HRI Ethics Nao pill (MAARS) American remote ATLAS reminder - - PDF document

2/21/19 1

HRI Ethics

Modular Advanced Armed Robotic System (MAARS) – American remote control gunbot

ATLAS

Nao pill reminder robot

Ethics and Robotics

ATLAS

Nao pill reminder system

(MAARS) – American remote control gunbot

Robear transfer robot

What is Ethics?

• The study of right and wrong ac-on
• Descrip1ve (or compara1ve) ethics:
• How do people think about these issues?
• What are the laws? What are the cultural norms?
• Prescrip1ve ethics:
• How should one act?
• What should the laws and norms be?
• Applied ethics
• Iden1fying the correct course of ac1on for real problems

Neither absolute nor rela1ve Extremely near-term ques1ons!

Normative Ethics (“Ethics”)

• Consequen1alism/u1litarianism (consequence based)
• Maximize the overall good
• ‘The greatest good for the greatest number. ’
• Deontology (duty based)
• It is moral to follow rules, and intent is what maRers.
• ‘Because it’s the law.’
• Contractualism (contract based)
• Do exactly those things where a ra1onal agent would

want to live in a world where everyone did those things.

• ‘Golden Rule’

Robot Vision

• You’re building a vision system:

Cybernetics

2/21/19 2 Categories of Harm

• Robots can physically harm people.
• How?
• Negligently
• Deliberately
• ..?
• Robots can impinge on people’s self-determina1on.
• Robots can fail to act as expected, causing…
• Physical harm
• Emo1onal distress
• Disappointment

“vic1mless” ethics is rarely “applied” This is a huge category – but is it the most likely?

Meta-Questions

• Ques1ons we will not answer today:
• What do “right” and “wrong” mean?
• Who gets to decide what’s right and wrong?
• How do/should those decisions be made?
• What should we do about things that are wrong?
• We’ll use commonly understood ideas of wrong:
• It’s wrong to harm people
• Physically, emo1onally, financially…
• It’s wrong to discriminate against people
• It’s wrong to steal from people
• It’s wrong to invade people’s privacy
• It’s wrong to be unfair to people

“Without extenuating circumstances,” and understanding that sometimes there’s no “right” alternative

Categories of Harm

• Robots can change our defini1on of “humanity.”
• Robots can have rights that are impinged upon.
• Robots can discriminate.
• Robots can do environmental damage.
• Robots can increase the have/have-not gap.
• For people
• For na1ons

What Robots, and How?

• Military Robots
• Caretaker robots
• Elderly
• Children
• Assis1ve robots
• Explora1on robots
• Factory robots
• Surgical robots
• Cyberne1cs
• Search-and-rescue
• Automated cars
• Bipeds and

Quadrupeds

• Why is ATLAS

Scary?

Big Questions

• Can computers “hurt” people?
• What about robots?
• Can a machine be

“unfair” (discriminatory)?

• Why do we, as robo6cists,

care?

• What are some immediate

issues, right now? Sure. Even more so. Sort of. There’s a GIGO aspect. Ethics and morals, legal liability

Topics

• Drive discussion with an example: Self-driving cars
• And generalize from there

2/21/19 3 Self-Driving Cars

• Cars can hurt or kill people.
• How many fatali1es is acceptable?
• Is it enough to not cause accidents?
• People cause accidents!
• ~38,000 deaths per year in the U.S.
• Lately it’s been going up
• How many of you text and drive?
• Do cars have to be perfect? Just beRer than humans?

Somewhere in between?

Harder Questions

• What about naked self-driving cars?
• No control mechanisms inside at all
• Should it be legal for a person to drive?
• Even if cars are demonstrably beRer at it?
• Why?
• Because I wanna?
• Because we dislike giving up control?
• Even if you accept the risks, what about my rights?
• Who’s legally liability? ß this is a big question

that will affect the future

The Hardest One

• When an accident is inevitable…
• Should the car occupants get hurt?
• That is, the person who paid for it?
• If it’s not their fault?
• Would you buy a car that could hurt or kill you?
• If it could be avoided by hur1ng or killing someone else?
• Can you buy any other kind? What’s the difference?
• But consider:
• Would you swerve to avoid a kid in the road?
• What about a baby stroller?
• Who should be deciding these things? Uber?

Harm in HRI

• Therapy: Taking it away
• Privacy rights and vulnerable popula1ons
• Physical contact
• Discouraging bonding
• Decep1ve studies
• Sales agents
• Haves and have-nots
• Diversity

Code of Ethics

• Dignity: Respect privacy, frailty, and emo1onal needs
• Design: Be transparent, predictable, and trustworthy;

convey status; allow op1ng out

• Legality: Respect all laws; all decisions reconstructable;

always seek informed consent

• Social: Avoid decep1on; consider emo1onal bonding;

disallow human morphology; avoid –ist behaviors

• Do we like it?

2/21/19 4 Ethical Governor

• A component that intervenes in ethical cases
• Uses rules to provide ac6ons

Rules

• Drawn from exis1ng literature
• …by computer science students
• “More sensi1ve signals [are] hard to detect …

currently, but we can add those more sensi1ve signals to our architecture later by developing the

• technology. “
• Anger-based
• Emo1onal or physical withdrawal based

Rule Structure Evaluation and Conclusions

• Evalua1on
• Reviewed by an OT expert
• Qualita1ve analysis à changes
• Actual user trials planned
• Pros and cons
• PD-specific
• Early stages
• is it scalable? (No. Not at all. Does it maRer?)
• How did they do wrt. Riek?

Overall

• What are our responsibili1es?
• What about cultural context?
• Caregivers vs. pa1ents
• Diversity, again
• Military applica1ons

Disabled, Able, and Super-Able

• Disabled:
• “Having a physical or mental condi1on that

limits movements, senses, or ac1vi1es.”

• Compared to who? I can’t lik 180 kg.
• 20-20 vision is compara1vely new.
• Prosthe1cs: pacemakers, cochlear

implants, bionic limbs, re1nal implants

• Beyond able-bodied…
• For work: TALOS; EKSO
• For informa1on: Google Glass
• For fun?