ever run across the need to express
26:26
my feelings through the word want?
26:29
Tori Dominguez-Peak: Yeah. Exactly. Clue number two, and
26:31
you were pretty right about this, was Em dashes. Yeah. It's
26:34
not just any Em dash because I I I feel you, like, it sucks that
26:38
em dashes have become like this weird red flag and it's like, I
26:41
like a good em dash.
26:42
Jason Oberholtzer: Sure.
26:43
Tori Dominguez-Peak: She said that there's a type of em dash
26:46
that is a red flag to her.
26:48
Megan Fritts: An acquaintance of mine, another philosophy
26:50
professor, he calls those epiphany dashes, where you go
26:54
from an ordinary, you know, thought like it's not just a
26:58
walk em dash, it's a brainstorming session. This is,
27:02
you know, you're having an epiphany.
27:04
Jason Oberholtzer: Interesting. Yeah. So there's like an emotive
27:06
component to the Emdash when it's used this way.
27:09
Tori Dominguez-Peak: Some some sense it sounds like a LinkedIn
27:11
post. You know what I mean? Like, has that type type of
27:14
cadence for it.
27:15
Jason Oberholtzer: Right. It's like It's sort of a it's like
27:16
copywriting usage not Yeah. Yeah. Yeah. Oh, that's super
27:21
interesting.
27:21
Tori Dominguez-Peak: And then obviously, the third main clue
27:23
is like she runs it through one of her software programs and
27:26
it's like, bing, AI generated. But yeah, mean, it's kind of an
27:30
intensive process.
27:32
Jason Oberholtzer: Yeah. It sounds exhausting. I guess for
27:34
everyone, I suppose. Like, are the students having a good time
27:36
while this is happening?
27:38
Tori Dominguez-Peak: They don't seem to be big fans.
27:40
Jason Oberholtzer: Sure.
27:41
Tori Dominguez-Peak: As you can imagine, a lot of them have
27:43
reacted by saying, well, I don't get why you've banned it because
27:47
my other professors and other classes don't care. So like, why
27:50
should it matter?
27:51
Jason Oberholtzer: Oh, jeez. Alright. Well, at least they're
27:53
still learning something and that something is emotional
27:55
manipulation.
27:57
Tori Dominguez-Peak: Yeah. I guess they gotta learn to read
27:58
the syllabus.
27:59
Jason Oberholtzer: Yeah.
28:00
Tori Dominguez-Peak: But it does kind of bring up something which
28:04
is that, okay, some professors don't care. Professor Fritz very
28:09
much does. Yeah. And so college students are kind of navigating
28:12
this landscape where it's like in the same semester, they might
28:15
have someone who's a real stickler about this stuff. And
28:18
they may also have a different teacher who doesn't care.
28:21
And it seems like they have to navigate all these individual AI
28:26
policies.
28:27
Jason Oberholtzer: Or just not use AI? That's one way to
28:30
navigate all of them.
28:31
Tori Dominguez-Peak: Or just not use it at all. But I asked Fritz
28:35
about this like, oh, do you talk with other professors about
28:38
handling this? And she said that she actually sits on a couple of
28:41
AI related committees at her university.
28:44
Megan Fritts: I think instructor uniformity and solidarity on
28:48
this issue is pretty important for our students. I thought it
28:52
was a good idea, I was excited to to to try to make this
28:56
policy. But what ended up happening is that people just
28:59
had such different views on AI use in higher education that it
29:06
kind of just turned into debate every every meeting, and we we
29:15
have not yet made any kind of a policy.
29:18
Jason Oberholtzer: Okay. Well, I suppose that's predictable. It's
29:20
academia. It's meetings. It's consensus.
29:24
That is not necessarily easy to do, but, like, you know, I'm not
29:27
gonna tip my hand on where I stand on this. I'm sure everyone
29:30
is in deep mystery here. But like, if it's working for
29:33
professor Fritz, like, just let her do her thing. It seems like
29:37
it works.
29:38
Tori Dominguez-Peak: Yeah. I mean, she thinks that she feels
29:40
that her system works well for her. Yeah. But her way of doing
29:45
things relies on knowing what these students write like, and
29:49
knowing what they don't write like, and they're writing as a
29:52
sort of fingerprint. But what happens when you can't really
29:56
tell and you have to do some detective work?
