Discussion Topic: 1/28 Guest speaker comment1/28 Guest speaker comment
What did you learn from today's speaker, Dan Russell? What was the
best question from the audience? Did you ask it?

Reply from Yassin Mudawi

I learned more about the evolution of AI over the years and how
various machines implemented aspects of what we enjoy in LLMs
today. For me, the most fascinating part was learning about how vastly
more complicated Google search is than I previously assumed.


Reply from Joanne Zhao

My main takeaway from Dan Russell's talk was the importance of making
AI systems easy to use and understandable for users. His example of
the AI program he wrote to translate a Xerox task into concrete steps
stuck with me, as the perfect accuracy of his program did not matter
because users couldn't follow those steps. I thought the best question
from the audience was the one about Russell's thoughts on Waymo cars
in SF, and more broadly, how to ensure technology isn't negatively
disruptive to the communities it is deployed in. He had a unique
perspective given he lived in close proximity to Waymo HQ, so he was
used to seeing Waymo cars on the road for many years. His response was
centered around the accuracy/safety of Waymo cars and the camera
capabilities they have that human drivers don't. I wonder what other
potentially negative impacts on communities could be besides the
safety concerns he focused on.

Reply from Gavin Onghai

I learned about the power of search and how so many people
overestimate their search abilities. I also learned more about the
importance of the user experience.

One question that I thought was good was about how the CS curriculums
would change in the future. Yassin Mudawi asked the question. Dan said
that alot of the assembly programming, he believes, will become
outdated. I'm not sure if I fully agree, since everything is based
upon assembly programming.

Reply from Lakxshanna Raveendran

I learned a lot about user experience from the beginning of this
lecture. Some of my main takeaways were about the importance of user
testing and not relying on "intuition", especially because if your UI
is poor, then nobody can access the true function of your product. The
Xerox machine example was very helpful. I liked the question (that I
did not ask) about how he saw CS curricula in colleges changing over
the next few years. As a student and a CS peer mentor, I've already
seen such significant differences in both coursework and how students
approach their psets. Like Dan said, I hope for the sake of future
students that learning assembly code is phased out of the core
curricula.

Reply from Xinyuan Zhu

He shared insights from his 30-year career as a "techno-cognitive
anthropologist" studying how humans interact with search and AI. A
major takeaway was that poor user interfaces can block even the best
AI, which he illustrated through his failed copier planner and the
social robot Jibo. He also highlighted a massive "Dunning-Kruger
effect" in search: while most of people feel confident in their
abilities, most search less than once a day and often stop at the very
first result without verifying information. The best audience question
addressed the gap between industry and academia; Dan noted a
staggering "$100 billion gap" in compute power alone, meaning
universities cannot train models at the same scale as giants like
Google or OpenAI. I did not ask any questions during the session.

Reply from Morgan Go

Dan's comment on unintuitive UI being a roadblock for AI really stuck
with me as someone who is passionate about UI/UX design. I think using
that perspective may help with how we choose to utilize AI (Maybe as a
test dummy to see how usable a product would be to a random person)
and his experience with emerging and changing technologies also helped
decrease any anxiety around the future of computer science/what we
should study next. His real-life examples of people trying to work
around the system (Man with the cart to prevent traffic in his
neighborhood) put into perspective how important good design is, and I
loved his response to the questions related to Waymo and it's recent
popularity boom (I did not ask this question but was very grateful
someone asked about it as someone who lives in a city with LOTS of
waymos)

Reply from Guangxing Cao

1. I learned that even very capable AI systems can fail if the user
experience is weak, because people rely on the interface to build a
correct mental model of what the system can and cannot do.

2.  Best Question:

how do we make sure that the research we do is not overbearing?
(because it forces researchers to think about ethics and user consent,
not just whether a method works.)

Reply from Cixuan Zhang

We learned that the User Experience (UX) of AI systems is critical and
must be integrated from the very beginning—rather than added as an
afterthought—to effectively support human reasoning and mental
models. To me, the best question from the audience was whether he uses
any voice-based LLM tools while driving. Dan said definitely not,
because the UI is awful. Even though I wasn't the one who asked, his
answer really resonated with what I’ve always thought, which is why
that was my favorite part of the Q&A.

Reply from Olivia Ye

I learned about a lot of side projects that Google does on the side,
as it was neat to hear form someone who ideated one himself and then
pitched it. I also enjoyed hearing about what people's mental model of
Google was, because I had never stopped to ask it myself.

I liked someone else's question about self-driving cars and their
reliability, as it was interesting to hear he would get in a Waymo but
not Tesla. I wish I had asked myself.

