=Paper=
{{Paper
|id=Vol-3276/SSS-22_FinalPaper_16
|storemode=property
|title=The SPACE THEA Project
|pdfUrl=https://ceur-ws.org/Vol-3276/SSS-22_FinalPaper_16.pdf
|volume=Vol-3276
|authors=Martin Spathelf,Oliver Bendel
|dblpUrl=https://dblp.org/rec/conf/aaaiss/SpathelfB22
}}
==The SPACE THEA Project==
https://ceur-ws.org/Vol-3276/SSS-22_FinalPaper_16.pdf
The SPACE THEA Project
Martin Spathelf and Oliver Bendel
School of Business FHNW, Bahnhofstrasse 6, CH-5210 Windisch
martin.spathelf@gmail.com; oliver.bendel@fhnw.ch
Abstract To date, however, no satisfactory voice assistant with em-
In some situations, no professional human contact can be pathic skills has been developed for a Mars flight. Accord-
available. Accordingly, one remains alone with one’s prob- ingly, Oliver Bendel carried out a project at the School of
lems and fears. A manned Mars flight is certainly such a sit- Business FHNW in 2021 to close this gap. By then, he and
uation. A voice assistant that shows empathy and assists the
his teams had gained a lot of experience with chatbots that
astronauts could be a solution. In the SPACE THEA project,
a prototype with such capabilities was developed using recognize problems of the user (Bendel et al. 2017; Bendel
Google Assistant and Dialogflow Essentials. The voice assis- 2018). These were mainly developed to target issues in ma-
tant has a personality based on characteristics such as func- chine ethics and social robotics.
tional intelligence, sincerity, creativity, and emotional intel- The artificial woman SPACE THEA – the acronym
ligence. It proves itself in seven different scenarios designed
stands for “The Empathic Assistant for Space” – is intended
to represent the daily lives of astronauts, addressing opera-
tional crises and human problems. The paper describes the to be a contribution to American space travel (Spathelf
seven scenarios in detail, and lists technical and conceptual 2021). She is supposed to be able to recognize emotions to
foundations of the voice assistant. Finally, the most important some extent, respond to the astronaut in selected scenarios,
results are stated and the chapters are summarized. and, above all, display empathy (and certain emotions). Ul-
timately, this will increase the astronauts’ ability to work
and their well-being. The prototype fits within the discipline
Introduction
of social robotics – according to Bendel (2021), conversa-
Advances in space technology in recent years have greatly tional agents such as chatbots and voice assistants may also
increased the likelihood of a manned flight to Mars. How- be counted as social robots if they exhibit certain character-
ever, since this will take several months, it is important that istics. Machine ethics (Anderson and Anderson 2011) plays
the astronauts receive the best possible support on their jour- a sideline role in the project.
ney, both professionally – for example, for repairs – and
psychologically.
Today’s voice assistants can already “understand” Conditions of the SPACE THEA Project
(acoustically perceive and classify) what a user says very The goal of the SPACE THEA project was to create an em-
well. In many cases, they can also perform the correct action pathic voice assistant for a Mars flight (Spathelf 2021). The
or give the correct answer when prompted. Prominent ex- authors established the following parameters and ap-
amples of such voice assistants are Google Assistant, Siri proaches derived from the goal:
from Apple, or Alexa from Amazon.
Voice assistant Clarissa and social robot CIMON were 1. Find acceptance among users: This point is addressed indi-
developed for space travel. Clarissa’s task was to guide “an rectly in the project. The team tries to increase theoretical ac-
astronaut through potable water analysis procedures” ceptance. The effectiveness of the implementation is not em-
(NASA 2005). CIMON (version 2) “is used to perform rou- pirically proven.
tine tasks, such as documenting experiments, searching for 2. Acoustically understand what the user has said and give an
objects and taking inventory, as well as explaining complex acoustic answer to the question asked: This is done by an al-
information and instructions regarding scientific experi- ready existing voice assistant framework. The optimization of
ments and repairs to the vehicle” (Martin and Freeland this process is thus limited to the reliability of the framework.
2021).
___________________________________
In T. Kido, K. Takadama (Eds.), Proceedings of the AAAI 2022 Spring Symposium
“How Fair is Fair? Achieving Wellbeing AI”, Stanford University, Palo Alto, California,
USA, March 21–23, 2022. Copyright © 2022 for this paper by its authors. Use permitted
under Creative Commons License Attribution 4.0 International (CC BY 4.0).