29:59
Jason Oberholtzer: Mhmm.
30:00
Tori Dominguez-Peak: So I found a detective.
30:02
Jason Oberholtzer: Alright, listeners. This is the most
30:04
podcast break we will ever do after these messages, a
30:09
detective.
30:26
Rui Sousa-Silva: Even though we learn the same languages from
30:29
the same books and we learn we can find the same words in the
30:33
same dictionaries, the way each one of us uses language is
30:37
different. So we have a let's call it a different style of
30:42
using language.
30:45
Tori Dominguez-Peak: So this is doctor Rui Sosasilva. He's from
30:49
Portugal. His whole training and job is as a forensic linguist.
30:54
Oh. So like, literally, his field is all about confirming
30:58
the identity of who wrote what.
31:00
Jason Oberholtzer: Woah. That's a rad job.
31:03
Tori Dominguez-Peak: I know. It's such a cool job. And I
31:04
literally didn't even know this job existed until I started
31:07
writing this episode. And I was like, this is amazing. Wow.
31:11
And being able to identify the nuances of what someone sounds
31:17
like, that identity is called an idiolect.
31:21
Rui Sousa-Silva: So ideolect is your own way of speaking or
31:27
writing the language. So it's as if your DNA was related to the
31:31
way you use language.
31:33
Jason Oberholtzer: I believe it.
31:34
Tori Dominguez-Peak: So I mean, like, legally, criminally,
31:38
historically, a forensic linguist is the person you call
31:42
to match the fingerprints of someone's writing. Right?
31:46
Jason Oberholtzer: Woah.
31:47
Tori Dominguez-Peak: And so at the same time, he's also a
31:49
lecturer. He also teaches, which means that he also has to deal
31:53
with the issue of students using AI in his class. Really also has
31:59
students write in his class sit down and write. I know you're
32:03
not using AI because I can look at you writing. Right?
32:06
But when they do that, he's noticing something different
32:10
than Professor Fritz.
32:12
Rui Sousa-Silva: What we see nowadays, people interact with
32:14
generative AI so much that people are starting to write
32:20
like machines. With some of my students, I know that they are
32:25
sitting an exam and I know they were the ones who wrote the text
32:29
and still when I read the text it sounds as if it was generated
32:33
by a machine. And that's because we accommodate with other people
32:38
and we accommodate in the same way with the machines we
32:41
interact with. So we tend to accommodate so much to the
32:45
machine that we learn so much from the machine that we start
32:48
start writing like machines. So this is a challenge at the
32:51
moment.
32:52
Jason Oberholtzer: So is he saying that because we're
32:54
ingesting so much writing that has been made in this process
32:58
that we are starting to regurgitate Yeah. That Yeah.
33:02
That checks out.
33:04
Tori Dominguez-Peak: That's wild though, isn't it? Yeah. Because
33:06
then it makes me think about professor Fritz's class and her
33:09
methods, like, what if people just start writing like
33:13
ChatchyPT?
33:14
Jason Oberholtzer: Oh, boy.
33:15
Tori Dominguez-Peak: Then at some point, it's just gonna get
33:17
harder to tell.
33:19
Jason Oberholtzer: Alright. You've made an un virtuous cycle
33:21
here. I see what has happened.
33:24
Tori Dominguez-Peak: You see what has happened. Yeah. And so
33:26
because Rui is a forensic linguist, like he is the person
33:29
you can tell who wrote what, I was like, can you please give me
33:33
an example of influencing the way of like how machines
33:37
influence the way people write. And he said that when you speak
33:41
to ChatGPT in Portuguese, because he lives in Portugal, it
33:45
will sometimes reply using a Brazilian Portuguese dialect,
33:51
and that has consequences.
33:53
Rui Sousa-Silva: Yeah. One example is the way when when
33:56
you're writing in English, you usually say, if you want to to
33:59
list a set of points, you'll say, firstly, such and such.
