AI in Education:

Defining AI Chatbots
DATE

NAME
TITLE
UNIT
CONTACT

1
Today’s primary
learning objectives
• Increase feelings of acceptance, curiosity,
and motivation regarding generative AI
tools.
• Define common terms and concepts in AI.
• Summarize how generative AI tools work.
• Try using an AI chatbot in an informal and
exploratory setting.

2
 Acknowledgements

• ChatGPT4 for summarizing, brainstorming, and

suggestions on the content.
• Midjourney & DALL-E to generate images
• Designer feature within PowerPoint to suggest slide
layouts and stock images.
• Grammarly for spelling, grammar, conciseness, word
choice, and so on.

• Smartsheets for automating registration.

• Outlook and Zoom for event coordination.

• Google for web search...

For the sake of transparency and to model best practices, here are some AI
tools that I used when developing this workshop.

I used ChatGPT, and Midjourney and DALL-E, Designer in PPT. Oh yeah, I

also have Grammarly installed. And if we mean algorithms and machine
learning, not strictly LLM. Then I technically should say I used Google a lot too.
I also used Outlook for event planning, it has AI features too. And Smartsheet
for the registration forms. Zoom! Zoom has AI features too.

Yes, I am being melodramatic! But my intent is to suggest, hopefully in a

humorous way, that there are fuzzy and ambiguous lines of distinction here.
And warm us up to this topic. This is an ambiguous, and often contradictory
topic. We are entering a twilight zone.

3
 Let’s be flexible

(Image generated by DALL-E, 1/17/23)

This is a big topic so we can’t cover it all. Also, because it is so emergent, nobody has
all the answers. Those answers are emerging in places like this. So, thank you for
joining the discourse!

Everyone is at a different starting point. Some may be learning the very basics, some
may be already thinking about bigger issues, or have begun adapting your course.
Some of this you may already have heard, other parts might be new to you. Wherever
you are, let’s be mindful that different people are at different points and be flexible
about where we might end up today.

The benefit of being together here is that we can share our thoughts with each other.
This workshop will generate more questions than answers. It is a starting point for
you to engage further with each other, CTL, and your own learning.

My expectation is the we will improvise a bit and hop around different topics that
interest you. So let me know as we go, what you are thinking, if we want to skip
ahead or circle back. Maybe even things that aren’t on here yet!

4
 Why should you care
about AI?

• Our perspective and expertise is important

to the discourse
• The landscape is evolving quickly
• Our students and instructors want guidance
• Using AI could improve how we do our work

• Other groups on campus are, as expected bringing their perspective to it, whether
it’s computer science, business, medicine, and so on. Also, in the bigger discourse,
Higher Education is being scrutinized. Think Claudine Gay or Liz Magill or even
Michael Tessier-lavigne. People are asking, are we preparing students for success?
Is Stanford leading the way responsibly? Is a college education valuable? What role
does education play in our society?

• I’m of an age where WWW and web search came to prominence when I was in
high school and college. I’m a 90’s kid. Some of us might remember dozens of
different search engines, Lycos? Alta Vista, Ask Jeeves? Dogpile? It seemed to be
evolving so rapidly. I remember teachers saying “Don’t cite anything from the web!
It’s not a valid source” Then the next semester saying, “You can cite web sources
from journals or news articles, but Wikipedia is not a valid source!", then later
“Wikipedia is okay, but be sure to check the references!” It was constantly
evolving.

• Students want guidance. For many of us, this is not our first rodeo. But for some of
our students, this might be. And for some of us, this might feel like a first rodeo. If
we abdicate then for-profit will fill that void, students will go and do it anyway. We

5
 owe it to our students to give them guidance. .

• Lastly, AI can help us improve how we do our work. Web search, googling it, is now
an indispensable part of how we do our work. We couldn't do our work without
web search. People smarter than me are predicting that AI ubiquity, AI everywhere
is coming soon. And it’s going to become a routine part of how we do work like
web search has become. And it will have a big impact on how we work, so it makes
sense for us to get ahead of it.

