Preliminaries
Ethics...

• is no spoil-sport
• helps exploring early on the values that are often implicit
• can help a responsible innovation process
• helps us find making explicit which kind of future we want

Differentiating AI

• weak AI
• machine learning
• deep learning
• general AI
• consciousness
• ...
  is true AI even achievable? / How to know when we achieved it?

Impact of progressive automation on work
Changing working conditions; the kinds of work humans will do...

Asimov’s Laws of Robotics

• A robot may not injure a human being or, through inaction, allow a human being to come to harm.
• A robot must obey the orders given it by human beings except where such orders would conflict with the First Law.
• A robot must protect its own existence as long as such protection does not conflict with the First or Second Laws.

UK’s Principles of Robotics

• Robots should not be designed as weapons, except for national security reasons.
• Robots should be designed and operated to comply with existing law, including privacy.
• Robots are products: as with other products, they should be designed to be safe and secure.
• Robots are manufactured artefacts: the illusion of emotions and intent should not be used to exploit vulnerable users.
• It should be possible to find out who is responsible for any robot.

Human-Centered AI
Designing an AI system that gives human excitement, more enjoyment, more interest and empowers them to do the things they want to do.
AI in service to humans rather humans being in service of AI is the key difference
John Shawe-Taylor

II
To treat all persons fairly and with respect, to not engage in harassment or discrimination, and to avoid injuring others.

III
To strive to ensure this code is upheld by colleagues and co-workers.

2.
improve the understanding by individuals and society of the capabilities and societal implications of conventional and emerging technologies, including intelligent systems;

3.

• avoid real or perceived conflicts of interest whenever possible
• disclose conflicts of itnerest if they exist

1.

• hold paramount the safety, health, and welfare of the public
• comply with ethical design & sustainable development practices
• protect privacy of others
• disclose factors that might endanger the public/environment

Reflection upon wider impact of work supporting the public good

Ethical decisions occur during the planning for and conducting of research, not just when applying results to new technology

tendency of unsimilar nodes to attach to each other

a large number of nodes having a small number of links and a few of them having many

minorities in
underrepresented/disadvantaged
position

tendency of similar nodes to attach to each other

a large number of nodes with a large number of links

number of edges that are incident to the vertex/node of a graph

established nodes rather have
a new connection than new nodes

long term neighbouring nodes
delete bias between each other

some bias stays while neighbouring nodes converge monotonically

Allowed Privacy-Violation
leads to intimacy and personal relationship
(love, friendship, cooperation etc.)

Justices

• distributive
• relational
• prodedural
• interactional
• retributive
• transactional
• restorative/transitional
• epistemic
• social
• intergenerational
• global
• climate
• gender
• environmental
  ...
• Algorithmic justice
  //any context in which different viewpoints can come into conflict

site of justice
to which entities/agents do justice claims primarily apply?

• governments
• institutions
• companies
• groups
• individuals

principles of justice
according to which criteria do we decide about (re-)distribution

• desert/effort (individual responsibility
• maximal benefit
• need (sufficiency/priority)
• contract
• equality
• sustainability
• authoritarian
• ...

"The Good"
the positive demands - to be secured

• everyone receives his/her due
• "to each according to his need, from each according to his ability" (Marx)
  
  

→
+
all receive their due

-
flawed reality

"The Bad"
the negative demand - to be avoided

• reduce unchosen disadvantage/hardship
• avoid rewarding the irresponsible
  
  

→
+/-
unjust hardship/advantage
clustering of advantages/disadvantages
enhancing general injustices

//access to school for everyone gives access but says nothing about the quality of school in lower socioeconomic areas

"The Good"
the positive demands - to be secured

• secure interactions among all 'on a footing of equality'
  

  • inter-individual
  • institutional

• secure that all have enough to interact as equals
  
  

→
utopian reality: ability to connect and interact as equals across the world

Stuctural Injustice
when social processes

• put large categories of persons under a systemic threat of domination or deprivation
• enable others to dominate/have wide range of opportunities for developing and exercising their capabilities

Marginalization

• discimination
• acces denied
• silent voices
• silenced additionally through algorithmic bias and epistemic injustice

"Specific" Concern
distinctive intersection between social injustice & technology

• digital sphere mirrors/perpetuates/increases the existing relational injustices IRL
• implicit bias/algorithmic bias; epistemic injustice

Motivation

• making profits VS provision o essential services/catering to basic human needs (interaction, information, ...)
  
  

→
problem with distribution

One-Sidedness
not actively listening to people negatively affected (Silicon Valley techno-utopianism VS. experiences of diverse people underrepresented in Silicon Valley → abusive speech against Black Women)

→ problem of relation
(fairness/distribution alone does not capture it)

accountability, legal oversight, top-down regulation
→ digital sphere == public sphere where basic goods are distributed

Categorize

• what we regard as normative
• what people think is normative in one culture
  
  

→ Information consistent with stereotypes will be

• given more attention
• rehearsed more often
• → remembered better

positive Stereotypes
// Benevolent Sexism/Racism/...

→ both forms, positive & negative stereotypes legitimize discrimination against the group in question

Stereotypes
a generalization of a group in which certain traits are assigned to virtually all members regardless of actual variation between members

microinvalidations
ignoring/excluding/devaluing thoughts/feelings/perception of others

microinsults
complimenting people that seem foreign for their good language skills

Social Identity
part of persons self-concept that is based on his/her identification with the belonging to a certain group or other social affiliation

Ethnocentrism
cultural or ethnic bias:
the belief that one's own culture is superior to others // lives the correct way of living

Outgroup Homogeneity
the perception that individuals in the out-group are more similar to each other (homogenous)than members of the in-group are

Blaming the victim
the tendency to blame individuals (make dispositional attributions) for their victimization, typically motivated to see the world as a fair place

• the stronger the belief in a fair world, the more the tendency to blame the individuals
  → often happens for people who rarely experienced discrimination

Male Robot

participants were:

1. ascribe more agency related traits
2. seeing the female robot as more capable for stereotypical male tasks
3. more likely to choose a math task to work on together with the robot than a verbal task

Race
refers to physical differences that groups and cultures consider significant

→ no biological/scientific basis for race

Ethnicity
shared cultural characteristics such as language, ancestry, practices and beliefs

Robot Shooter Bias

1. participants were quicker to shoot an armed Black agent than an armed White agent / simultaneously faster to refrain from shooting an unarmed White than an unarmed Black agent
   → regardless of whether human or robot
   
   
   

2. short response window with accuracy instead of latency
   similar to experiment 1: participants were faster to not shoot at an unarmed White than an unarmed Black agent - No shooter bias on error rates

Female Robot

participants were:

1. ascribe more communal related traits
2. seeing the female robot as more capable for stereotypical female tasks
3. equally likely to choose a math task or a verbal task to work on together with the robot

Biased Components

• sensors are biased for white skin color
• NLP is biased for male voices
  NLP is biased against dialects, slang, childrens voice, older adult's voice

Biased Adaption
robot might unintentionally favor one group member over another
based on

• components
• performance
• beter training data for specific users

multi-identity robots/systems

• one distributed system for multiple robots
• one robot with multiple identities
  
  
  

robots that fit the personalized expectations of the user/user groups
→ would need to use stereotypes to tailor behavior -> fosters stereotypes

In-Group designers
having more diverse// more specified in-group designers within the process of design

Designing against established structural Discrimination

1. identify groups
2. ask in which way they could be in an adavntage/disadvantage

ideas

• ability to choose how you personally want to use the service
• //focus on specific customer groups (?)
  

  • common practice