ASU professor talks use of AI, ChatGPT in universities and its future
Summary
TLDRThe video discusses the impact of AI on university students, especially with tools like ChatGPT. The professor explains how AI has evolved and how it is used in academia. While AI helps students with tasks like improving writing, it also raises concerns about plagiarism and misinformation. The professor highlights the ongoing cat-and-mouse game between educators and students adapting to AI's capabilities. They mention that despite challenges, there's potential for positive use in education. The discussion touches on the need for institutions to develop strategies to handle AI's influence in academic settings.
Takeaways
- ๐ก AI is fundamentally reshaping how students approach research, learning, and exams, offering both new opportunities and challenges.
- ๐ฅ๏ธ The professor explains that AI's recent surge in impact is due to the rise of internet data, smartphones, and machine learning algorithms, making tasks like image and text generation more accessible.
- ๐ Students can now use AI tools like ChatGPT to assist with writing and research, but this raises concerns about plagiarism and academic integrity.
- ๐ AI systems like ChatGPT don't retrieve exact passages but rather generate plausible completions of prompts, often introducing factual inaccuracies known as 'hallucinations.'
- ๐ Academia is grappling with how to respond to AI's influence, with some universities considering banning these tools while others explore how to integrate them responsibly into education.
- ๐ The rise of AI writing tools has blurred the lines between authentic student work and machine-generated content, leading professors to rethink plagiarism detection and writing assessments.
- ๐ The professor highlights AI's widespread influence, noting that ASU students are actively using these tools, and its impact is visible in Silicon Valley and beyond.
- โ ๏ธ Large language models (LLMs) can sometimes produce misinformation, fabricating sources or generating fictional stories, creating new challenges for verifying facts in both academic and legal contexts.
- ๐ง A watermarking system could help identify AI-generated content by manipulating word choices in generated text, but its adoption depends on cooperation from AI developers.
- ๐ฎ Despite concerns, the professor believes society will adapt to AI's presence, drawing parallels to how people learned to spot AI-generated essays in academic contexts over time.
Q & A
What is the primary focus of the professor's department?
-The department focuses on computer science and artificial intelligence, particularly on technologies that assist humans in performing intelligent tasks. It also covers general computer science topics like hardware, software, and software engineering.
How does the school collaborate with the outside world?
-The school collaborates with various companies and government agencies, often through sponsored research. The professor's research, for example, is funded by federal agencies like the Department of Defense (DOD) and the National Science Foundation (NSF). Students also contribute by interning at industry partners, particularly in Silicon Valley.
What significant changes have occurred in the field of AI over the past few decades?
-AI has evolved from performing tasks that humans understand, like playing chess, to tackling complex tasks like image and text recognition. The recent availability of large datasets from the internet has allowed AI to learn tacit knowledge that humans find difficult to articulate, like recognizing images or generating text.
How is AI impacting students' ability to write papers and take tests?
-AI tools like ChatGPT can assist students by improving their writing, especially for those who are non-native English speakers. However, there is also concern about plagiarism, as AI-generated essays can be difficult to detect since they paraphrase rather than directly copying text.
What challenges do large language models (LLMs) like ChatGPT pose in academic settings?
-LLMs present challenges in detecting plagiarism since they generate original-looking content by paraphrasing rather than copying. They also have issues with 'hallucinations,' where they fabricate information, making it harder for educators to trust the output.
How can AI tools improve student writing, and what concerns exist?
-AI tools can help students refine their rough ideas, making their writing clearer and more coherent. However, there is concern that students may rely on AI to do their work for them, especially in areas like plagiarism and over-reliance on AI-generated content.
What proactive steps are universities taking to address the challenges posed by AI?
-Some universities are restricting access to AI tools, while others are exploring ways to use AI proactively, such as asking students to fact-check or correct AI-generated content. There is also ongoing research into watermarking AI-generated text to identify its source.
