Individual Assignment 1

1.1 Concepts, definition and history of AI and interaction with AI

The history of AI
AI came about during the second world war when the allies started making machines with
the objective of breaking the Germans communication devices. According to Grudin (Grudin
2009) the term Artificial Intelligence was first used in 1956 by American mathematician and
logician John McCarthy.

Discuss briefly similarities and differences between the HCI design

guidelines and the Human-AI interaction guidelines.
Definition from T.Bratteige and G.verne 2018:
“AI is a subfield of computer science aimed at specifying and making computer systems that
mimic human intelligence or express rational behaviour, in the sense that the task would
require intelligence if executed by a human” (T.Bratteige 2018)
This definition was made in 2018 by Tone Bratteige and Guri Verne both belonging to the
section for digitalization at the institute for informatics at UiO.

Definition from Merriam-webster dictionary:

“the capability of a machine to imitate intelligent human behavior” (Merriam-Webster 2020)
This is one of the current definitions of Artificial Intelligence in the renowned
Merriam-Webster dictionary.

Definition from Cambridige dictionary:

“the study of how to produce machines that have some of the qualities that the human mind
has, such as the ability to understand language, recognize pictures, solve problems, and
learn”
This definition is from the University of Cambridge one of the world's oldest and most
respected universities

My definition of Artificial intelligence:

“A computer system with the ability to extract information from different sources and process
it in order to complete some form of task”
My definition of AI is based on that an AI is a system that can take in information from either
the physical or digital realm and process it as needed in order to perform that task that is
expected from it.

Microsoft and how it presents AI

IN5480 - Individual assignment 3
snorreod
I have looked at how Microsoft presents AI on their page. They try to present it as a positive
tool and base for platforms that will change the world for the better. They do this by listing
examples of AI being on the frontline of helping humanitarian organizations. Microsoft also
tries to reassure the people who have fears about some of the implications of AI´s with
claims that they are developing it responsibly with a focus of making lives easier and more
effective by relying on AI as a supplement.

Westworld and how it portrays human interaction with AI

I will describe the interactions between robots and humans in the hit HBO show Westworld.
The base concept of the show is that human guests visit an amusement park inhabited by
extremely life-like robots. Here the humans are free to do whatever they want, a lot of them
choose to experience the parks with robots on scripted adventures where they can choose
to either be a hero or villain. But there are others who sees the park as an opportunity to
vent out a lot of things they wouldnt be able to do in the real world, like for example brutally
raping and murdering these very life-like robots.

1.2 Robots and AI systems

The origin of the Robot

The word robot comes from the czech word robota meaning slave. The first use of it in the
context of a human built obedient autonomos machine was in the czech play R.U.R from
1920.

Two definitions of “robot” along with mine

The Cambridge dictionary´s definition of the word Robot:
“a machine controlled by a computer that is used to perform jobs automatically” (Cambridge
2020)
Cambridge definition of the word Robot can be said to be quite vague, but since there is no
consensus of what a robot actually is, it is a good starting point in that it defines the most
agreed on basic concept of a robot.

Britannica dictionary´s definition of the word Robot:

“Robot, any automatically operated machine that replaces human effort, though it may not
resemble human beings in appearance or perform functions in a humanlike manner. By
extension, robotics is the engineering discipline dealing with the design, construction, and
operation of robots.” (Britannica 2020)
This definition also focuses on the basic concept that a robot performs a task but also
includes some tidbits in saying that a robot both can and can not resemble a human in both
appearance and how it accomplishes tasks.

My definition of Robot:

1/11
IN5480 - Individual assignment 3
snorreod
“A machine that can perform at least one task. A Robot often has a human-like element in
appearance or how it accomplishes a task that makes it different from what humans normally
would just call a machine. “
I think this definition of robots highlights there is a sort of mental difference between what we
would just call a machine and what we would call a robot, and that it often lies in that the
robot completes its task in a more advanced human-like manner.