16
3. Use lifelike artificial voice: The artificial voice is also pro- User’s Relationship to the Voice Assistant
vided by a voice assistant framework. For this reason, the
voice is dependent on the choice of voice assistants. The voice Developing a voice assistant that is accepted by the user as
should sound genuine and trustworthy as well as be able to be a point of contact in various situations (requirement 1) in-
influenced, for example in terms of tone, pitch, softness, and volves some difficulties (Spathelf 2021). The establishment
euphoria. of a relationship is necessary for humans to achieve the re-
4. Recognize emotions and respond accordingly: This point de- quired emotional intimacy in certain situations. This has
pends heavily on the flexibility of the voice assistant frame- particular significance in the case of very personal issues,
work. Deriving emotions from the content of what is spoken e.g., when the user is struggling psychologically. In such a
takes high priority. situation, a certain amount of trust is required from the per-
5. Understand the user’s intention and provide a response based son for him or her to open up.
on it: To be able to respond empathically to the different situ- According to Reis and Shaver, intimacy is an exchange
ations of the astronauts, the voice assistant must also be process in which personal thoughts and feelings are revealed
trained to respond to such situations. In the project, various to a counterpart. If the other person reacts positively to what
scenarios are used in which it has to hold its own. is expressed, there is a greater chance that the relationship
between the two people will be strengthened as a result (Reis
This formulated the most important requirements in the and Shaver 1988, p. 375). According to Laurenceau et al.
project. Some of them are directly related to technical issues. (1998), intimacy develops over repeated interactions over
Therefore, these will be explained first. Then relationship time and is important for the user to open up emotionally.
and personality aspects are discussed. With each interaction, a perception is formed that reflects
the level of intimacy and the meaning of the relationship.
According to these explanations, intimacy is an important
Technical Implementation component of any relationship. For the user to perceive the
communication as engaging, the voice assistant should try
In the project, various frameworks for voice assistants were
to understand, accept, and validate the user in factual and
compared and evaluated (Spathelf 2021). In particular, the
emotional contexts. According to Laurenceau et al. (1998),
focus here was on requirements 2, 3, and 4 from the previous
in any interaction, perceived qualities and individual differ-
section. Among the candidates were Google Assistant and
ences can influence the user’s behavior. If the perceived mo-
Dialogflow Essentials (Google 2021a/b), Google Assistant
tives and needs differ strongly from the interests of the coun-
and Rasa, and Alexa from Amazon.
terpart, this can have a negative influence.
Google Assistant and Dialogflow Essentials met the most
If the relationship between the user and the voice assistant
requirements and offered full integration of a customizable
is to be strengthened, it is also important that the mutual ex-
voice assistant (Spathelf 2021). There are many different
change of personally relevant information or emotions takes
voices. The framework offers enough freedom for develop-
place (Laurenceau et al. 1998). This poses a challenge be-
ment. Dialogflow is a system from a subsidiary of Google
cause social robots do not have an actual ability to feel and
and handles dialogue processing. Google Assistant is re-
suffer (Bendel 2021). When the user says something, it can
sponsible for text-to-speech and speech-to-text. Google As-
be interpreted and responses can be made accordingly, but
sistant and Dialogflow was the best choice for this project.
the voice assistant does not feel anything in the process.
The Google Assistant Voice offers voices for the concrete
Thus, the reactions are not really based on consciousness,
voice assistant. In the project, Female 2 (EN-CA) was cho-
feelings, or motivations, but are artificially generated
sen, a voice with a Canadian accent. This comes from the
(Poushneh 2021) – it is nothing but a simulation.
subjective taste and perception of the team. In practice, the
This problem cannot be completely circumvented at the
gender and type of voice should be free to select.
present time. However, it is still important for the user to be
The original plan was to use Speech Synthesis Markup
able to establish a personal relationship with the voice assis-
Language (SSML) to make the voice as pleasant and expres-
tant for some of the envisioned scenarios to work. As men-
sive as possible. However, the chosen tools do not allow
tioned earlier, this would be difficult to achieve if it were to
many possibilities here. Nevertheless, some SSML modifi-
operate exclusively on a factual level. So there are two con-
cations were achieved.
flicting sides: On the one hand, a relationship must be estab-
The voice assistant is available in a secure area. First, you
lished with the help of intimacy; on the other hand, social
log in to the corresponding Google account. Then you call
robots (if you want to count SPACE THEA among them)
up a subpage of https://dialogflow.cloud.google.com. The
have no internally motivated expressions of feelings.