How might future tools help detect AI-generated content in academic work?
-One idea is to watermark AI-generated content by using a technique that prevents certain words from being used. By knowing which words were blacklisted during generation, professors could potentially detect AI-generated work, as human writing wouldnโt follow the same constraints.
How are students currently using AI tools like ChatGPT?
-Students are using AI to paraphrase texts, generate summaries, and assist with research. However, there are cases where AI tools fabricate citations or other information, which can lead to academic issues.
What are some of the limitations of AI tools like ChatGPT in generating accurate information?
-ChatGPT and similar tools often generate the 'most likely' completion for a given prompt, which can result in the creation of false information, such as fabricated citations or inaccurate descriptions. This makes it unreliable for fact-based tasks.
Outlines
๐ง Introduction to AI in Education
The speaker begins by introducing the topic of AI's impact on university students, particularly as they approach finals. They interview a professor from the school of computing and AI, who explains the department's focus on technologies that aid humans in intelligent tasks. The professor also discusses the department's research and its collaboration with federal agencies and industry partners. The conversation highlights the rapid advancements in AI and how they have been integrated into the curriculum and student life.
๐ AI's Impact on Student Research and Plagiarism
The discussion shifts to how AI has made research easier for students but also facilitated plagiarism. The professor explains that tools like Turnitin are used to detect copied work, but AI technologies like chatbots can generate original text that is difficult to trace. This has led to a new challenge in academia, as educators must now discern whether student work is original or AI-assisted. The professor also notes the positive use of AI for improving writing skills, especially for non-native English speakers.
๐ The Evolution of AI in Academia
The professor reflects on the evolution of AI, noting its origins in the 1950s and the significant advancements in the last 10-15 years due to the availability of large datasets on the internet. He compares AI learning to human learning, where systems now can generate text and complete tasks in ways that were previously challenging. The conversation also touches on how AI has changed the landscape of student assignments and exams, making it easier to generate text but also raising concerns about the authenticity of student work.
๐ Combating Plagiarism with AI
The discussion delves into the strategies professors are using to combat AI-assisted plagiarism. The professor suggests that students should verify the information generated by AI systems and correct any inaccuracies. He also mentions the idea of watermarking AI-generated text to identify its origin. The conversation highlights the need for students to critically evaluate AI-generated content and the ethical implications of using AI in academic work.
๐ฎ The Future of AI in Education
The professor speculates on the future of AI in education, suggesting that educators and students will adapt to the new challenges posed by AI. He shares anecdotes about how quickly professors have learned to identify AI-generated essays and predicts that the cat-and-mouse game between educators and students will continue to evolve. The conversation concludes with the professor's thoughts on the potential for AI to both enhance and complicate the educational experience.
Mindmap
Keywords
๐กArtificial Intelligence (AI)
๐กLarge Language Models (LLMs)
๐กPlagiarism
๐กChatGPT
๐กHallucination
๐กTurnitin
๐กParaphrasing
๐กWatermarking
๐กSilicon Valley
๐กExplicit vs. Tacit Knowledge
Highlights
AI's impact on university students as they approach finals.
Professor's name and title in the School of Computing and AI.
Department's focus on technologies that assist humans in intelligent tasks.
Research sponsored by federal agencies and industry partners.
Students' significant presence in Silicon Valley and their impact.
AI's evolution since 1956 and the shift from explicit to tacit knowledge.
The role of internet and mobile technology in AI advancement.
Challenges in detecting plagiarism with AI-generated text.
Use of AI to improve writing skills, especially for non-native English speakers.
Academic concerns over AI-generated content and plagiarism since the release of ChatGPT.
Strategies to identify AI-generated content, such as checking for factual accuracy.
The potential for AI to 'hallucinate' and create false information.
The idea of watermarking AI-generated text to detect its origin.
Students as early adopters of AI technology for various purposes.
The use of blacklisted words to identify AI-generated text.
