Please enable JavaScript.

Coggle requires JavaScript to display documents.

ETHICS OF AI - Coggle Diagram

- - - - meaning of ethical concepts
      - ontology and epistemology
    - - Determining a moral course of action - practical
    - - what someone who can be held accountable is obligated to do in a specific situation
  - - - Immediate
        
        Here and now
        
        Security
        
        Deepfakes
        
        privacy and transparency in AI systems
      - Medium term
        
        impact on military, medical, and judicial use
        
        Impact on education systems
      - Longer term
        
        ethical goals of developing and implementing AI in society
        
        Do we want to prioritise social harmony, what would this expense?
    - - unlike a squirrel AI will at some point likely be capable of abstract thought with it's own autonomy - this is very different from a squirrel
      - When genuine abstractly thinking AI develop they will be able to think orders of magnitude beyond what we can
      - ISAAC ASIMOV (1942)\
        
        3 LAWS from I robot
        
        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.
      - Currently we don't think of the machines we are using as moral agents as they do not possess the capability of autonomy and abstract thought
    - - case study
        
        why weren't the data practitioners looking how they could not represent this demographics
        
        took the working model and didn't cross check it because they were excited about it
- - - - Generally big moral values
        
        happiness, freedom, wellbeing
        
        Can be used to explain the goodness in other things - Aristotle
    - - Money
        
        Can be used to pay for medical care for someone
        
        Only valuable because it can be use for other things
        
        Value determined by purchasing power
    - - moral values are primary
      - economic, aesthetic, epistemic (knowledge related) values also play a role
        
        what is an example of an epistemic value?
        
        empirical grounding
  - - - Statistical regularity
        
        Computer scientists wear black t-shirts a lot
      - Social norm
        
        what people in a group believe to be appropriate actions in that group
      - Moral norms
        
        prescriptive or proscriptive rules with obligatory force beyond social or statistical expectations
        
        DONT USE AI FOR BEHAVIOUR MANIPULATION?
        
        May not always be enforced by law
      - LEGAL NORMS
        
        x is the law
        
        this does not make x a moral principle
        
        laws in North Korea
        
        Not always moral
        
        THE US JUDICIAL SYSTEM AND RICH PERPETRATORS OF OPPRESSION
  - - - The data doesn't have any moral implications at base
        
        FACTS ALONE DON'T SOLVE ETHICAL PROBLEMS - ethical discussion is needed
      - But what does the data mean?
        
        A bias in the medical data Alex was talking about led to the death of children because it was biased against kids with rare diseases
    - - describe how the world should be - ought to be
      - machine learning systems should be transparent
      - the machine learning system is a black box systems = descriptive
- - - - Do good
      - Create beneficial AI
      - Thought experiment
        
        Shifting from reactive to preventative medicine
        
        Benefits = sickness prevention is an effective way to improve the health of individuals
        
        Better impact estimation and healthcare planning
        
        Reduction of healthcare costs
        
        Problems
        
        Privacy and security concerns
        
        Require a lot of data and hospitals are notorious for poor data governance
        
        Where is the line of non-acceptable intrustion?
        
        How is consent to be given?
        
        What about high risk patients who cannot give consent?
        
        1 more item...
        
        who is responsible?
        
        If the city has information about potential health risk of a patient and does not act upon this data, is the city guilty of negligence?
        
        Utilitarianism
        
        Saying yes the city should seek an ethically and legally acceptable way to use these methods, there are too many benefits vs the risks
        
        A version of consequentialism
        
        the consequences of any action are the only standards of right and wrong
        
        Morally right = actions that produce the greatest balance of benefits over harm for everyone affected
        
        Where could this be a problem?
        
        How do you weigh it?
        
        3 more items...
        
        Utility tool kit
        
        what are the various actions we could perform?
        
        what is our estimation of the benefits and harm that would result from each action
        
        choose the action that provides greatest benefits after costs have been measured and taken into account
        
        1 more item...
        