5
 How do you feel when thinking
about the future of AI in education?
Why do you feel that way?

6
 Navigating new
technology
• Connect to community
• Have a growth mindset
• Take care of your well-being
• Engage in the discourse

Some of you might be like, “That’s so cool. It can be trained to do all kinds of cool
things.” Or maybe “OMG, that is so dangerous! How can you trust it?”

Whatever your emotional responses to AI, it is all valid. We’ve been through periods
of tech disruption before. Remember when Wikipedia and Google first came out?
Remember when they said open online courses were going to revolutionize
education? Remember the emergency pivot to teaching online?

We have strategies that help us navigate any big change or disruption. Growth
mindset, community, and well-being. And joy is a part of that!

We know how to navigate this. We just gotta roll up our sleeves and do it again.

7
 Facial Social media
Spam filter
recognition feed

Web search Maps &

Autocomplete
How many AI- algorithm Navigation

powered tools Grammar check Auto-captions

Smart home or
have you used speakers

lately? Movie/song Driver

Automated
customer
recommendations assistance support

Optical
Voice to text character Photo filters
recognition

8
 Common AI Terms

Generative AI
Model
Large Language Model (LLM)
Chatbot
Deep Learning
Training Data
Algorithm
Prompt

AI is a big catch-all term. Like “transportation” can mean vehicles, infrastructure,

policy, etc. Let’s examine a few common terms.

Generative AI – “Generative AI” is AI that leverages the prediction powers of AI to

generate things: language, images, audio, media, and so on. AKA AI generators

AI model A model is an AI software program that has been trained on datasets to

perform a specific task.

Large language Model - So a LLM is a complex model trained on vast amounts of

data that generates language that resembles human-generated language.

Chatbot – A program that runs on an LLM. The interface is in a chat or text message
interface, where a person interacts with the program in a back-and-forth conversation
via typed text.

Deep learning - A subset of machine learning inspired by how biological brains are

9
structured. It uses a technique by which a computer can learn without being directly
programmed with rules. Deep learning layers these techniques for sophisticated
output

Training data—Labeled data used in the training process to "teach" an AI model or

algorithm to make a decision. For example, with an AI model for self-driving vehicles,
training data may include images and videos in which traffic signs, pedestrians,
bicyclists, vehicles, and so on are labeled.

Algorithm—A set of instructions or rules for performing a computation. Developers

typically design algorithms used in AI to progressively iterate themselves, which we
can consider a form of machine learning.

Prompt—Instructions entered by users to direct an AI generator to generate an

output or complete a task.

9
 How machine learning works

2. Select a
3. Humans 5. Create
1. Start with model to 4. Test the
identify train the software
labeled data model model applications
patterns

Other
training
methods

1. I give an AI model a zillion photos of people. I’ve labeled some of the photos as
“men” and some as “women”.
2. The model is gonna analyze the color, shape, angle, and so on of every bit of these
photos.
3. I ask it to guess if a photo is a woman or man. When it guesses right, I say “Ok”,
when it guess wrong I say “No”. Each time it guesses it updates its own code.
4. After many rounds, I might add more labels. Maybe I add more diverse photos.
Maybe different quality of photos, or people with different hairstyles or clothing. I
don’t really know how it exactly chooses it. I’m not telling it “photos of women must
have these characteristics”. It guesses, and I validate. Over time it gets so good at
guessing, that I can assume it “understands” what a photo of a woman or man is
(even though I know it doesn’t understand in the same way a human might, but I can
rely on it to guess correctly) Then I repeat this for all kinds of different faces or
characteristics, like age, nationality, etc.
5. Assuming I eventually get this facial recognition AI model and algorithm working
reliably, maybe I create an application that uses it for medical diagnostics. Or maybe
an app that uses it for security purposes.

10
You probably can already find shortcomings in this process.

10
 How to confuse
machine learning

CHIHUAHUA OR MUFFIN SERIES. (n.d.). Karen Zack ! Creative™. Retrieved April 8,

2024, from https://www.karenzack.com/work/recognition-series

Nesher86. (2021, May 14). How to confuse “Machine Learning” [Reddit Post].
R/ProgrammerHumor. www.reddit.com/r/ProgrammerHumor/comments/nceia0/h
ow_to_confuse_machine_learning/

"

The original image "Chihuahua or muffin?" by Karen Zack was reinterpreted as an

insightful jab at machine learning in the meme "How to confuse machine learning”
posted by Reddit user Nesher86.