The commercial applications of AI beyond academia.
The adaptability of educators and students to AI-generated content.
The future of AI in education and the potential for new detection methods.
Transcripts
well yeah so it's like it's become very
you okay okay well thanks for talking to
us if it's okay just an overview we we
kind of started this wondering
how much AI may or may not be changing
the landscape of students at
universities as we head to finals so
that's kind of how we started this thing
if it's okay I'll ask you a little bit
um first first give me your your name
first and last in the title please okay
um I am
and I'm a professor in the school of
computing and AI
um spell that Professor for me so I get
it right first and last name
yeah so what what happens in your
department here what kind of things do
you do this this is the Department of
computer science and artificial
intelligence augmented intelligence and
so basically we are interested in
looking at technologies that
um help humans in you know whatever
considered intelligent tasks in addition
to General computer science topics such
as you know Hardware software software
engineering and so on so that's the
sorts of things that we have that happen
and there's a bunch of work and research
going on and in all sorts of AI
artificial intelligence topics um in in
the department in the school
yeah be careful with your hands you're
bouncing it okay oh yeah okay okay how
does this school and what you do here in
end up translating to the outside world
you team up with companies or government
agencies yes so lots of that so most of
the research is sponsored by you know
the federal agencies in my own case you
know multiple federal agencies DOD
reports NSF and
and also lots of us get
gifts or the grants from more like
contracts from local industry or you
know National Industry partners and so
that's one main way of course you know
as a educational institution our biggest
way we actually help the industry and
the world in general is our students so
our students go on internships um so in
fact it's you can't go to Silicon Valley
and throw a stone without hitting any
you know Sunday well essentially I mean
so computer science ASU computer science
students are all over Silicon Valley so
so that's those are like the main
impacts that we have
tell me about
um
how fast and how far we've come with
something like AI because to us out in
the outside world it seems like
overnight boom it was there
um from your point of view so yeah I
think as a field AI has been around
since 1956 actually I think that's you
know there was like this meeting when
they acquired the name and I personally
have been working since my undergrad day
so close to 40 years
um the main thing that changed is in the
beginning we were getting computers to
do what are intelligent tasks that we do
but we know how we do them things like
playing chess we not only play chess we
have the rules of Chess and we know how
to play chess we can tell people how to
play chess
um but there are many other things we do
such as walking such as swimming such as
singing such as painting that we do
that's extremely hard to explain in in
words those sorts of things we basically
could not get machines to do in the last
10 15 years basically with the Advent of
web um you know internet and in a cell
phone technology Etc this is like a tons
and tons of data of on about all these
activities that's on the web and so you
can actually train systems using that
data in a way that's not that different
from what happens to our kids you know
the kids basically in the beginning
learn to walk learn to you know in the
swim or whatever basically by trial and
error and by looking at the world and
it's only the things like math that we
teach them later so the funnier thing is
that kids went that way from you know
tacit what's called asset knowledge to
explicit knowledge AI systems did
explicit knowledge tasks before like you
know in back in late 90s we already
defeated casparov in in chess but now we
can actually recognize movies recognize
pictures and develop you know generate
pictures you know corresponding to any
prompt and most recently generate text
corresponding to any prompt and that
essentially is all of writing so pretty
much all of conversation all of writing
is I ask you to you know tell me
something about the following topic and
you start talking and then I interrupt
and say something else and then you
continue and so so it becomes easier to
generate these sorts of things now how
is this sir changed the landscape of
what students are doing to write their
papers and take tests and do finals
I I think so you know technology has
been you know evolving quite a bit over
the last many years and so when Google
came along already uh students can do a
lot more research a lot easier without
having to actually step out and go to
the library they can essentially sit in
front of the computer and look at
Wikipedia look at this and all these
things that was the positive side the
negative side is that if they didn't
really want to do their own thinking
they can copy from Wikipedia copy from
various books and submit those things