34:03
Secondly, such and such. And in Portuguese, usually you wouldn't
34:09
use the literal pronunciation of the adverb. But people are now
34:14
doing that and that's because interestingly Brazilian
34:17
Portuguese does that and because when you look at language
34:22
variants, I mean in Portugal you've got about 10,000,000
34:27
speakers, if you go to Brazil there are 200,000,000, so for
34:31
generative AI engines they feed on languages.
34:35
So they they are more likely to feed on Brazilian Portuguese
34:39
Tori Dominguez-Peak: than There's just more Brazilian
34:40
Portuguese language data out And so when Chateapiti speaks
34:44
Portuguese, it sounds Brazilian.
34:46
Rui Sousa-Silva: It's usually Brazilian Portuguese, even
34:48
though nowadays you can ask to write in European Portuguese but
34:51
every now and then there is a word in Brazilian Portuguese for
34:54
example. So the fact that it was based on Brazilian Portuguese
34:59
and the fact that Brazilian Portuguese uses that literal
35:02
translation of firstly, secondly, thirdly, now people
35:06
are writing like that.
35:08
Tori Dominguez-Peak: Oh, that's so interesting.
35:10
Rui Sousa-Silva: Even native speakers of European Portuguese
35:13
are writing like that at the moment.
35:14
Jason Oberholtzer: Wow. As I'm hearing that, I'm just thinking
35:17
that that is not necessarily the canary in the coal mine, but
35:20
there's probably some better metaphor for it just being the
35:23
visible object of something that is also happening under the
35:26
layer of language and thinking. If we are regurgitating
35:30
different cultural grammar rules, we're probably also
35:33
surfacing other imports that we don't know are imported from
35:37
different places.
35:39
Tori Dominguez-Peak: Yeah. I mean, Brazilian adverbs are
35:41
pretty low stakes. I mean, I think it's kind of amusing. But
35:45
remember when I said that forensic linguists identify the
35:49
DNA of language.
35:50
Jason Oberholtzer: Yeah.
35:50
Tori Dominguez-Peak: And they do that in criminal settings. Can I
35:54
outline a scary scenario for you, Jason?
35:59
Jason Oberholtzer: I'm braced. I'm ready.
36:00
Tori Dominguez-Peak: Okay. So doctor Sosasilva told me that
36:04
one of the things that his colleagues talk about all the
36:07
time is the use of AI in criminal activity. So like using
36:12
generative AI to impersonate someone's writing style to write
36:16
something incriminating.
36:18
Jason Oberholtzer: Sure.
36:18
Tori Dominguez-Peak: So for example, somebody does not like
36:21
you. And so they write a strongly worded threat to a
36:26
politician, but they write it Jason Operholzer style. And they
36:30
like, maybe they make a sock puppet social media account and
36:33
they impersonate you, and they post it on there. And you get a
36:36
visit from the police, and they're like, you wrote this.
36:39
Jason Oberholtzer: Yeah.
36:40
Tori Dominguez-Peak: How do you prove that you didn't?
36:42
Jason Oberholtzer: That's a good question. I would probably at
36:45
this point, try to point to like the corpus of available writing
36:52
that I would have on hand. Like, I'm trying to really take a
36:54
cynical view of this and just assume that the the
36:57
infrastructure is weighted against me on this one, and
36:59
there's now a letter out there signed by me that says, hey,
37:02
buddy, I'm gonna kill you. And to disprove this, I don't think
37:09
I would have successful time attacking the language on a word
37:11
by word basis. I would probably have to compile my own corpus of
37:15
writing, and I'd probably have to divulge repositories of data
37:20
that I would otherwise want to keep secret, like private
37:23
messages and be like, I will take all of my eye messages and
37:28
put them in a model.
37:29
And you can see the way that I communicate and you can see like
37:32
all the communication I have had, all the cynicism I have had
37:35
around politicians and the government. And you tell me
37:39
where in this trajectory is there the leap to a murderer.
37:43
And it is less about word choice and more about state of mental
37:47
well-being. Like, is this the and then I'm going to write a
37:50
politician and murder them trajectory? And here's every
37:52
piece of written correspondence I have available to you.