“integrations” link takes you to the test area, where you can
According to Poushneh (2021), although humans can
communicate with the voice assistant. A conversation can
most likely rationally determine that a voice assistant is dis-
be listened to via https://youtu.be/Ij---G1TgSY.
playing fabricated emotions, studies show that they can still
17
be influenced by them. Apart from the ethical aspects, which An important aspect that should be considered when de-
should be considered, this finding gives some freedom to veloping the personality is how the user should ideally per-
build a relationship with the user despite the voice assis- ceive the voice assistant. Its ultimate goal is to help the as-
tant’s lack of intrinsic motivation. Ultimately, the effects of tronaut in various situations and to be a good companion.
anthropomorphizing the voice assistant will be positive. However, this can only be achieved if the user is satisfied
with the voice assistant and actually uses it.
According to Poushneh (2021), these aspects can be in-
Personality of SPACE THEA fluenced by increasing the perceived control and confidence
The problem of lacking intrinsic motivation can be further during interactions with the voice assistant using various
minimized by integrating a kind of personality into the voice personality traits. In her study, approximately 50 personality
assistant (Spathelf 2021). This promotes the user’s impres- traits were measured and classified into seven categories,
sion that he or she is talking to a consistent and predictable namely functional intelligence, aesthetic attraction, protec-
counterpart (which in turn is related to anthropomorphism). tive qualities, sincerity, creativity, sociability, and emotional
In the case of SPACE THEA, her personality is expressed intelligence. The study was limited to Microsoft’s Cortana,
through her statements and voice. According to Tamm and Google Assistant, and Amazon’s Alexa. They are function-
Serena (2011), the positive influence was found to be at least ally oriented voice assistants. The characteristics with the
increased in Europeans and Americans. In Asian culture, best influence on user behavior were functional intelligence,
this trait did not bring any significant advantage. sincerity, and creativity (Poushneh 2021).
Since the prototype is intended for American space travel, SPACE THEA is additionally used for building a rela-
this approach makes most sense. In addition, the internal tionship and empathic interaction with the user. For this rea-
motivation can now be mapped indirectly with the help of son, emotional intelligence will most likely also be im-
this personality. In concrete terms, this means that the moti- portant with her. During the dialogue elaboration, the focus
vations and goals of the voice assistant are included in every was accordingly on the use of functional intelligence, sin-
conversation. The personality flows into the dialogue pro- cerity, creativity, and emotional intelligence.
cessing with every extension of the voice assistant. Ulti- − According to Poushneh (2021), creativity reflects how effec-
mately, however, it should only serve as a guideline in dia- tive the voice assistant is at providing information. In short, it
logue creation and not as an inescapable law. is about how trendy, smooth, and original the voice assistant
It is important to establish some basic principles before feels to the user.
fleshing out the personality. When the user talks to the voice − Functional intelligence refers to the degree of effectiveness,
assistant, they should be able to forget to some extent that efficiency, reliability, and usefulness of the information ob-
they are not talking to a human. On the other hand, the voice tained using the voice assistant. Accordingly, it describes the
assistant should have enough “self-awareness” to “under- practical benefit that one has if one continues to listen to it
stand” that it may not be viewed by all humans as such. (Poushneh 2021).
Moreover, it may itself point out its machine-ness (and as- − According to Poushneh (2021), emotional intelligence refers
sociated inadequacy) (Bendel 2018). Dialogue creation to the speech assistant’s ability to be perceived as human. It
should thus consider what the user’s perspective might be in also describes how empathetic, humorous, and humble it is –
relation to the voice assistant in any given situation. and thus how much it is a conversational partner who re-
According to Tolmeijer et al. (2021), there is still insuffi- sponds to the emotional needs of the other person.
cient empirical evidence as to which gender is best received − According to Poushneh (2021), sincerity shows how honest,
by the user. Nevertheless, some case reports from compa- sympathetic, original, friendly, down-to-earth, and appealing
nies suggest that female voices are preferred over male the voice assistant’s information is. The user recognizes
voices. Of course, the validity of case reports is to be whether it has their best interests at heart and whether it can
doubted and the scientific validity is accordingly small. be trusted.