and this became kind of a cat and mouse
game and so Technologies are developed
on the educational side which can spot
directly copied passages right like
there are there are companies like
turnitin and various other things which
will compare people's essay students
essays or even students code to known
code repositories and known tax
repositories they can get cut what
changed now so that is already going on
right now and what changed now with the
Advent of these sorts of chat GPT style
Technologies
is they don't directly index and
retrieve entire passages it's sort of
like your and my memory if we read about
something and we are telling it to our
friend we give the gist of it in our own
words mostly because we don't actually
remember the exact passage that we read
we are just essentially able to complete
you know the gist these systems almost
work that way ah essentially they are
stringing together words uh that would
be a logical completion of the prompt or
the question that you gave them what
that essentially means is that the essay
they put together for example will never
be anywhere on the Wikipedia you can't
prove that the students have actually
copied it verbatim from anywhere else
um and so that obviously is a new kink
in this cat and mouse came again
everything has a huge positive side for
example
students can use these sorts of systems
to improve their writing you know they
can put together their rough thoughts in
maybe English that's somewhat rough
especially when you have lots and lots
of students who for whom English is not
you know not necessarily a first
language and then they can have their
chat GPT style llms large language
models to rewrite the text in fact I can
tell you that I get tons of mails from
International students asking for
internships asking for researchers and
positions Etc and ever since chat GPT
the quality of English has become more
like Midwest now so it's no longer
doesn't sound like there is no such
thing as Indian English or Chinese
English it's all whatever chat gbt
rights sort of a thing that could be
seen as a positive thing because they
are getting their message across but in
terms that I can understand but the
other side of it is they can also use it
to for things like plagiarism
are professors around here you can
Target professors worried about that
around here I yeah in in general
Academia has been quite worried since
like December because I think November
30th is when open AI put Chachi PT out
into complete open
um usage and and the university started
hyperventilating as to how especially
certain kinds of things like SCA based
evaluation you know in English and so on
it becomes a much harder thing and and
so the universities are actually having
including it ASU they are having you
know
um think tanks and discussion groups to
figure out what is the best way to
handle this you you know some have taken
extreme steps in some universities like
we don't allow access to chat GPT
um but others are taking actually a more
interesting proactive steps for example
one of the things that's well known is
as I said charging PPT and are are llms
the large language models and I keep
saying GPT but there are many other
things like Bard in Google and so on all
of these are essentially dynamically
generating the answer and it's not the
answer is not really factual are are
intentionally a lie it is just a most
likely completion of what the question
was so if you said this question it's
almost like if I say he you know stepped
onto the road and looked left end and
then you will complete in your mind
right because that's the most likely
completion except they're doing this you
know with much bigger context you know
given the last 3000 words what would be
the most likely words that we need to
string together
when they do that they can actually make
up stuff you know it's been called
hallucination but essentially everything
that is we do this human memory is very
valuable which is why in courts when
Witnesses are not trusted like right
away because we our memory was almost
this way we string together pieces to
make into stories sometimes the stories
are not actually completely true and and
these systems have the same exact
problem and so some professors have
actually used this to say okay you can
ask chargpt for the answer to this
factual question but then you need to
correct it
do you see what I'm saying because it's
like it it sounds it sounds very
authentic because that's part of our
problem we tend to think that people who
speak smoothly also know truth right
that part is like a cognitive flaw that
you know human we have we humans have
that we think that people who are well
dressed people who speak
smoothly must know what they're doing
that's why most of the Khan jobs are by
smooth trackers not like people with
right but that has to change now because
everybody can speak smoothly with you
know charging in the background and so
from the professor side we need to
actually look at the content which is
longer it takes more time and from the
student side you can ask them if you use
chat GPT if it's a factual question it's
a question which actually has you know
answers that are true versus false you