37:55
And just hope that I've got a doctor like the good Portuguese
37:59
doctor doctor Sosa Silva on my side who can help me make a
38:02
better argument about that material than whoever's on the
38:05
other side.
38:06
Tori Dominguez-Peak: Yeah. I mean, definitely, you would like
38:08
to call doctor Sosa Silva. Right?
38:10
Jason Oberholtzer: Yeah.
38:11
Tori Dominguez-Peak: They can analyze the text. They're like,
38:12
this is this isn't quite reek of Jason. There's something just a
38:17
little bit off of this.
38:18
Jason Oberholtzer: Speaking of word choice, can we do better
38:19
than reek of Jason?
38:20
Tori Dominguez-Peak: Reek of Jason.
38:23
Jason Oberholtzer: For want of a better term, I suppose you can
38:25
keep reek.
38:26
Tori Dominguez-Peak: So, yeah, this is where forensic linguist
38:28
come in and then you don't get arrested, hopefully. Right?
38:32
Okay. But as generative AI gets more sophisticated, Rui thinks
38:37
that his work is going to get harder.
38:40
Rui Sousa-Silva: The developments in generative AI
38:42
will make it more complicated for forensic linguists to
38:47
attribute texts, which in turn will mean that forensic
38:52
linguists will need to do more research and to further their
38:55
research and to have more fine grained methods of attributing
38:59
authorship. But there will always be a distinction between
39:04
the way humans produce text and the way machines generate texts.
39:08
So things generative AI will evolve, forensic linguistics
39:13
will evolve, but eventually we will always be able to pinpoint
39:19
differences between the texts.
39:21
Jason Oberholtzer: So very similarly to the classroom here,
39:25
this seems like it is just creating piles of work for
39:27
everybody.
39:28
Tori Dominguez-Peak: Yeah. It seems like things are just gonna
39:30
get harder for everyone, which is kind of a bummer. And I know,
39:34
like, the threatening the politician, like, that's a very
39:36
dramatic example, obviously. But like you said, with the
39:40
classroom, the idea I keep coming back to is cognitive
39:45
offloading and like, not doing all these mental processes and
39:50
offloading it to AI, which then I guess makes it harder for
39:53
forensically. Like, there's been studies from Microsoft, from the
39:58
SBS, Swiss Business School about how people who use generative AI
40:04
regularly tend to score lower on markers of critical thinking.
40:08
Like, there's actual data we have now. Mhmm. Again, I'm
40:12
painting kind of a scary picture to you. It's like, okay, so our
40:16
critical thinking may be getting compromised by a technology that
40:20
is also getting more sophisticated at pretending to
40:24
be us. And we're also starting to become influenced by the way
40:29
it writes.
40:30
Mhmm. Like, that's just such a weird trifecta.
40:34
Jason Oberholtzer: Right. But doesn't the cycle also work in
40:35
the other direction? Like, we are the corpus of information
40:39
that the generative models need to continue their work. And as
40:43
we lose cognitive function because of offloading, the
40:46
material that we are able to feed depreciates in value as
40:50
well, which one imagines leads to worse outputs from machines,
40:55
which we are synthesizing and further inhibits our ability to
40:58
think and provide a reasonable corpus of information updated to
41:01
the moment from which Yeah. The models can select.
41:04
Tori Dominguez-Peak: It's kind of like this feedback loop.
41:06
Yeah. Us feeding it and then it influencing us, and then all of
41:11
a sudden, we're just all speaking Brazilian Portuguese,
41:14
right?
41:15
Mike Rugnetta: Yeah.
41:16
Tori Dominguez-Peak: And getting accused of crimes we didn't
41:18
commit. And professor Fritz's concern about the feedback loop
41:23
is that it affects everyone differently.
41:26
Megan Fritts: A lot of people defend, letting students use AI
41:31
in their work, by saying that they see it as a tool for
41:35
equity. Maybe for students who, had a less privileged primary
41:41
education or students for whom English is a second language, I
41:44
would contend the exact opposite is true. That what this is doing
41:48
is setting the stage for genuine reading and writing skills
41:51
becoming something that is really only accessible to the
41:56
elite class, those with a lot of money and leisure time to
42:00
cultivate them intentionally. And so that's really something
42:03
that concerns me quite a bit.