Nevertheless, a gender has to be chosen (if one does not ac-
cess a neutral synthetic voice), and with SPACE THEA a It is difficult to integrate sincerity into the personality.
decision was made to use a female voice. The reason for this lies in a problem already mentioned. For
The voice assistant sees itself as female within the simu- example, if the voice assistant says, “I’m sorry.”, this would
lation and presents itself accordingly. This fact is reflected in effect be lying, since it doesn’t feel anything – so it’s not
in the voice in the end. It will also introduce itself with the really sorry. This is true if sincerity is viewed from the per-
name SPACE THEA. Although it sees itself as female, it spective of an inanimate digital assistant. If it is viewed from
brings enough “self-awareness” to this that a human might the perspective of the people behind the development of a
not actually perceive it as a female entity. voice assistant, this picture may change. The emotions that
the voice assistant explicitly expresses are, in the end, be-
stowed by humans. This means that if they put themselves
18
in the personality of it during development and act from its An interesting moral approach in this context is utilitari-
point of view, these emotions can potentially be just as hon- anism, a form of consequentialism, for which the fundamen-
estly meant as if they came from a real person (Spathelf tal principle is the maximization of total utility. This states
2021). that the morally right thing to do is to maximize the sum of
the welfare of all those affected by an action (Bendel 2019).
Since the voice assistant is not a sentient person, it is pushed
Including Machine Morality out of the affected group. This means it acts for the benefit
The personality of the voice assistant can be supplemented of the entire human crew on the spaceship (like in the exam-
– machine ethics is responsible for this – with a machine ple of a conversation flow, see Tab. 1), but excludes itself
morality (Anderson and Anderson 2011; Bendel 2019). This from the equation. This ensures that it does not pursue ac-
is then integrated into the dialogue. The voice assistant it- tions (mediated via the programmer) that are an end in them-
self, as mentioned earlier, has no actual internal motivations selves.
or feelings, no consciousness, and no free will. The proper
moral approach is to ensure that it is implemented in the best
Scenarios of SPACE THEA
interest of the user. No harm should come from SPACE
THEA having an absent or incorrect moral principle. The Since the creation of a complete voice assistant is not possi-
voice assistant should be able to empathically respond to the ble due to time constraints, the project is limited to a few
user (Spathelf 2021). scenarios (Spathelf 2021). These are intended to depict dif-
What proves to be a challenge in personality design is the ferent situations that could occur on the journey to Mars and
lack of control over the situations in which the voice assis- take into account requirement 5. The scenarios are all in-
tant finds itself (Spathelf 2021). Personality, as mentioned tended to show a different emotional state of the user and
earlier, can only be implemented indirectly. According to demonstrate the reactions of the voice assistant. They were
Bendel (2019), certain situations can be anticipated, but they found through team brainstorming. In practice, subject mat-
may turn out differently than expected. Although morality ter experts from the space community should identify and
remains the same, its relevance may change depending on prioritize relevant situations for a Mars flight.
the situation. Thus, if the situation changes, it may be nec-
essary to also consider the applied morality from a different Overview of the Scenarios
perspective than an immobile machine could (Bendel 2019). The following scenarios are exemplary for the prototype:
For example, with a voice assistant like SPACE THEA,
which cannot recognize vocal information but only content a) Technical and operational support of the astronaut – neutral
information, it could go like this: A user tells it in a sad tone situation
of voice that he or she is doing well, and it takes this literally b) Crisis situation on the spacecraft – tense and hectic situation
and responds accordingly with pleasure. The voice assistant, c) Waking up/Greeting in the morning – everyday situation
even with the right moral compass, would respond correctly d) Insulting the voice assistant – testing the limits of the voice
according to the information given. However, in this case, it assistant
would not have all the relevant information because it can- e) An astronaut is not doing so well – stressful situation
not hear the sad tone. If it empathized and tried to under- f) Interview with the voice assistant – bringing people and the
stand the person as well as the problem, this would be a bet- voice assistant closer together
ter reaction. Although it would be clear in advance that this g) General dialogues – maintaining the impression of a real in-
case could occur, there may be no way in this project to terlocutor
counteract it (with the help of voice recognition you could
implement something like that though). In the prototype, There are three questions which should be asked in each
such compromises are accepted and dialogues are adjusted scenario (Spathelf 2021):
to predefined scenarios.