need to go through what it said and
figure out what is right and what is
wrong to give you an example if I go and
ask GPT 4 which is like the latest and
greatest system give me a 500 word bio
of Subaru Karma party that's myself
right you know because that's the most
narcissistic thing to do and it will
give a 500 word bio
it would look more or less kind of
correct and it turns out that it makes
up a whole bunch of things but I would
know that
do you see what I'm saying so for
example you know when at the last I
asked it it said I'm a president of a
particular Professional Organization I
was a president of triple AI which is
you know artificial intelligence
organization but it said I was a
president of computational linguistics
Organization for somebody outside they
can't tell you know it looks you know
maybe it's right do you want someone to
say but I do so the point is so if
suppose somebody were to make a
biography for some you know question I
can make them check double check with
Google so still by the way it turns out
that Google when it points you to
um credible sources like Wikipedia is
still a lot more factual than chargpt
and these llms can be and currently
that's just the reality they are
basically like you know worse than our
memories and our memories are valuable
you know we can remember a few things
but if you want me if I ask you exactly
what happened yesterday in your life you
would basically unintentionally
non-develop greatly make up a whole
bunch of stuff you know it's not
necessarily that you're trying to lie
it's just because you haven't indexed it
it's not a video in your head you are
putting pieces together the order will
be mixed up you may have done one
activity before other activity but
you'll remember the other way around
that's exactly what these systems wind
up doing
um and and so you can use that you know
basically with the students and say if
you're using it you know double check
that there are errors there are other
ways in which you can actually catch
plagiarism in terms of so one idea would
be if you use you know a large language
model to produce some text can there be
a watermark on it so that I can tell
that this part was produced by these
kinds of systems this significant
research that's going on there right now
and one particularly interesting idea is
essentially because these systems have
this ability that suppose if I were to
ask you to tell me your life story
without using the following 15 words
it's hard for us in fact I think you
know there are like TV games of this
kind saying you know say whatever you
want to say but without using Volume 15
words you can make charity PPT to write
an essay without using the following
twenty thousand words
so you know English approximately has 50
000 word functional vocabulary so I can
say don't use this Blacklist words right
right and then the the thing that comes
out it's paraphrased and so you can't
tell the difference it will be still
English okay but then if you know what
the Blacklist words are then you know
whether in fact it was generated by the
large language model are normal people
because normal people don't know what
the Blacklist is so they would actually
start using those words this is a this
is actually a pretty good way of
checking a watermarking the text except
for this to work the people selling
these systems need to want to help you
know and it's not very clear that they
want to because if they're selling this
service they probably would want people
to able to be able to say we wrote this
because the bigger Money Maker is the
large group of students around not just
students it's actually it's I mean we
are talking because we are in the
University we are talking about the
students but really this stuff was not
made for students I mean students are
the smartest ones they are always ahead
of the car when they try to use the
technology for themselves but this is
for anybody this is for writers this is
for liars you know every day you read
that Liars instead of writing like an
entire case summary they tend to write
some you know bulleted Point stuff um
and then chat GPT are these large
language models converted into English
in fact there's a joke now that you know
people will take five bullet points that
they want to send an email to their
friends and give it to charge GPT it
will make it into a two-page letter very
nicely Written Letter and then those
friends will ask Chachi PPT to convert
into five bullet points right so between
us we are just increasing that you know
that traffic on the internet but in in a
weird sense you know we have this
problem you know as humans we want
things to be well written but we also
want gist of it in bullet points you
know if you anyway wanted bullet points
why can't I just send you bullet points
you will be offended if somebody writes
you a letter with just bullet points but
it wants you need to have this so you
know Allah is full of this and so
basically they can have this this sort
of uses of you know these kinds of
Technologies and even there and that
could be a reasonable use
um right in the case of students of
course you might be more interested in
plagiarism detection but you know and