42:05
Jason Oberholtzer: Okay. I'm beginning to see why there's
42:06
such difficulty forming consensus around this. I mean,
42:09
initially, I'll be honest, my reaction was yeah. It's obvious
42:13
that in a space where you're supposed to be practicing
42:15
thinking and metastasizing your own thoughts that AI is just not
42:20
helpful. It's not there for any reason except for you to finish
42:23
the paper, which is a representation of the thoughts
42:26
you were supposed to be having, and it is like the wrong
42:28
takeaway from what you're doing in the classroom.
42:31
But now that they're applying these frameworks around
42:34
accessibility, I can see it becoming a little more
42:38
complicated. I guess a lot of it boils down to how much you think
42:41
the role of the academy is to prepare you for work,
42:45
employment, and how much it is to help you engage with how you
42:49
think and learn. But I'm beginning to see why the stakes
42:55
are a little more complicated than perhaps they feel at first
42:58
blush.
42:59
Tori Dominguez-Peak: Yeah. And then when you think about
43:01
professor Fritz's role in the committees and how it's just
43:05
really hard to come to agreement or make any sort of policy, I
43:10
think this is something that they're gonna continue to
43:12
wrestle with for a long time.
43:26
Megan Fritts: As for myself, my policy won't be changing. And
43:30
that's, you know, that's about all I can do about that.
43:44
Jason Oberholtzer: Tori, thank you as always for bringing in a
43:48
really insightful piece here.
43:51
Tori Dominguez-Peak: Yeah. Thanks so much for letting me
43:52
talk about it.
43:53
Jason Oberholtzer: Yeah.
43:54
Tori Dominguez-Peak: And I just wanted to shout out thanks to
43:56
professor Fritz and to doctor Sosa Silva.
44:00
Jason Oberholtzer: At least one of whom is probably going to be
44:02
getting me out of jail over the next couple years. So pre thank
44:05
you for that one.
44:06
Tori Dominguez-Peak: Pre thank you for that one.
44:09
Jason Oberholtzer: Tori, where can folks find you and all of
44:12
the writing that definitively comes from your own brain on the
44:15
web?
44:16
Tori Dominguez-Peak: You can find me at toori d p nine
44:18
8.bluesky.social. And you can also find my podcast about video
44:24
games that I make of my own brain and play with my own brain
44:30
at press-startpod.bluesky.social.
45:15
Jason Oberholtzer: Tuesday, September 23, 10:30AM.
47:35
Mike Rugnetta: That is the show we have for you this week. We're
47:37
gonna be back here in the main feed on Wednesday, October 8. We
47:40
are proud and thankful and extremely lucky to have the
47:45
member community that we do. Without the support of our
47:50
members, this show would not and could not exist. So I just wanna
47:56
say thank you.
47:58
If you would like to become a NeverPost member and join this
48:02
community for as little as $4 a month, you can do that at
48:06
neverpo.st. Where also if a membership is a little too big
48:11
of a commitment in these strange and trying times, you can also
48:15
tip us a one time any dollar amount, and we promise that we
48:20
will spend every last red cent on arcade games chewing gum and
48:26
baseball cards. Become a member at neverpo.st. Never Post's
48:39
producers are Audrey Evans, Georgia Hampton, and the
48:41
mysterious, doctor first name, last name. Our senior producer
48:44
is Hans Buto.
48:45
Our executive producer is Jason Oberholzer, and the show's host,
48:48
that's me, is Mike Rugnetta. And then this warning flashes on the
48:55
light meter. Inside the house, a pilot light is always burning in
48:59
the oven's eyes. The low roof is pulled down over the eyes like a
49:04
hat. And underneath the warnings, light motif networks
49:07
of subterranean lines run like the nervous system or bloodlines
49:12
or fractures spreading from tectonic lines of fault.
49:16
In distant coasts, heavy and light petroleum is piped across
49:20
state lines and gas, electric, oil, and water lines convey
49:25
their vital humors to the house. Excerpt of nervous systems by
49:31
Greg Williamson. Never Post is a production of charts and leisure
49:35
and is distributed by Radiotopia.