What about people who appear differently in different contexts? Where do all the
training images come from? Is it representative of diverse kinds of people? What
about the biases of the people validating and training the AI?? Or people who
appearance and identity doesn't fall into those boxes?

What if your head looks like a blueberry muffin?

AI’s that detect and generate images are different than the language AI. And LLM are
much more complex and sophisticated. Instead of looking at the colors and shapes of
images, AI generate language by using mathematical representations of the statistical
relationships between words.

But the machine learning process is similar. Remember that is a tool. It has
idiosyncrasies, strengths and weaknesses.

11
It is fire. It can cook your dinner and it can burn your fingers. So, its important to
understand a little about how it works

11
 AI-generated language is unique

• Super-advanced auto-complete and word

prediction
• Average of enormous amounts of data
• “Randomly” and “statistically” generated
• Little, if any, self-awareness and contextual
awareness

AI is “intelligent” in a way that mimics human

intelligence but is fundamentally very different from
human intelligence.

In the end, LLMs generate language in a unique way. It is quite different from human
intelligence.

In some ways “artificial intelligence” is not a good choice of words to describe what
these tools are. They are not intelligent nor have intelligence. But I guess “AI” just
sounds cooler, and it stuck.

12
 Human-generated
language is unique

• Self-exploration
• Self-expression
• Relating to others
• More than just conveying information

Use AI to enhance human-

generated language not replace it.

Humans have many qualities that AI don’t, (So far at least)

• We are biologically embodied
• We are each unique and different
• We are part of societies and cultures
• We have personal histories, memories, and psychological experiences.

This all factors in to how we use and generate language.

13
 AI-generated
language
&
Human-
generated
language

I believe they can complement each other.

I’m reminds me of the AIs that play chess. Computers could defeat world champion
grandmasters of chess in the 90’s. Then why do more people play chess today than
ever before? Maybe it’s that AIs made it easier for people to learn and become better
chess players. And maybe having an AI that could beat anyone, helped people
appreciate the value of playing with another person?

So how do you beat an unbeatable AI? Simple, humans teamed up with AI. These
days human-AI teams, cyborg or centaur chess teams, can beat those once
unbeatable AI or human grandmasters alone.

Another example is going to the gym. We can make robots that are better than
humans at lifting things. But it’d be foolish to say “Hey let’s invent a way to automate
the gym, so the weights lift themselves and people won’t have to go to the gym
anymore.” That’s not the point of going to the gym! Or of going to school! But we are
saying “Can technology make exercising better?” “Can it make going to school
better?”

14
The point here is that I believe that if we communicate to students and hold in
ourselves this view that human-generated language is unique and more than just
about transmitting information, we might be more inclined to us AI to enhance rather
than replace our voices. And that we might more easily view writing or generating
language as a way of learning, not just a way to demonstrate learning at the end.

Writing does not equate thinking deeply. Just because a student wrote something,
doesn’t mean they thought deeply about it. Especially in a world with AI. And
conversely, a student can think deeply about something and have great ideas and
thinking, but written words alone don’t capture that. Especially in a world with AI.

One thought I offer is, how might you make it so that students still learn, even if they
use an AI to generate words? If a student has a long rich conversation with a chatbot
about ideas for an essay, but in the end, the AI generates some of the words,
sentence structure, and grammar that the student verifies, critiques, adapts, and
edits. Is the student still learning what you intended? Maybe, depends on what they
are learning.

There is no reason anymore why anyone with AI should be turning in an unstructured

essay with poor word choice and unclear sentences. A chatbot can do that. But
unique voice, original ideas, depth of thought, personal expression, human
relatedness, contextual nuance. That is what humans bring to the table, whether
writing it themselves or coaxing it from a chatbot.