and so this Blacklist whitelist thoughts
of approaches our way are a way of you
know capturing that a couple more things
that I'm done
um you're right yeah
how much do you think uh students are
just using this technology in general or
is it still new enough I think again I
think typically the students I'm in the
College of Engineering our students are
of course the smartest in being the
early adopters of All Tech in general
you know I mean I'm sure I mean we all
know that our kids are faster with
technology
um than than you know the older people
right and so so people are already using
this and they're using it for all sorts
of uses one of the most interesting
things is these systems were generated
as text completing systems that's all so
like you know it was there for a long
time for five years back onwards as
you're typing on your iPhone sometimes
it car you know basically suggests the
word because it assumes that this is the
word you are likely to type instead of
suggesting the word and the spelling now
they can suggest sequence of words
that's essentially the how the way the
technology will develop but people found
that it can actually be used you know
given that all of conversation all of
you know language activity is saying
something I say something and you say
something that is connected to that and
I say something that is connected to
what you said and so you can use llms
you can use this to do that and so
students have essentially actually one
of the early ones who have figured out
that you can use these sorts of
technique systems
to kind of paraphrase things to ask
instead of going to Google and ask tell
me all the research papers in this area
and give me a summary of what happened
in this area previously they had to go
old old days they had to go to the
library more recently they had to click
on Google Now they just asked chat GPT
it will give a summary but when it gives
a summary oftentimes it's not necessary
that it's actually true it just the
summary is the most likely sequence of
words sometimes it's actually correct
sometimes it's not they've been cases
where charging Beauty would generate
citations that is actual paper names
with the names of journals they were
published in all made up these journals
don't exist these papers don't actually
exist so I actually did the same thing I
did one more narcissism question where I
asked charging Beauty for you know in an
area I work in human every eye systems
what are some of the best papers hoping
that you'll say you know one of my
papers it did say one of my papers
except I never wrote that paper
so I quickly actually have to write one
and put it on archive so that people
will cite it so the point again is
students are finding very interesting
ways they have actually been cases where
recently in the news for example people
have found based on what Chachi PPT said
people got into trouble like it made up
an entirely convincing story that's a
mayor was a criminal and that poor guy
wasn't or that a law professor actually
harassed his students sexually harassed
students and there was no such thing if
you ask a leading question it gives a
nice answer to it and it has its
completely there's this old movie called
absence of malice and that's playing out
in in screen right now because chat GPT
doesn't mean to hurt you mean to help
you it is just trying to complete the
sentence and all the meaning and all the
truth Etc is in your head when you
actually look at it and so students
are finding the you know both the uses
and also kind of finding the pitfalls of
it and and I believe that it's it's a
it's an interesting new world right now
yeah you can make it cat Matt Mouse game
cat and mouse gas will move on exactly
to the next level yeah what do you think
is is next coming um down the pike for
all of this type of thing I I think this
type of knowledge my senses will get
used to it in fact one of the most
interesting thing hopeful things that
I've heard is this philosophy Professor
I mean after all philosophy you would
think philosophy I say should be easy to
make right um and this philosophy
Professor basically had like two things
to say when in back in December when the
charging came he said by January he was
like trying bucket saying suddenly all
my students essays looks so much more
coherent and I know that they clearly
are not writing this stuff okay by March
he was saying that I can spot a chat GPT
essay from a mile away
do you understand what I'm saying
because we get used to it I mean it's
like
we do always we can still tell the
difference between people who are faking
it and people who are actually who know
what they are talking about maybe you
can be taken a couple of days a couple
of times but if you are actually
interacting with the person who is
Faking it mostly without knowing what
they're doing you
Browse More Related Video
Dampak AI di Dunia Pendidikan Bak Pisau Bermata Dua
Macht Chat GPT Schรผler dumm? Oder schlau?
10 Ways Students Are Using AI to CHEAT
Jenni AI vs PaperPal: Best AI Tools for Academic Writing in 2024
Summit Fernando Dรญaz Chief Learning and Technology Office Mentu GEF 2024
AI Solutions for Teachers and Professors
5.0 / 5 (0 votes)