annesgjo 1 Interaction with AI: individual assignment Module 1 Concepts, definition and history of interaction with AI History of AI The term artificial intelligence, AI, was first used in 1956 by John McCarthy. AI originates in mathematics and engineering and was at first hand focusing on devising better algorithms. In the 1960 and 1970 AI researchers believed that machines would be so intelligent that they could compete with humans on equal terms before the twentieth century. Through the last decades AI have had a changing interest and actuality. Today we can say that AI was way oversold and is not in anywhere near the predictions that were made. (Grudin, 2009) What is AI? “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.” (Cambridge dictionary, 2019). By comparing the human mind to how a computer works seems kind of far off doesn’t it? How the human brain works is in no way related to how the different processes in a computer works. Mike Loukides and Ben Lorica is questioning the definition of intelligence: “Defining artificial intelligence isn’t just difficult; it’s impossible, not the least because we don’t really understand human intelligence. Paradoxically, advances in AI will help more to define what human intelligence isn’t than what artificial intelligence is.” (OReilly, 2016).
15
Embed
Interaction with AI: individual assignment...annesgjo 1 Interaction with AI: individual assignment Module 1 Concepts, definition and history of interaction with AI History of AI The
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
annesgjo
1
Interaction with AI: individual assignment
Module 1 Concepts, definition and history of interaction with AI
History of AI
The term artificial intelligence, AI, was first used in 1956 by John McCarthy.
AI originates in mathematics and engineering and was at first hand
focusing on devising better algorithms. In the 1960 and 1970 AI researchers
believed that machines would be so intelligent that they could compete
with humans on equal terms before the twentieth century. Through the
last decades AI have had a changing interest and actuality. Today we can
say that AI was way oversold and is not in anywhere near the predictions
that were made. (Grudin, 2009)
What is AI?
“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.” (Cambridge dictionary,
2019).
By comparing the human mind to how a computer works seems kind of
far off doesn’t it? How the human brain works is in no way related to how
the different processes in a computer works. Mike Loukides and Ben Lorica
is questioning the definition of intelligence:
“Defining artificial intelligence isn’t just difficult; it’s impossible, not the
least because we don’t really understand human intelligence.
Paradoxically, advances in AI will help more to define what human
intelligence isn’t than what artificial intelligence is.” (OReilly, 2016).
annesgjo
2
By this I want to go back to the earlier definition by John McCarthy:
“It is the science and engineering of making intelligent machines,
especially intelligent computer programs. It is related to the similar task of
using computers to understand human intelligence, but AI does not have
to confine itself to methods that are biologically observable.” (McCarthy,
1998).
A computer is only intelligent in that way that can make a dissection
between what is right and wrong. I will now try to make a definition of my
own:
AI is a way of tricking us into believing that machines can take their own
decisions and, in some way, have the same (or even more) intelligence as
humans.
AI at Google
Google has a vision to bring the benefits of AI to everyone. They present AI
as something that can help and be used by everyone, every day for
anything. Their research can be accessed by everyone and they are a big
believer of sharing their knowledge. The technology should be used to help
people by being social beneficial, fair, accountable and works for everyone.
(Google, 2019).
AI in Black Mirror
Black Mirror portraits a twisted and scary future where human innovation
has gone totally wrong. Every episode is based on a new type of high-tech
innovation which pushes the limits of the society and makes us question
our use of technology. The episodes explore the negative effects of AI and
makes you want to take a reality check. Human interaction with AI is
portrayed as a danger to the society and the series is trying to show us that
annesgjo
3
we have a dark future ahead if we don't change the course we are in right
now.
Robots and AI systems
How robot came about
Robotics is a field in change and what the term robot means have changes
over decades. The Robot Institute of America defined in 1979 a robot as “a
reprogrammable, multifunctional manipulator designed to move
materials, pars tools, or specialized devices through various programmed
motion for the performance of a variety of tasks”. (Thrun, 2004). Robotics
have been categorized into three different categories; industrial robotics,
professional service robotics, and personal service robotics.
What is robots?
NASA defines a robot as “a machine that is built to do a certain job again
and again, or to do work that might be dangerous for humans.“ (NASA,
2019).
LEO Robotics have a more detailed definition: "A robot is an actuated
mechanism programmable in two or more axes with a degree of
autonomy, moving within its environment, to perform intended tasks.
Autonomy in this context means the ability to perform intended tasks
based on current state and sensing, without human intervention." (LEO
Robotics, 2019).
It is hard to define what a robot actually is, but i like how Joseph
Engelberger puts it:
"I can't define a robot, but I know one when I see one."