        Why is it hard to compare risks and benefits?
        
        risks and benefits are valued by value commitments, subjective and diverse preferences, practical circumstances, and personal and cultural factors
        
        You could choose a general benefit model or a biased model, both have ethical problems
        
        Harms and benefits are not static
        
        Marginal diminishes in a way that can be difficult to predict
        
        Specific harm or benefit may have different utility value in different cirucmstances
        
        1 more item...
        
        Real world situations are typically so complex that it is difficult to foresee or compare all the risks and benefits in advance
        
        aDVERSERIAL PATCHES TO CONFUSE AUTONOMOUS MILITARY AI
        
        Utilitarianism over-simplifies individual rights
        
        What if someone wants to express their right to not know about being in a high risk category, for example they suspect they might have a neuriological condition that degrades but don't want it to burden their present life and how they live it
        
        Nozick's utility monster
        
        what is utility really?
        
        a numeric measure of an underlying good
        
        1 more item...
        
        pleasure or wellbeing - hedonists would say these are the same
    - - Do no harm
        
        Can moral problems things that can be fixed by technical systems?
        
        Prison system as a case study says no
        
        Who controls, governs, and is affected by AI?
        
        Evgeny Morozov
        
        TECH SOLUTIONISM
        
        problems created by tech can be solved by more tech
        
        Where does this run amok?
        
        Oversimplification of deep ethical questions
        
        Pro-life vs pro-choice is a case study of how polarising these problems can be and how sticky they are to untangle
      - What are the negative consequences and risk of AI?
        
        Ethical dilemmas and moral questions: the train killing a baby or 5 adults, which track do you choose? - Driverless cars?
        
        Physical and social harm
      - Minimisation of discrimination
        
        A flattering poem about Joe Biden but not about Donald Trump
      - Minimisation of privacy risks?
        
        The dad of a pregnant teen being sent pregnancy discounts from a supermarket that had collected his daughter's data and knew before he did
      - Often discussed threats
        
        intentional misuse
        
        malicious hacking
        
        technical measures
        
        risk management and risks
- - - - Whilst it is the algorithm's fault
      - It is not morally responsibl
      - Automated vs autonomous systems
        
        Autonomous - learns and adapts to dynamic environments. Evolves as it's environment changes
        
        Automated - follows predefined instructions
      - What is accountability?
        
        Acknowledgement of responsibility for actions, decisions, and products
        
        Responsibility can be legal or moral
        
        AI Ethics - Dimensions of accountability
        
        Which individuals are accountable for the impact of algorithms or AI? Who is responsible for what effect within the overall socio-technical system?
        
        A feature of he societal system that develops, produces, and uses AI
        
        A feature of the AI system itself
      - Who should be blamed and for what?
        
        moral agents are morally responsible for their actions
        
        Responsible for their own acts and omissions
        
        If letting someone die = not-helping = omission
        
        Omissions are not morally equal to actions
        
        Moral responsibility requires: moral autonomy and the ability to evaluate the consequences of actions
        
        Moral autonomy is an agent's capacity to impose the moral code on oneself in a self-governed way
        
        Autonomy requires: capacity to rule oneself, authenticity of desires that cause action, sufficient cognitive skills
        
        Moral responsibility and Immanuel Kant
        
        Actions are based on our own will to utilise a moral law to guide our decisions
        
        We owe ourselves moral respect in virtue of our autonomy
        
        We owe a similar respect to all other persons in virtue of their capacity
        
        The problem off individuating responsibilities
        
        Why is it difficult to set criteria on responsibility?
        
        Quality of responsibility differ
        
        An actor is responsible for a specific action
        
        But may be hard to keep humans in the loop when AI starts decision making
        
        The quality of responsibility is dependent on the stakeholder
        
        If you choose not to disclose something to ensure that a violent murderer definitely goes to jail I would not say this is immoral
        
        Technology can take the persuasive form
        
        Seatbelt beeper
        
        Only if the action is done voluntarily if done intentionally and is free from controlling influences
        
        Persuasive tech should comply with the requirement of voluntariness to guarantee autonomy.
        
        Voluntariness presupposes a sufficient understanding of the use of specific technology
        
        2 more items...