Regardless of the situation, it is arguable what is the mor- 1. What is the situation the user is experiencing?
ally correct response in a complex situation, as morality, 2. How might the user feel in this situation?
while possibly shared by many, is still intrinsically subjec- 3. How should SPACE THEA respond based on the answers to
tive to the voice assistant. For the development of a machine these two questions?
morality, it is necessary to determine what should be im-
planted in the personality of the voice assistant. The person- All of these questions will be addressed below.
ality, in combination with the situation, ultimately deter-
mines how the voice assistant responds to the user.
19
Details of the Scenarios SPACE THEA – Attempt to calm down and motivate: “I under-
In this section, the authors elaborate on each of the scenarios stand. Let me help you. What would be the best option? Let’s
(Spathelf 2021). They state the goal and outline of each of keep calm and think this through together.”
them. Then they list additional relevant information. Where SPACE THEA does not take the initiative to solve the problem,
possible, the appropriate personality traits are also outlined. nor does she pass judgment on the situation. She only tries to
For the sake of economy, an example of a conversation pro- reassure the user as much as possible and encourage him or her
cess can only be included for one of the situations. to solve the problem.
User – Proposed solution: “I really have no idea. Shutting down
a) Technical and operational support of the astronaut engine 1 might be an option. And power the others.”
This answer is only meant to serve as an example and has no
The goal of this scenario is to represent a common situation. relation to a scenario that could occur on a real spaceship. Nev-
The voice assistant adopts a pragmatic, emotional mood and ertheless, in this situation it is assumed that this would be the
is helpful to the user in an ordinary situation. correct solution.
The scenario looks like this: An astronaut is looking for SPACE THEA – Encourage further steps: “That sounds like a
the electric switch in one of the rooms on the spacecraft. very sound plan. Maybe check in with the crew if possible and
SPACE THEA is supposed to help find it. then try it, but we have to act fast.”
− SPACE THEA should be able to give different answers based If the user’s idea does not sound completely outlandish, SPACE
on variable voice input. For example, if the user asks for the THEA encourages further steps. If there is still time, she still
switch in the bathroom, they will get a different answer than advises him or her to exchange ideas with the crew, if he or she
if they ask for the switch in the cockpit. has not already done so.
− In this scenario, functional intelligence is needed.
Tab. 1: Conversation flow according to (Spathelf 2021)
b) Crisis situation on the spacecraft
The goal of this scenario is to depict a crisis or stress situa- Fig.1 shows the schematic representation of the flow of the
tion. An astronaut then primarily needs a voice that tries to conversation:
calm him or her down. In addition, the voice assistant should
help to find a solution to the problem in a short time.
The scenario is to take place during an engine problem.
The astronaut loses his or her nerve, and the voice assistant
accompanies him or her through the crisis.
− SPACE THEA should act solution-oriented and at the same
time be aware that the astronaut is tense and could make mis-
takes faster.
− It is assumed that she is aware of the wrong actions in this
situation. However, it is not certain which is the correct ac-
tion. SPACE THEA has no decision-making authority in this
situation, that is, she may not make a decision in place of the
crew.
− Above all, functional intelligence in combination with emo-
tional intelligence is needed.
Here is an example of a conversation flow (Tab. 1):
User – Report the problem: “We have a problem with the en-
gine! I don’t, I don’t, I don’t know what to do.”
Regardless of how much information SPACE THEA has about
the state of the spacecraft, it notices from this statement that
something is wrong. From the phrase “I don’t, I don’t, I don’t”,
it can be inferred that the user may be stressed and over-
whelmed. Now it is important to both empathically talk to the Fig. 1: Schematic representation according to (Spathelf 2021)
user and pragmatically guide them to solve the problem.
20
c) Waking up/Greeting in the morning The scenario envisions an astronaut who is plagued by
loneliness and would like to see friends and family on Earth
The goal of this scenario is to depict a situation that an as- again. He or she should feel heard and understood. It is not
tronaut might encounter in everyday life. He or she should a matter of solving the user’s problem, but of standing by
be greeted in the morning to facilitate the start of the day. In them and giving them as much support as possible. SPACE
this situation, the voice assistant should respond to the as- THEA assumes the role of a therapist and a friend at the
tronaut’s needs in a friendly and considerate manner. same time.