I want to take a different approach to define a robot:
Robots is a physical machine that can do your job better than you.
annesgjo
4
The relation between AI and robots
However you define a robot it’s not the same as an AI. A robot can be
controlled by an AI system and therefore be an artificial intelligent robot.
But an AI system don’t need to have anything to do with robots, and robots
don’t need AI to function.
AV1
AV1 is a communication robot made by No Isolation and is specifically
designed for children with a long-term illness. The robot will take the place
of the child that can’t attend school, and be their eyes, ears and voice in the
classroom. The owner of the robot can control it through an app on their
phone or tablet. If the child wants to ask a question, they can press a
button on the app and the robot will have a flash on the top of the head.
You can also change the eye expression of the robot.
Universal Design and AI systems
What is universal design?
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.” (The centre
for excellence in universal design, 2019).
Universal Design is something that you should find in every part of the
society, ether it’s to get into a building, take the tram or be able to read the
information on a web page. No exception and maybe even more important
when it comes to AI and robots. It’s about that everyone should feel
included and be able to use a system. If a system is using face recognition
it should be able to recognize people with dark skin on equal terms as
people with light skin.
annesgjo
5
“...simply put, universal design is good design”. (The centre for excellence in
universal design, 2019).
The potential of AI
annesgjo
6
Module 2
Characteristics of AI infused systems
Amershi er al. (2019) describes AI-infused systems as ' systems that have
features harnessing AI capabilities that are directly exposed to the end
user'. Key characteristics of AI-infused systems can be learning, improving,
black box and to be fueled by large data sets. I will furthermore describe
and discuss these characteristics. Amershi et al. (2019) have developed 18
generally applicable design guidelines for human AI-interaction that I will
use to exemplify these characteristics.
Learning
AI-infused systems can be designed to be dynamic; this means that it is
designed for change and will learn and develop while it is being used.
Often this can mean that the AI-infused system is not fully developed
when you start using it but will give you a better user-experience and a
more personalized experience after it learn from your use. In this case G1,
make clear what the system can do, is important to bring into the design
of the AI-infused system. When interacting with a chatbot it will inform you
of what kind of information it can give you. This will make sure that the
user knows what to expect and will not be disappointed. Furthermore, it is
important to make clear how well the system can do what it can do (G2)
so that the user can use the system in the best way. Using the chatbot-
example, it can be beneficial to inform the user how to write so the chatbot
understands in the best way. This will work as a kind of “user manual” so
that the user knows how to interact with the chatbot and “speak” the same
language. The more the system is used the better it will get since it learns
from every time it is used.
Improving
An important aspect of the user experience for AI-infused systems is that
the user can correct if the AI-infused system doesn’t give the best results or
annesgjo
7
show the wrong information. If the system supports efficient correction(G9)
will make it easy for the user to edit, refine or recover. When the user is
shown the advertisement or news that is not suited or wanted you should
easily be able to correct by for example hitting a “show less/don’t show“
button. In other contexts, it would be beneficial for the user to be provided
with multiple options if the AI-infused system is uncertain about the user’s
goals (G10). When you are designing an AI-infused system you should take
uncertainty into consideration because you cannot predict how the user
will use your system.
Black box
Often can AI-systems seem fairly magical and the user may ask themselves
why the system behaved as it did. To give the user this explanation the AI
system should make clear why the system did what it did (G11) by bringing
explainability into the design. This can be done by showing the data the AI
system used or have the option to press “why am I seeing this
advertisement?”.
Fuelled by large data sets
Many AI-infused systems are designed to collect your data and use this to
give you a better user experience. By collecting and learning from your
actions and use over time the AI system can give you a more personalizes
experience. Tesla for example collects data from all of their users to develop
autonomous software.
annesgjo
8
Search for photos on iPhone
One AI-infused system I recently where made aware of is Apple’s
embedded image search function. Here you can easily search for words,
persons or places and the image
recognition will show you all the photos
that fits your search. This way you can
easily find a specific photo using one or
multiple keywords.
To make this possible I believe that the
image recognition is fuelled by large
data sets from everyone that own an
iPhone or iPad, and practice on all of the
photos the users take. My experience is
that the image recognition is fairly
accurate, and it does not take me long
to find the exact photo I am looking for.
The search function will get better each time you use it and learn by using
the image recognition on your photos, and therefor fits the characteristics
of a learning AI-infused system. The system makes it clear what it can do
(G1) by showing the user keywords that
can be searched for and by making
predefined groups of keywords.
annesgjo
9
The system supports efficient correction(G9) and makes it easy for the user
to correct the search function if it makes mistakes. The image recognition
is an improving AI-infused system and
makes the user experience better by
optimizing the recognition of people. The
function “bekreft ekstra bilder” gives the
user an option to go thru photos the image
recognition is uncertain about and in this
way you can teach the AI system how a
person look in different angles etc.