The relation between AI and Robots

I think AIs and Robots are very closely related and that the main difference might be that a
robot is more at times the physical manifestation of an artificial intelligence. Even though it
doesn't have to be given that it exists robots without any form of artificial intelligence. TBut
given that it is a robot with artificial intelligence, manifestation can take a lot of forms like for
example that of a smartspeaker that speaks back to you like a human or a boston dynamics
robot that resembles how a human actually looks and moves.

The Paro therapeutic sea

The Paro therapeutic seal is used with a lot of different groups that are in need of a
companion that can both relax them and make them feel less alone. The way the robot is
used, is that the patient treats it like an actual living seal so they pet it and hug it.

1.3 Universal Design and AI systems

Universal Design
Definition of Universal design by universaldesign.ie:
“Universal Design is the design and composition of an environment so that it can be
accessed, understood and used to the greatest extent possible by all people regardless of
their age, size, ability or disability” (Universal Design 2020)
I think this a good definition based on what I have learned about universal design and its
focus on designing stuff that can be used by everyone regardless.

The potential of AI with respect to humans

Depending on how AI develops it can be used to improve upon human attributes like
perception, movement and cognitions. An example could be a blind man getting a device
with sensors that could read his surroundings, perhaps even better than an able-bodied
person, and then perfectly guide him where he needs to go.
It is not unreasonable to think that AI also could be used to make prosthetic limbs like arms
and legs that worked perfectly in conjunction with the rest of the body.

2/11
IN5480 - Individual assignment 3
snorreod

The potential of AI inclusion and exclusion

AI has the potential to both be an including and excluding force. Given that the AI is not
developed in conjunction with different parts of society and the world at large it might end up
not having the necessary dataset to be able to service all sorts of different people. Thereby
giving it biases to only work properly with the people that are similar to the ones that
developed it. An example of this is an AI that has problems recognizing people with different
skin color or accents then what it is used to.

Human and machine understanding

In both WCAG 2.1 and Human-AI guidelines the terms understand and understanding is
used. Both of them use them in the context of presenting the user with information in a way
that he can actually process it and base his actions on it.
In that context a machine can “understand” given the right dataset.

1.4 Guideline for Human-AI interaction

Microsoft guideline with a different example

Make clear why the system did what it did.

It’s important that the system informs the user why it did what it did. Because in the event
that the user uncovers a bias of some sort in how the AI solves a task. The AI can get fixed.

Comparing HCI and Human-AI interaction

I decided to compare microsoft's AI guidelines with Nielsen and Molich's 10 User Interface
Design Guidelines.
The HCI
- Visibility of system status
- “Make clear why the system did what it did” matches this in that it focuses on
showing where the system is and how it got there.
- Match between system and the real world
- Matches the “Match relevant social norms” in that the system should be
expected to be able to communicate with a user in a given context.
- User control and freedom
- Matches “Support efficient correction” in that it is easy to fix when system or
user has made an error.
- Consistency and standards
- This matches the point “Update and adapt cautiously”. Both the hci guideline
and the ai guideline is warning us about making changes that can result in the
user having troubling get back into the system as it has changed beyond what
can be expected from him.
- Error prevention

3/11
IN5480 - Individual assignment 3
snorreod
- This matches “show contextually relevant information” in that we should limit
the chances for the user to make a mistake.
- Recognition rather than recall
- Sort of matches “Show contextually relevant information” in that we need to
display task-related information in way that it is easy to recognize it and not
having to recall where it is.
- Flexibility and efficiency of use
- Connected to “Learn form user behavior” in that the system should be able to
learn from the users behavior so that it can improve upon its efficiency in
doing the tasks it knows the user frequently does.
- Aesthetic and minimalist design
- Matches “Show contextually relevant information” the main focus should be
to present the users with the information they need for their current tasks, not
unnecessary information that might distract them from finding what they need.
- Help users recognize, diagnose and recover from errors
- Is connected to “Make clear why the system did what it did” in that it should
be clear to the user why the system behaved the way it did.
- Help and documentation
- This can be connected to the initial guidelines with “Make clear what the
system can do” and “Make clear how well the system can do what it can do”
in that program easily tells the user what it can do and how well the system
can do it.