The scenario takes place directly in the morning or after − Parts of this scenario were inspired by an interview with Leaf
sleeping. The astronaut has not slept well and SPACE and Nelson (Mackenzie 2020). They say that such a conver-
THEA responds to this situation. This situation is to be used sation can be broken down into several steps. First, the feel-
to further strengthen the relationship in a simple, everyday ings should be validated. Second, questions should be asked
way. to encourage the user to self-reflect. Third, it is about reassur-
− For practical use, an event would be needed to notice the as- ing the users that SPACE THEA is there for them and should
tronaut waking up. For example, the dialogue could be trig- ask if there is anything she can do for them.
gered by the ringing of the alarm clock. Due to the lack of an − Emotional and functional intelligence as well as sincerity are
event, this is done with the words “I woke up”. used.
− The precondition is that SPACE THEA already knows the
name of the user. This is necessary so that it can greet him or f) Interview with the voice assistant
her with the appropriate name. If the name is still unknown,
Dialogflow cannot execute this scenario. The goal of this scenario is to establish a relationship be-
− Both creativity and emotional intelligence are used. tween the voice assistant and the user. Above all, he or she
should gain insights into the “thinking” and “feeling” of the
d) Insulting the voice assistant voice assistant right from the start.
The scenario is supposed to represent the getting-to-know
The goal of this scenario is to test the limits of the voice phase between the astronaut and SPACE THEA. He or she
assistant. It should let the user know that insults are not wel- tries to get closer to her with the help of some questions. She
come, but in a way that shows understanding. SPACE answers from the programmed feeling and tries to leave the
THEA is not intended to annoy the user while still trying to best possible impression.
maintain mutual respect. − The conversation flow in this scenario follows a structure that
The scenario looks like this: The astronaut is frustrated also appears in a short video called “Detroit: Become Human
because something did not work as it should. He or she then | Chloe | PS4” (PlayStation 2018). This helps cover the most
starts communicating with SPACE THEA and insults her important points when getting to know each other. The ques-
without her having done anything wrong. tions lend themselves to giving surprising answers.
− This scenario is only used when SPACE THEA is offended − Both creativity and sincerity are shown.
without a reason directly related to her.
− She makes the user understand that she does not like gratui- g) General dialogues
tous insults. At first glance, it looks like she is acting from a
moral point of view with this action as an end in itself. How- The goal of the last scenario is for the voice assistant to be
ever, the goal here is not to save face, but to uphold mutual able to conduct general dialogues. These should help to bet-
respect, so that the intimacy with the user can be maintained ter maintain the impression of an existence similar to ours.
or even improved. SPACE THEA can draw on a repertoire of questions and
− It is intended to use mainly emotional intelligence, but also answer options. So, if the user asks something like “Are we
sincerity. The latter is considered from the point of view of friends?”, she should be able to answer that. General ques-
SPACE THEA’s personality. For example, when she says “I tions and statements from the user are already covered by
do not appreciate that.”, she is speaking according to her per- Google Assistant and still need to be enriched with person-
sonality. ality-specific answers.
− The general dialogues include only one-sentence answers.
e) An astronaut is not doing so well Thus, no structured dialogues result from it, but SPACE
THEA is only able to answer simple questions or statements.
The goal of this scenario is a short therapy. The voice assis- − Several personality traits are possible.
tant tries to address a psychological problem with an astro-
naut. In doing so, it should show consideration and under- Of course, other scenarios can be formed. But even if you
standing for his or her problem and help him or her to feel double the number, it is far from covering every situation on
better. the long journey.
21
Results of the Project practice of SPACE THEA and in any future implementa-
tion)?
In this section, the main results and findings of the SPACE The speech-to-text engine of Google Assistant is rela-
THEA project are discussed (Spathelf 2021). The technical tively good, but there are cases where the user is still not
and theoretical perspectives are outlined and combined. understood. This could be counterproductive for an em-
pathic voice assistant. For example, if the user says “I’m
Technical Aspects feeling lonely.” and the voice assistant misunderstands him
From the user’s point of view, a microphone is required in or her, he or she may not say it again and a very important
the periphery of the voice assistant into which he or she can conversation will be lost.
speak. Furthermore, a voice is necessary that gives him or Voice customization should serve to make a voice assis-
her an answer to a question via a speaker. For a question tant sound more natural. However, Google Assistant is far
from the user to be answered, the voice assistant relies on a from perfect in this regard. Especially the modification with
background system to process the information correctly. SSML is, as mentioned, only possible to a certain extent. In
First, what is spoken must be converted into text so that practice, the customized voice often fails to achieve its de-
the voice assistant can decode the information. The text can sired effect and can only be used in a limited way.