Reply from Diana Shyshkova

From Dan Russell's presentation and his work on studying user behavior
during information retrieval, I took away one key conclusion: how
important user interfaces are. Even the most powerful tools or models
will not be of much use if people don't know how to use them
effectively. His statement that more than 90% of users don't even know
how to use such a basic function as Ctrl+F truly struck me and
highlighted the huge gap between technical capabilites and actual user
behavior. This clearly showed that good design and usability are just
as important as the underlying technologies

Reply from Tracy Chen

I learned that we should not blindly trust our intuitions when using
AI systems. Even simple tools like spell-check can be
content-sensitive, and small wording changes may lead to different
outputs. He also emphasized that AI systems are often updated without
notice, and if nothing goes wrong, users may not realize that the
system’s behavior has changed.

Reply from Esha Garg

I learned a lot from Dan's discussion last week, but I think that my
favorite part was his emphasis on what is not AI. Today, the term AI
gets thrown around a lot, both in the computer science world as well
as outside of it, such as the business world. It was cool to see how
for a seemingly simple Google Mail mechanism of checking if you've
attached a file is simply a semantic search to see if you have
included the word "attach" rather than an "AI" methodology. I really
liked learning more about the inside of how Google works.

Reply from Lucas Liu

People are somehow too confident about their knowledge but they dont
even ctrl f.

Reply from Yna Owusu

I learned a lot about being a better searcher. The best question was
about the outlook of software engineering as we know it. No I did not
ask it.

Reply from Andy Ma

I was very surprised to learn that most people here at Yale are
unusual in ways that I hadn't considered (knowing how to use Ctrl+F,
Googling/using ChatGPT many times per day, etc.).

The question that stuck with me the most was my own about how the
conflict between AI and news outlets will be resolved.

Reply from Weixing Zhang

I learned about the importance of detailing and understanding user
needs. I also learned about the swipe-to-text feature. A question that
I thought was good from the audience was what kind of car he drives
and his insight on self-driving cars like Waymo versus self-driving
Teslas.

Reply from Isabelle Millman

The most interesting thing I learned from Dan Russell is how few
people know how to use find to search pages for things. The best
question from the audience was "what car do you drive?"

Reply from Zikang Chen

A lot comments on AI and the last part of the AI written music. That
really shocked me. The most fun question to me was the question about
vehicle. The insights on auto driving and new energy cars are still
thoughtful. I did not the one who asked this question.

Reply from Emmett Seto

From the guest lecture today, I learned a lot of fun facts. For
example, he mentioned that his friend was the one who created the
swiping effect in texting and earns a nickel everytime it is
used. Additionally, something else that I learned that was interesting
was that Gen Z the first place they always look are the comments of a
video. I thought his overall talk about HCI was interesting and the
way our current generation interacts with certain commands. One of the
questions asked was that currently a lot of startups are essentially
GPT wrappers. If the only difference is the user interface, will these
startups survive? I had asked this question after class.

Reply from Thomas Chung

I thought the coolest part about Dan Russell coming in was hearing
such an inside perspective on the industry that is also extremely well
connected. It was so funny when he asked us if we we had ever used a
technology then casually dropped that he knew the person that made
it. I think this ultimately built his credibility on the future of AI
and its integration in human society. I thought the best question from
the audience was not asked by me but pertained to how we should
utilize AI in our own futures, and how it should be absolutely used as
a tool but not relied upon.

Reply from Hubert Wang

Dan’s argument that AI should explicitly disclose its guardrails is
compelling. In my own experience with image generation, I’ve often
encountered unexpected or wrong outputs, wondering what internal
constraints were at play. One of the questions asked was regarding
Waymo and AI's application in autonomous driving. I followed up with
Dan after the lecture to explore what the future holds as autonomous
vehicles become more prevalent. Clearly this field still has room for
improvement and needs a lot more testing, but it's on the right track.

Reply from Omar Espiricueta

My favorite thing I learned was the idea that we are not
"normal". When designing user interface systems, we must remember that
we are not the target audience, nor are many of our peers. To make a
successful product, you must design with the masses in
mind. Understanding people's behavior, tendencies, and desires is far
more critical to a helpful, successful product than a "cool"
interface.

Reply from Kemi Omoniyi

Finding information online is less about having “smarter AI” and more
about designing a UX that supports how people actually think and ask
questions. He emphasized that confidence often outpaces skill, so
users may accept the first plausible answer, especially when the
system isn’t obviously wrong, and normalize subtle errors.