References:
https://www.interaction-design.org/literature/article/user-interface-design-guidelines-10-rules-
of-thumb​ Guideline by Jakob Nielsen and Rolf Molich

https://www.microsoft.com/en-us/ai?activetab=pivot1%3aprimaryr6

T.Bratteig, G.Verne. Does AI make PD obsolete? Exploring challenges from

Artificial Intelligence to Participatory Design. Association for Computing Machinery (August
2018)

Grudin, Jonathan. AI and HCI: Two Fields Divided by a Common Focus. AI magazine 30,
no 4 (September 18, 2009)

Merriam-Webster.com Dictionary, s.v. “artificial intelligence,” accessed September 10, 2020,

https://www.merriam-webster.com/dictionary/artificial%20intelligence.

http://universaldesign.ie/What-is-Universal-Design/​ accessed September 10, 2020

https://dictionary.cambridge.org/dictionary/english/​ accessed September 10,2020

https://www.britannica.com/technology/robot-technology​ accessed September 10, 2020

4/11
IN5480 - Individual assignment 3
snorreod

Appendix 1:
Based on the feedback I received from the last Individual Assignment I have tried to
structure this assignment to have a bit more logical and consistent structure,
hopefully making it a bit easier to read.

Individual Assignment 2
1 AI-infused systems

1.1 Identify and describe key characteristics of AI-infused systems

There are a lot of key characteristics of an AI-Infused system. One of the most
important is the ability to learn. Either from huge data sets or from the users

5/11
IN5480 - Individual assignment 3
snorreod
behavior. Based upon what it has learned it should also try to improve, meaning
making less errors as time goes by and as it gets the feedback and information it
needs to properly fulfill its tasks.
But errors are in fact themselfs a characteristic of the system as it is nearly
impossible to make a system without any sort of error. These errors can be
manifested in the system in a lot of different ways either technical like the system
crashing or social like the system not functioning properly for different kinds of
people then the ones that built the system.
Another characteristic seen from the user perspective is that AI-Infused systems
often function like a black-box. Meaning that you do not know what the systems
actually does with your input and why it delivers the output it does.

1.2 Discuss the implications of these characteristics for the example

system, in particular how users are affected by these characteristics.
I have picked the AI-Infused word-processor Google Docs. The AI-Infused
functionality in the program is its autocorrect and word suggestions functionality.
Concerning the key characteristics mentioned above I think the learning one is
probably the most relevant one. This because of the gigantic scope of data and
users Google has at its disposal to properly train the Google Docs AI. Making the AI
able to distinguish between for example Norwegian and English in the same
sentence and only correcting based upon if the actual words were written wrong not
the mix of languages.

2 Human-AI interaction design

2.1 Main take-aways from Amershi et al. (2019) and Kocielnik et al.
(2019)

The main take-away from Amershi et al (2019) concerning interaction design is that
with the rapidly increasing amount of solutions implementing different types of AI the
interaction design community needs to try to keep up in order to shape the solutions
in a way that fits the users. Amerishi et al contribution to this is the 18 guidelines that
they have developed to help clarify some problems they see and make the solutions
more accountable for the users.

Kocielnik et al (2019) main take-away concerning interaction design is the

importance of managing users expectations and some techniques that can help in
doing so. The reason for this is because of the negative effects, expectations not
built on the actual reality of the system, can have on the users satisfaction and
willingness to actually use the system.

6/11
IN5480 - Individual assignment 3
snorreod

2.2 Discussing design guidelines

For the first design guideline I have chosen g4: “Show contextually relevant
information”. An example of this is in Google Docs is how the system helps users
finish commonly used phrases based on the rest of your text by semi-filling them out
for the user, so that the user can choose whether they want to take the system's
advice or write it in their own way.

For the second design guideline I have chosen g8: “Support efficient dismissal”. The
system incorporates this by letting the user tell the system to ignore false-positives
that the system's autocorrect function has picked up, which can happen to for
example some names probably because of social biases. Another instance where it
incorporates this guideline is the ease you can correct it when it actually tries to
autocorrect something that it shoulnt have and you can by either returning or
pressing the backspace key dismiss change.