Human AI-interaction design
Summary of Anashi et al. (2019)
In this paper we are introduced to 19 generally applicable design guidelines
for human-AI interaction. Through multiple rounds of evaluation with
people within the field and different popular AI-infused products they have
validated these guidelines. Their goal is to address the challenges for
designing user interface in the context of AI and frame the opportunities
within the field.
annesgjo
10
Summary of Kockielnik et al. (2019)
This paper investigates the expectation of the end-users of AI-infused
systems and how this affect their perception an acceptance of such
systems. Some of the main factors that affect the users are external
information, knowledge and understanding, and firsthand experience. In
their work they used a Scheduling Assistant with an AI system for
automated meeting detection and comparing how the focus on high recall
and high precision would affect the user’s satisfaction of the system. Their
finding indicate that high precision should be the main focus rather than
high recall as the users feel more satisfied with a false negative, and that it
is easier to correct than a false negative.
Chatbots
Some of the key challenges in the design of chatbots and conversational
user interfaces discussed by Luger & Sellen(2016) and Følstad & Brandtzaeg
(2017) are to meet the user’s expectations or mental model of what the AI
system can do. The lack of system feedback made the users uncertain
about how to interact with the system and their knowledge about the
limitations and possibilities to the system. Luger and Sellen (2016) discuss
key factors that limited the user’s interaction; understanding of what the
system can do, what it was doing and how it was doing it. By using G1
“make clear what the system can do” you could limit these challenges,
and the user would be more confident on how to interact with the system.
Users of chatbots and conversational agents states issues with both trust of
the system, and its ability to learn. If G2 “make clear how well the system
can do what it can do” was implemented properly it would be better
received by the users as they would be familiar with its limitations and
open for it to get better by doing mistakes.
annesgjo
11
Module 3
Collaboration and levels of automation
Philips at al. (2016) argues that human-animal teams can serve as an
analog for developing effective human-robot teams. They serve as an
inspiration to look at the robot as part of an interactive team rather than
just a tool. Humans and animals have worked side by side for decades and
are capable of completing a wide variety of work by leveraging the unique
capabilities of each team member. The article explores the future of robots
where they can serve as team members alongside soldiers, working to
achieve common goals and complete team tasks.
By basing human-robot interaction on human-animal interaction we can
benefit of people’s mental model. As people already have knowledge about
how to interact with animals, this would be beneficial in their first meeting
with the robot. Philips at al. (2016) argues that this relationship between
physical forms and mental models has been shown to influence initial
perceptions of robot trustworthiness.
Replace physical capabilities: Big Dog
By looking at how human-animal teams are structured to give the human
physical benefits we can start to identify team structure elements in
human-robot interaction. The robot Big Dog is an example of this where
the robot serves as a pack transporter and give a more efficient locomotion
for transportation. The design is based on a large dog and benefit of not
having wheels in uncertain terrain. Endsley (2011) present a variety of
automation levels from manual control to full automation, and I would
argue that Big Dog have the level of automation described as Automated
decision making. The robot generates recommended options along with
humans, the system selects best and carries it out. In the human-robot
annesgjo
12
interaction the human serves as the direction giver, and the dog will
choose the best way to get from A to B and follow along with the human.
Augment and extend natural human functions: The FRIEND
Service animals such as dogs can help people with impaired vision, and in
many cases help to enhance an individual’s independence and execution
of daily activities that they would not be able to perform safely on their
own. This is a field robots are starting to emerge, and will be at great
means for people to live more independently. The FRIEND robot is an
example of a care providing robot that has the purpose to help people that
suffer from paralysis of other muscular disorders. The user will not be as
depended of other humans and can manage to do simple everyday tasks
that seem rater hard for them. The robot will also reduce stress for the user
by letting them retain a greater degree of self-determination. The FRIEND
robot have an automation level described by Endsley (2011) as action
support, where the robot aims at doing each action as instructed.
annesgjo
13
References Amershi, S., Weld, D., Vorvoreanu, M., Fourney, A., Nushi, B., Collisson, P., ... &
Teevan, J. (2019, April). Guidelines for human-AI interaction. In Proceedings
of the 2019 CHI Conference on Human Factors in Computing Systems (p.
3). ACM.
Bratteteig, Tone. Verne, Guri. (2018), Does AI make PD absolute? Exploring
Challenges from Artificial Intelligence to Participatory Design. Belgium:
Hasselt and Genk. Cambridge dictionary. (2019). Artificial intelligence.