Reply from Alba Quintas Núñez

I think that Dan Russell was a very inspiring speaker. I believe that
his approach of the revolution of AI in our lives was very unique and
interesting. Sometimes we tend to believe that, when it comes to
developing new technological tools, the key component of their success
is strictly tied to how effective and sophisticated these technologies
are. However, Russell raises a very important point: the user
experience is fundamental. It does not matter how competent an AI
system is if the average person does not understand or know how to use
it. In the last part of his talk, he proceed to leave us with the
message that, as much as AI can be really good, we still need humans
to think behind it, we still need a person to develop that user
experience.

I think the best question from the audience (I did not ask it) was:
What do you think is going to happen with AI in the CS curriculum of
higher education in the future? Even though it is impossible to
predict exactly what is going to happen, Russell recognised that this
tool is already completely changing this subject, but he still
encouraged us to never stop reasoning and thinking.

Reply from Tony Chang

I learned how important the connection between the users of AI and the
AI model itself actually is for the user experience. For example, when
the model behaves unexpectedly, it would be widely beneficial to
provide an explanation for the strange behavior. However, it also
becomes a slippery slope as to how much should be disclosed to the
users, as sometimes volatile and unsafe information could leak. It's
shown with ChatGPT and users phrasing their query in a "nice" language
to get the dangerous information they desire. I think the best
question from the audience was how he saw the CS curriculum changing
in the coming future, as it has become a widely relevant topic with
the release of AI. I thought that it was good to see the perspective
from someone who has worked closely with academia and industry,
especially one acquainted with AI and user experience. I did not ask
this question.

Reply from Felix Zou

The general overarching lesson is the UX is important. This may be
somewhat common sense, but the examples provided of UX failures on
Google apps (as well as the first printer one) do reinforce this
point.

Another point that Dan Russell is trying to drive home is how unaware
the majority of people are of certain tools/shortcuts like
control-F. This is new to me, as I have never thought of myself as
being particularly keyboard-shortcut-savvy but somehow actually rank
in the top 10% just by knowing control-F. In any case, it speaks to
how easy it is to miss the experience of the majority of users if one
judges UI design using one's own familiarity.

Towards the end he also discusses AI trust, which are certainly
relevant and timely, though he does not provide any useful guiding
framework and/or solution. This is understandable, since getting
symbolic human intelligence to trust purely statistical ML systems is
never easy, and perhaps is not practically possible without making
deceiving the human.

In terms of questions, I liked the first one about what car he
drives. It is funny and simultaneously relevant to the topic. I did
not ask the question.

Reply from Will Yang

Prof. Russell offered great perspective on the human-machine
interaction aspect. As computer science students, we typically focus
on designing efficient and effective systems that calculate complex
equations well or predict the next token perfectly; however, we often
ignore that these applications and programs are ultimately used by
humans (at least in the near future). Because of that, it's incredibly
important to also think about UX design and how people will actually
use the program, about whether it's intuitive and helpful enough for
the target users. That's my central takeaway from the guest lecture.

A very good question was asked about how advances in AI and LLMs might
affect the future CS curriculum, which I think is especially relevant
as the required skill set continues to evolve. I didn't ask that
question, but I did ask another one about the gap between academia and
industry, which I also care deeply about.

Reply from Yide Jin

From today’s speaker, Dan Russell, I learned that the hardest part of
useful AI is not only the model. It is the system design. A real
“connected” AI system should let tools communicate with each other,
keep updating based on new information and user feedback, and try to
understand the human goal over time. Just as important, it should
explain its actions in plain language so users can trust it, catch
mistakes, and steer it easily.

The best audience question was about the gap between universities and
industry, and how we turn strong academic ideas into real products and
real impact. I did not ask it, but it resonated with me because this
gap looks different across countries. In China, there are more
structured industry and academia integration models. For example,
Shandong University has school affiliated enterprises such as Shanda
Dareway Software Co., Ltd. (山大地纬软件股份有限公司), and its
leadership includes university faculty. The company’s president, is
also a Shandong University professor. The company’s chairman, is also
described as a Shandong University professor. This kind of close
integration can reduce the “last mile” barrier and make it easier to
build AI that is deployable, explainable, and user friendly in real
settings.

Reply from Sean Lee

I learned the importance of building intuitive and transparent UI/UX
for AI systems. The user interface serves as a critical bridge between
users and the underlying system, so it is essential that it accurately
reflects what the system is actually doing. If the interface is
unintuitive or opaque, users may fail to take full advantage of the
system or even misunderstand its behavior. Dan Russell also emphasized
the importance of testing UI with real users, since there is often a
gap between how engineers expect users to interact with a system and
how people actually use it in practice.

I thought one of the best questions from the audience was about how
university CS curricula might change over the next few years. Because
CS education plays a major role in shaping the next generation of
engineers and programmers, I thought that this question highlights a
potential shift in how we teach and prepare students, which could have
a significant long-term impact.

