AI & Ethics

Since AI is an emerging market there is a race of all nations of the world to get better than the others in the field of AI –> so Tech is crucially important ethics don´t matter as much

Ethics are difficult to explain. We humans do have a certain intuition on what is right or wrong. But most of the time we can’t explain them proper to, e.g. a machine

Europe:

• high-quality AI talents and research

• struggles to translate research into business applications

• Ethics as an enabler

US: best job market

• talent attractor

• most AI companies/start-ups

• most private equity and venture capital funding

China:

• rapidly catches up

• less strict data protection

• large data bases due to Social Scoring

There are seven key elements of trustworthy AI in Europe

52 experts worked on the skript

In summary AI should be lawful , ethical, robust

Human agency and oversight

Technically robustness and safety

Privacy and data governance

Transperancy

Diveristy, non-discrimination and fairness

Societal and environmental wellbeing

Accountability

Fundamental rights –> there should be mechanisms which feedback if a fundamental human right is insured by an AI

Human agency –> human dicision-making should be informed and atonomous ragarding AI-systems

Human oversight –> there should be governant oversight over AI-systems (HITL, HOTL, HIC)

Avoidence of unfair bias –> discriminating biases should be removed, diverse working teams and diverse data can prevent biases

Accesibility and universal design –> AI-apllications should be user-friendly independent of the users age, culture or gender

Stakeholder participation –> communication with all people of interest of a AI-project

sustainable and environment-friendly AI –> SDGs, help adapt/fighting Climate Change

societal impact –> AI should not endager the phiscal and mental wellbeing of a single individual or communities

society and democracy –> when it comes tp political decisions with the help of AI it should be often times checked

There is a moral machine expermiment of MIT

It is the largest AI study of the world –> over 40 Million people participated from all over the world

The participants should choose in various scenarios and ethical dilemmas, how they would choose themselves

These ethical dilemmas were created based on the difference of peoples age, gender, social status, fitness and the legal/illegal behavior of involved people or the amount of number. There were even scenarios were one should chose between an animal and a human

Most people spared more frequently:

• humans over pets

• mroe over few

• young over old

There were als differences between the different cultural clusters which participated in the survey –> Westeren (young, fit), Eastern (lawful), Southern (higher status, females)

In the survey mostly there were mostly young, male, academic participants –> bias

Since the results of such surveys have the risk of potential biases, or ambigous rules of morality (e.g. we agree that 5 lives or worth more than one…but our own life ist most-precious to us) and the result often conttradicts to our basic law of equality of all human beings (e.g. sparing a young or a female human rather than an old male) it is probably not a good idea

It’s a decision between two moral imperatives, when neither of which is in some way more preferable or accaptable

The Deontological ethics –> only considering if the action in it self is wright or wrong

The Utilitarian ethics –> decide on how many people will die

There should be no distinction between people regarding their age, gender, religion, ehtnicity, etc.

If there is the possibility to spare human lives one should do that –> 5 lives are worth more than one

People who are not involved should be spared –> who defines involvment

scholarships of students

• humans would agree that the distribution of scholarships schould be fair

• but how do we define “fair”? Is it equality (same ressources) or equity (same opportunities)

care-robot

• user begs for painkillers but robot does not get approval of a doctor

• how should the robot decide? User-based or expert-based

househould-robot

• teenager of a family does something viscious, robot witnesses it

• How should the robot react? Should he tell the parents or should he withold the information on purpose

fatal car crash

• A car cannot stop and rushes towards a pedestrian-passing and either kills a young girl or an old lady

• How should the selfdriving car decide? Should it spare young life? But then again, every human life should be equal

self-driving cars - when completed - cause probably less car crashes

but it somehow feels wrong to make a decision based on rules which were made beforehand and not deciding intuively like a human would do

Human: intuively, without reasoning, random

Machine: has to decide which life to spare and whom to kill, resoning, ethical rules

Behavior or decision-making which is commonly accepted in society –> morally right means basically means “good”

If tasks are morraly charged or an ethical dilemma AMAs can’t really make a good decision

It is unclear how one should program ethical code and which ethical approach one should chose, since humans do not have consens on what is right and what is wrong

It is really necessary that AI becomes moral?

What are consequences or moral AI? How can we punish AI if it made wrong decisions

Risk: people who built AMAs are not moral themselves –> selfinterest, bias

Humans are considered moral agents –> they have a intuition of what is wrong or right and feel responsibility. No other non-human thing or animal has this ability

There exists research about Artificial Moral Agents (AMA) –> machine ethicists

• Intuition of good and bad

• Reflecting on potential risks and consequences

• Ability to make the “right” choice

Advantages of AMAs:

• reduction of potential harm towards humans

• higher objectivity than humans

• Potentially higher acceptance since it is idependent

• Avoidance of improper use –> machine cannot do anything what the user tells it to do

• Better understandability of own “moral code”

Auditability –> AI should be able to be audited by anyone

Minimasation and reporting of negative impacts –> AI can help to minimise negative impacts, however if there are certain concerns of the decision-making of an AI-model whistleblowers, NGOs, etc. should be able to critizise

Redress –> there has to be responsibility behind an AI-model. If something goes wrong it should be redressed

Moral principles which determine a persons behavior and decision-making

Sub-discipline of Philosophy which deals with morality

Traceability –> data gathering and decision-making process of an AI-model should be protocolled

Explainability –> decision-making of an AI-model should be understandable by a human being

Communication –> AI-systems should identify themselves as non-human

Resilience to attack and security –> AI-systems should be protected of their vulnarabilities and not be exploited

Fallback plan and general safety –> their should always be a way out in case of problems, e.g. the machine should ask for a human opinion if it is unsure

Accuracy –> decisions, predictions, classifications should be accurate

Privacy and data protection –> privacy should be maintained for the whole process. Input, on-going and output data

Access to data –> there should be data protocols which state who does have access to data and why

Quality and integrity of data –> data should not be biased or wrong