14
You won’t know until you try.

15
Which
chatbot
should we
use?

16
 Which search engine
should we use in 1997?
A. Excite
B. Yahoo
C. Dogpile
D. Alta Vista
E. Lycos
F. Infoseek
G. Use your phone-a-friend

(Image generated by DALL-E, 1/30/24)

Phone-a-friend : ”Dude, whats the name of that movie, where Tom Cruise slides
across the floor in his socks and underwear singing that old time rock and roll
song!!?? It’s been driving me nuts. What is it??”
RISKY BUSINEESSSS!!!!!!

Maybe we’ve forgotten, but there was a time when search engines first came out,
where you couldn’t just find the answer to that question. You’d have to maybe put
“Tom Cruise filmography”, then search each film “synopsis of Far and Away” “synopsis
of Days of Thunder” and then even then.

It took a while for search engines to become that sophisticated, and for us to learn
how to to use them best. Same with AI. It’ll take us time to figure it out, but we will.
Best way to do so, is to try it for yourself.

Get in there and try it yourself. That’s the best #1 piece of advice I can give you. Try a
bunch of different things. Get creative, be weird, push it and see what happens.

17
 Which streaming
service should we use in
2020?

A. Netflix
B. Hulu
C. Amazon Prime
D. Disney+
E. HBO Max
F. Kanopy
G. Pirate Bay

(Image generated by DALL-E, 1/30/24)

It kind of depends right? What kind of shows do you like? How much you want to
pay? What devices do you use?

18
 Accessing AI chatbots

Google Gemini Microsoft Copilot

OpenAI ChatGPT
(aka Bard) (aka Bing)
• GPT3.5 (good) • PaLM 2 (good) • GPT4 (best)
• GPT4 (better) • Gemini (better) • Web search
• Free • Free • Free
• Can use Stanford email to • Use non-Stanford Google • Use non-Stanford
create account account Microsoft account

Similarly, there is no one correct answer with AI chatbots. There are the big three
(and you might add Anthropic’s Claude to that list), which is a good place to start. But
there are also many smaller or more specialized LLMs as well. There really isn’t a
clear answer to which is the best at this point.

For our purposes, which is to gain fluency with AI chatbots, we recommend starting
with one of these three.

ChatGPT
ChatGPT, developed by OpenAI, uses the Generative Pre-training Transformer (GPT)
large language model. As of July 2023, it is free to those who sign up for an account
using an email address, Google, Microsoft, or Apple account. You may also need a
valid phone number to verify your account. See ChatGPT help documentation for
more details.
Go to openai.com/chatgpt and sign up to access ChatGPT.
Bard
Bard, a generative AI chatbot developed by Google, relies on the Pathways Language
Model (PaLM) large language model. As of July 2023, it is free to those with Google

19
accounts. It is not enabled for Stanford University Google Workspace accounts. If
you'd like to access this tool, please use your personal Google Account. See the Bard
FAQ for more details.
Go to bard.google.com and sign in with your personal Google account to access Bard.
Bing Chat
Bing Chat, an AI chatbot developed by Microsoft, also uses the GPT large language
model. Unlike most other chatbots it can access and search the internet. It is
available from within the Microsoft Edge web browser. Sign in to a Microsoft Edge
account to allow longer conversations with Bing Chat. See Bing Chat help
documentation for more details.
Download Microsoft Edge and sign in with a Microsoft account to access Bing Chat.

19
Try using a chatbot

• Remember to:
• Jot down your thoughts and feelings
• Reflect on the chatbots performance
• Share your experience with others

(Image generated by DALL-E, 3/1/24)

20
Start with something fun

Brainstorm ideas for a

Create a recipe for a new
vacation to a fantastical
fusion dish
world

Play a guessing game

Plan a surprise party for
about an interesting
a special guest of honor
trivia topic

21
 Complete the following
prompt regarding AI: "I used to
feel AI was... Now I feel..."

22
Continue to engage

• Use AI chatbots for your work tasks

• Read the AI Teaching Guide on the
Teaching Commons website
• Attend upcoming CTL workshops on AI
• Talk about AI with colleagues and
students

23
Wrap-up activity

Tinyurl.com/KenjiSurvey
(Image generated by DALL-E, 12/18/23)

24