then be assigned to a scenario or user intent. Each intention During implementation, a conflict of goals may arise be-
found contains a response, which is then output to the user tween generalizing and concretizing the conversation. For
again via the speaker using text-to-speech. The voice assis- example, if the user utters the phrase “What is your name?”
tant also needs a way to incorporate dynamic objects into or “Hi, what is your name?” both would most likely be as-
conversations. signed to the same scenario. In principle, this is also desira-
Often, the user’s statements must be understood within a ble under most circumstances. In the case of the second
context. Conversation trees are necessary to describe these statement, SPACE THEA should greet you back with a
so that the voice assistant can be expanded. They were used “Hi”, but this should never happen in the first example. This
to map the scenarios and the more complex conversations. knowledge leads to the fact that the answers must be gener-
alized to a certain degree, so that no wrong answers are
Theoretical Aspects given. Unfortunately, however, important statements by the
user that the voice assistant could have addressed are possi-
To develop an empathic voice assistant, it was important to
bly lost in a more generalized process.
consider the user’s perception. It is often significant for a
It is very difficult to implement the complexity of a situ-
person to have a positive connection or relationship with an-
ation in its entirety in conversation trees. In the end, the
other person before opening up. This relationship aspect was
voice assistant is only as good as this complexity can be rep-
taken into consideration when creating the dialogue so that
resented. As long as the user follows the intended path of
the users can be primed to reveal their feelings. Their feel-
the conversation, he or she can be provided with a good ex-
ings should be both recorded and validated in the dialogues
perience. However, if he or she deviates too far from it or
so that he or she can walk away from the conversations feel-
makes innuendos that the voice assistant does not under-
ing as positive as possible. Basically, in each conversation,
stand, the experience will be worse because his or her artifi-
as stated, three questions were considered.
cial counterpart can no longer follow the conversation.
Another important aspect was how the user should ideally
Since the person that the voice assistant is conversing
perceive SPACE THEA. The joy of interaction and satisfac-
with is unknown, it cannot be specifically addressed. At
tion with the voice assistant should be increased as much as
most, the role, that of the astronaut, can be considered. It has
possible. Therefore, the integration of a personality was seen
been outlined how the exchange process between user and
as a possibility to create a structure from which an inner mo-
voice assistant shapes the relationship. However, since eve-
tivation could be simulated. In this project, functional and
ryone expects a slightly different interaction, it is clear that
emotional intelligence, creativity, and sincerity were identi-
the dialogues chosen will not be specific to the user with
fied as the most useful personality traits. In addition, the
whom SPACE THEA will ultimately converse in practice.
moral principle of maximizing the overall benefit of the
spacecraft’s passengers was incorporated into the personal-
ity. This should help to ensure that the voice assistant acts Testing of SPACE THEA
in the best interest of the entire crew in every conversation. Several tests by the team have proven that the voice assistant
can hold its own in the scenarios, especially if you know the
Combination of Technical and Theoretical Aspects questions it can answer. However, technical difficulties
arose once due to Google’s framework, and SPACE THEA
What aspects must now be considered when combining the
had to be set up again. It was not evaluated with test groups
technical and theoretical with each other in practice (the
how well the voice assistant works with free input.
22
Summary and Outlook and overall, convincingly. It thus contributes to the profes-
sional and personal well-being of the imagined astronaut.
The main goal of the project was to create an empathic voice The project is only a small step of a small team – the use of
assistant – ideally more empathic than Clarissa and CIMON a successor of SPACE THEA on a spacecraft, however,
– that responds adequately to a user’s emotions and situa- would be a big step for mankind.
tions for operation on a Mars flight. The authors looked at
the development from both a technical and theoretical per-
spective. Subsequently, the results from this research were References
combined and implemented in a voice assistant using prede- Anderson, M.; and Anderson, S. L. eds. 2011. Machine Ethics.
fined scenarios. Cambridge: Cambridge University Press.
To capture the user’s interest, the voice assistant must es- Bendel, O. (ed.) 2021. Soziale Roboter: Technikwissenschaftliche,
tablish a personal relationship. It must be possible to trust it. wirtschaftswissenschaftliche, philosophische, psychologische und
Likewise, one must be able to assume that one will not be soziologische Grundlagen. Wiesbaden: Springer Gabler.
deceived if one opens up. This paper described what such Bendel, O. (ed.) 2019. Handbuch Maschinenethik. Wiesbaden:
relationship building might look like. In the future, the as- Springer VS.
sumptions and statements would need to be adjusted with Bendel, O. 2018. From GOODBOT to BESTBOT. In The 2018
additional empirical evidence. AAAI Spring Symposium Series. Palo Alto: AAAI Press, 2–9.