3 Chatbots / conversational user interfaces

3.1 Discussing key challenges in the design of chatbots

There are a lot of challenges connected to developing chatbots . One challenge is to
manage expectations like stated in Kocielnik et al (2019). For example if the user
comes into the context expecting the chatbot to be able to answer his every question
he might be disappointed to find out that it is in fact limited to the more common
questions.

Therefore it is important that the chabot try to adhere to some of the AI design
guidelines given in Amershi et al. The most importants is probably “Make clear what
the system can do” so that the user don't have waste a lot of time on a solution that
will never and if the chatbot is setup more like a filter before the users come to an
actual human being make it possible to “Support efficient dismissal” so that the user
can easily dismiss the chatbot and go right to talking to an actual human. The
chatbot should also as much as possible follow “Remember recent interactions' ' so
that the user does not need to repeat already stated information that the chatbot
should already be aware of, as this can cause a lot of frustration for the user.

Another challenge when it comes to chatbot is shaping the language of the chatbot
so that it is understandable and actually helps the user in the task they are trying to
accomplish (given that it is a typical customer support chatbot). One way of making
sure of this can perhaps be to base it upon successful interactions actual human
beings in customer support has had with people, but even that might just be

7/11
IN5480 - Individual assignment 3
snorreod
propagating something that could have been explained in much simpler and
understandable terms.

3.2 Guidelines G1 and G2 in chatbots

Guideline G1 is as I mentioned earlier of major importance when trying to make the
user understand what the chatbot is capable of. Guideline G2 is also of importance
as the system should not necessarily expect that it has actually solved the users
problem, since it can make mistakes. And that is something users should be aware
of and possibly have the opportunity to correct.

References:
Amershi, S., Weld, D., Vorvoreanu, M., Fourney, A., Nushi, B., Collisson, P., ... &
Teevan, J. (2019). Guidelines for human-AI interaction. In Proceedings of the 2019
CHI Conference on Human Factors in Computing Systems (paper no. 3). ACM.
(​https://www.microsoft.com/en-us/research/uploads/prod/2019/01/Guidelines-for-Hu
man-AI-Interaction-camera-ready.pdf​)

Kocielnik, R., Amershi, S., & Bennett, P. N. (2019). Will You Accept an Imperfect
AI?: Exploring Designs for Adjusting End-user Expectations of AI Systems. In
Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems
(paper no. 411). ACM.
(​https://www.microsoft.com/en-us/research/uploads/prod/2019/01/chi19_kocielnik_et
_al.pdf​)

Liao, Q. V., Gruen, D., & Miller, S. (2020, April). Questioning the AI: Informing
Design Practices for Explainable AI User Experiences. In Proceedings of the 2020
CHI Conference on Human Factors in Computing Systems (paper no. 463). ACM.
(​https://dl.acm.org/doi/abs/10.1145/3313831.3376590​)

Yang, Q., Steinfeld, A., Rosé, C., & Zimmerman, J. (2020, April). Re-examining
Whether, Why, and How Human-AI Interaction Is Uniquely Difficult to Design. In
Proceedings of the 2020 chi conference on human factors in computing systems
(Paper no. 164). (​https://dl.acm.org/doi/abs/10.1145/3313831.3376301​)

8/11
IN5480 - Individual assignment 3
snorreod

Individual Assignment 3:
Industrial robot and Nano robot
The industrial robot is explained in Phillips et al (2016) as a tool to multiply the
physical capabilities of a human. An example of this sort of robot would be the typical
robot arm you often see in factories doing repetitive tasks that it can do faster and
more reliably than humans due to the fact that it doesn't have to concern itself with
stamina like a human would. This task would probably be put somewhere in the
middle of the right corner of the two-dimensional framework shown in Shneiderman
(2020).This due to the nature that these robots perform their tasks separately from
humans at different stages of the product's assembly line. Meaning that the humans
working around it probably don't have that much control over it, other than being able
to stop it in case of errors or emergencies.
.
The Nano robot explained in Phillips et al (2016) is explained as a tool to multiply
cognitive capabilities. The robot in the article, the Black Hornet, does this by allowing
it’s operator to program waypoints for it to take, allowing it’s operator to watch from
the video feed transmitting from the robot instead of having to for example peek
around a corner. If i had to place it in the two-dimensional framework shown in
Shneiderman (2020) this sort of robot would probably find itself in the upper right
quadrant somewhere. I based this on what I have understood from the way it
functions. In that humans have full control over the path of the robot, but that a lot of
the flying and sensor-reading are more automated. Meaning that the operators no
longer have any need for physically peeking around the corner, in order to get some
general situational awareness.