Reply from Ayushi Das

I really appreciated how Dan described his work at the intersection of
computer science, AI, and human behavior. His introduction of himself
as a cyber-tribal-techno-cognitive anthropologist was especially
interesting. I found it fascinating how he works with multiple
datasets and makes thorough alterations to interfaces to make them
more accessible and user-friendly. One of the most interesting moments
for me was his response to the question about how he sees the AI and
computer science curriculum changing over the next five years.

Reply from Yuwang Ma

I learned that human-computer interaction about how results are
explained, how evidence is highlighted, how filters and facets are
presented, how the system supports “backtracking” or “branching” lines
of inquiry can either help or hurt the user’s ability to explore. I
realized this is also important as algorithm is. I can remember a
question about "How can a search interface make its “reasoning”
legible without overwhelming users like showing why something ranked
highly, or what evidence supports a summary." I didn't ask it though.

Reply from Alex Lu

I learned that users don't have a good mental model of how AI works,
partially because underlying models can change very quickly. I also
learned that the average search user uses search less than once per
day, and that a majority of internet users do not know ctrl + f. The
best question was someone else asking how San Francisco has changed
since Waymo has started operating there. It made me realize autonomous
systems like robot taxis are maturing and being deployed out in the
world faster than I thought.

Reply from Omar Abdellall

I learned a lot about search and the deep research that went into
it. There were some mind-boggling statistics about how people use
search though, so I was surprised about the overconfidence of our
search abilities. The best question was asking about what car he drove
-- then following up to ask about autopilot & AI. I did not ask it.

Reply from Thomas Luong

I learned about his thoughts on human-centered perspectives on AI. His
work at Xerox PARC with the AI scheduler system project was
interesting because he described how his team walked through the
process of creating a UI that incorporated sample user input and
research --> motivating question of his work is "how can we study and
improve user experience with AI?"  - How do we make sure that the
research we do is not overbearing? --> continuously test these things
and make sure that what you're doing is useful for the team and useful
for humanity.

Reply from Arya Bhushan

One thing I learned from Dan Russell is how decision systems,
especially in the modern age of AI, can be a two-edged sword. While it
can be incredibly helpful in researching and consolidating
information, it can also often reinforce mistaken information when
people try to verify misinformation. This raises a crucial AI and UX
design issue of developing responsible AI.

My favorite question from the audience was "How do you think the CS
curriculum in universities changes over the next 5 years?"
Unfortunately, I did not ask that question.

Reply from Jiashu Huang

I learned that user experience is usually overlooked but a very
important part in AI software development. The best question is "how
would you predict the trend of future CS education in college". I was
not the one asking it.

Reply from Bende Doernyei

I truly enjoyed both the lecture and the dinner with Dan Russell. He
made me think outside of the box and see how many different ways
computer science and HCI tools could have evolved. It also made me
think whether LLMs deserve the current hype, since it is overfunded
compared to other, potentially more valuable AI developments (and
still very much inaccurate, see: "dutch composer" example). Over
dinner, his comments on delviery robots and Waymo security vs people
were especially interesting. I attended another talk the day after
about AI, and the speaker differentiated AI technologies that are
substituting smart humans, doing the same thing that smart humans
could do, vs. AI technologies that are doing things that even smart
people could never do, like gene analysis or identifying the author of
a book by looking at just the filler words. I think, consistent with
the thinking of Dan Russell, the second version has brighter future,
but the first one can help us focus less on substitutable skills while
learning greater ones.

Reply from Youxuan Ma

The most important thing I learned is that your AI product is only as
good as your UX design: if people can't understand how to navigate or
use it easily, no one is gonna use it and the product will fail
eventually. Also, another interesting thing I learned is that we are
really not "normal" people, and we have to put ourselves in other,
"normal" people's shoes when designing our product.

I think the best question asked was: "how do you see the CS
curriculums change/develop in 5 years?" I didn't ask it.

Reply from Annabelle Huang

Dan's friend invented the swipe type system, which is really cool. I
learned we cannot really type swipe low frequency names in it, because
words like proper names don’t work because they’re not in the training
set. I think the best question is: How do you see the CS curriculum in
the next five years? Dan has taught the human computer interaction AI
course many times, and the course curriculum has changed a lot in the
past few years. Dan thinks AI coding is getting better really fast, so
it opens up for interpretation whether down the future we will still
need to learn skills like assembly code, where it might be a skill
that we can leave AI to handle.

Reply from Sam Meddin

I really enjoyed learning about how much more complex the systems we
use are than we realize yet some things are much simpler and just as
popular like typing using the drag feature. I liked the question about
autonomous vehicles specifically teslas versus Waymo’s which I did not
ask.