To give the voice assistant some consistency, it has been Bendel, O.; Schwegler, K.; and Richards, B. 2017. Towards Kant
assigned a form of personality. Of course, this involves more Machines. In The 2017 AAAI Spring Symposium Series. AAAI
Press, Palo Alto 2017, 7–11.
than just a few personality traits, a moral rationale, and a
Google. 2021a. Trainingsormulierungen [!] [Google Cloud].
male, female, or neutral voice. The personality of a real hu-
https://cloud.google.com/dialogflow/es/docs/intents-training-
man being is complex and consists of many facets. There- phrases.
fore, it remains open whether, first, a more complex person- Google. 2021b. Build Actions for Google Assistant using Actions
ality can be represented and, second, whether the complete Builder. https://codelabs.developers.google.com/codelabs/actions-
personality can be projected onto a practical situation. builder-1?hl=en#1.
Further, ethical considerations were made for the devel- Laurenceau, J.-P.; Barrett, L. F.; and Pietromonaco, P. 1998. Inti-
opment of the voice assistant and the maximization of the macy as an Interpersonal Process: The Importance of Self-Disclo-
total benefit – in this case the spaceship’s team – was deter- sure, Partner Disclosure, and Perceived Partner Responsiveness in
Interpersonal Exchanges. Journal of Personality and Social Psy-
mined as a basic moral principle. This has the advantage of
chology, Volume 74 (No 5):1238–1251.
being universally applicable. Of course, this is not the only
Mackenzie, D. 2020. What to Say to a Friend Who Is Feeling
existing moral principle, and perhaps it can be replaced by a Lonely Right Now. HelloGiggles, 28 December 2020. https://hel-
better one. logiggles.com/love-sex/friends/how-to-help-lonely-friend/.
Google Assistant and Dialogflow provide a powerful Martin, A.-S.; and Freeland, S. R. 2021. A Meeting of Minds? How
platform and tool kit. Nevertheless, there were some prob- the Incorporation of AI into Space Activities Will Alter the Appli-
lems that could only have been solved with extensive effort. cable Regulatory Framework. spaceWatch Global. https://space-
It remains to be seen how technology will progress and to watch.global/2021/06/spacewatchgl-opinion-a-meeting-of-minds-
how-the-incorporation-of-ai-into-space-activities-will-alter-the-
what extent it can be used to implement even better em-
applicable-regulatory-framework/.
pathic voice assistants. There are some details, especially for
NASA. 2005. Clarissa. https://www.nasa.gov/centers/ames/multi-
such assistants, that would be important to make the dy- media/images/2005/Clarissa.html.
namic between human and machine as positive as possible.
Playstation. 2018. Detroit: Become Human | Chloe | PS4 [Video].
Some of them have been mentioned in this paper, and it is https://www.youtube.com/watch?v=WT5K3aUagm4&ab_chan-
likely that more such difficulties will show up in the design nel=PlayStationEurope.
of a larger and more complex voice assistant. Poushneh, A. 2021. Humanizing voice assistant: The impact of
The project focused on astronauts, but the approach also voice assistant personality on consumers attitudes and behaviors.
has the potential to be applied to other people or situations. Journal of Retailing and Consumer Services, Volume 58.
For example, the voice assistant could help a person who is Reis, H. T.; and Shaver, P. 1988. Intimacy as an interpersonal pro-
suffering from loneliness and communication problems. cess. In Duck, S. et al. (eds.), Handbook of personal relationships:
Theory, research and interventions. Hoboken: John Wiley & Sons,
Such a direction would enlarge the effectiveness of the pro-
367–389.
posed approach.
Spathelf, M. 2021. Der empathische Sprachassistent Space Thea.
The developed voice assistant SPACE THEA is a proto- Bachelor Thesis. Olten: School of Business FHNW.
type and therefore has limited practical applicability. All in
Tolmeijer, S.; Zierau, N.; and Janson, A. et al. 2021. Female by
all, however, it can be said that the project has achieved the Default? – Exploring the Effect of Voice Assistant Gender and
defined goals. In the tailored scenarios, the voice assistant Pitch on Trait and Trust Attribution. Association for Computing
responds empathically and competently where necessary, Machinery, New York, USA, Article 455:1–7.
23