Adjusting the level of autonomy

Industrial robot
Increasing the automation of the industrial robot would probably mean making it take
on more tasks in producing whatever product it is making. It would also probably
mean making it able to a larger degree to learn from what it is doing and optimize
itself without needing human intervention.This would probably lead to faster and
cheaper production, but would probably also lead to it replacing even more
human-tasks at it’s factory. An effect of this could be that workers get a more fulfilling
workday as repetitive and boring tasks are done by the robots, but it can also lead to
unemployment as the workers talent is no longer needed to produce a product.
There can also be some negative aspects to the AI itself deciding how to improve
itself, as there could be biases in how it measures improved performance like
neglecting a negative effect in the new procedure, for example that the industrial
robot wears out at a much faster phase then before.

9/11
IN5480 - Individual assignment 3
snorreod

Decreasing it’s autonomy would lead to it needing more human input in order to
produce the product leading to slower and more expensive production, but would
probably be seen as a good thing from highly specialized workers at the factory as
their service would be needed more.

Nano robot
Increasing it’s level of autonomy even further would probably go at the expense of
human control as the robot would probably decide the route and the places it would
monitor. These could both be a positive thing as it could perhaps find things the
operator couldn't sense in the same way as the robot, but could also make it worse
as the user would either not be confident enough to actually use the system or
maybe too confident leading to the operator trusting it blindly. These are both
scenarios that could lead to dangerous situations given the environment this robot is
meant for, where the operator's trust in the system can have life and death
consequences.

Decreasing the automation of the nano bot would probably lead to it being harder to
operate as quality of life features like it being able to hover on it’s own would
probably disappear. This would probably reduce its usefulness in the field as you
would need somebody to completely dedicate themself to controlling the robot during
operation, instead of it being able to do certain tasks by itself.

Explainability of the robots

To judge the robots' explainability as explained in Hagras (2018) we need to
examine how understandable the robot actions are for the user and how
understandable they need to be.

For the industrial robot I would hope that the current level of explainability is
sufficient enough. With sufficient I mean so that the person charged with controlling
the robot performance is able to understand how it performs its task. Meaning that
he should be able to look over how it is programmed to do the task, seeing things
like how it maneuvers and how much force it applies. So that he can see that it is set
up right to perform its tasks. And so that he is able to see that it is not performing its
tasks based on any biases that could have a negative impact on both the safety of
the manufacturing process and the quality of the finished product.

10/11
IN5480 - Individual assignment 3
snorreod
I would think that given that this is military grade equipment, where the users trust in
the equipment may be key in saving lives. Would mean that the system vendor has
had a big focus on achieving an acceptable level of explainability in order for the
military to be able audit it and for them to be able to properly train an operator in how
to use the system.

References:
Hagras, H., Toward Human-Understandable, Explainable AI, Computer, 51, 9, 2018,
28- 36 ​https://ieeexplore.ieee.org/document/8481251

Shneiderman, B., Human-Centered Artificial Intelligence: Reliable, Safe &

Trustworthy, arXiv.org (February 23, 2020). ​https://arxiv.org/abs/2002.04087v1
(Extract from forthcoming book by the same title)

Phillips, Elizabeth & Schaefer, Kristin & Billings, Deborah & Jentsch, Florian &
Hancock, Peter. (2015). Human-Animal Teams as an Analog for Future
Human-Robot Teams: Influencing Design and Fostering Trust. Journal of
Human-Robot Interaction. 5. 100. 10.5898/JHRI.5.1.Phillips.

11/11