Fair game

The ethics of telco fraud

Ethics Litmus Tests

now online!

www.ethical-litmus.site

Slides

summerscope.github.io/slides/fair-game

Section 1

Friend or fraud?

Lineup

  1. Developing a healthy cynicism about data sources
  2. Mapping knowledge states
  3. Designing for failure first
  4. Upstream prevention vs downstream mitigation

Ethics ops consulting

debias.ai

Three eye glasses with labels, User Harm, Historical Bias, Algorithmic Bias

A balancing act

Finding #1

Developing a healthy cynicism about data sources

Data (interpretation) pride comes before a fall

Deconstructing your proxy

A concept framework

Signal

arrow

infers

Activity

arrows

infers

Persona

A $600m mistake?

""[The dataset] is an index of Australian addresses. The dataset does not contain a list of premises," he said."

Another 300k NBN connections to cost $600m, CEO blames blowout on bad address data

Look at your raw data

Not just at dashboards

Data misinterpretation is easy, and likely under pressure

For example...

Design for cognitive load

Simple doesn't imply you're stupid

Litmus test

On a quick glance, is this easy to misunderstand?

Pointers for data interpretation

  • Using DSL (domain specific language)?
  • Add units
  • Absolute (total count) or relative (%)?
  • Is the data recency clear?

Finding #2

Mapping knowledge states

Beware the swamp of lazy assumptions

Implicit vs Explicit knowledge

Become a knowledge excavator

Naming matters

  • Build a shared vocabulary
  • Build shared mental models
  • Avoid name-space clashing

For example

International revenue sharing fraud
-vs-
Toll fraud

Personas

Person shouting through a bullhorn with lots of chat bubbles coming out

Defining your baseline

Data schemas

Grid with three columns, What did I think, New info, What changed

Changing our language
changes our minds

"We'll wait until we're 100% certain... no make that 99% certain"

Added uncertainty in the space of fraud

You really can't trust what people say

The antidote to suspicion

Setting the intention to be respectful

bUt wHAt iF iT's A BAd AcTor?

So what?

Keep the moralising out of it

Tip

Describe behaviours, not people

Finding #3

Designing for failure first

If you don't ask,
customers won't tell

Woven basket labeled Too Hard

Feedback loops must be

  • Intuitive
  • Contextual
  • Timely

Plan time for...

  • Customer support
  • Product/model improvements
  • Integrating your learnings

Be explicit

If you can't imagine consequences, you're not thinking hard enough

Shot

We believe this system has contributed to making Facebook the safest place on the Internet for people and their information.

Facebook Immune System by Tao Stein, Roger Chen, Karan Mangla

Chaser

"The goal is to protect the graph against all attacks rather than to maximize the accuracy of any one specific classifier. The opportunity cost of refining a model for one attack may be increasing the detection and response on other attacks."

Facebook Immune System by Tao Stein, Roger Chen, Karan Mangla

Harm mapping

Recommended starting point

github.com/summerscope/mapping-fair-ml

Failure-first design

  • UI interactions (Get help / This isn't right)
  • Email / SMS templates
  • Support scripts for conversations
  • Data schema design
    (capturing your learnings)

Litmus test

What if this happened to my most vulnerable customer?

Prepare for pushback

  • People don't like it when you make bad assumptions about them!
  • Making the implicit explict is going to be uncomfortable
Escape hatch

Designing an escape hatch

Litmus test

Could any customer go through my escape hatch, recover and remain a happy customer?

"Delays in feedback loops are common causes of oscillations. If you're trying to adjust a system state to your goal, but you only receive delayed information about what the system state is, you will overshoot and undershoot."

- Leverage Points: Places to Intervene in a System by Donella Meadows

Finding #4

Upstream prevention vs downstream mitigation

These sorts of things will happen. I’m very intentional about not saying abuse "might" happen - if it can, it will.
- Eva PenzeyMoog newsletter

Product use cascade

  1. What you can do in the product
  2. Intentions, framing, design
  3. Culture of your community
  4. ...
  5. Terms of use

Possible = permissible

Setting boundaries is design

Downstream is usually more costly than upstream

Section 2

Top-down vs. bottom-up

Responsible tech

No magic bullets

No magic bullets*

*Anyone who tells you otherwise is selling something

"As a result, developers are becoming frustrated by how little help is offered by highly abstract principles when it comes to the ‘day job’"*

From What to How: An Initial Review of Publicly Available AI Ethics Tools, Methods and Research to Translate Principles into Practices

"Goal: Consumers choose trustworthy products when available and demand them when they aren't."

Mozilla Trustworthy AI Paper

"AI must be designed to minimize bias and promote inclusive representation"

IBM Everyday Ethics for Artificial Intelligence

"AI systems should be safe and secure, and should serve and protect humanity"

Smart Dubai - Principles of Articificial Intelligence

Bike leaning on the side of a shed
"A string of empty profunditites"

- Andy Kitchen












— Gif by Barbara Pozzi

Document being thrown like a molotov cocktail, titled Principles of our company


How not to do principles

Commit to action

  • Assign strategy champion(s)
  • Set measurable targets
  • Provide examples
  • Allow for uncertainty

Section 3

Final thoughts

Findings

  1. Developing a healthy cynicism about data sources
  2. Mapping knowledge states
  3. Designing for failure first
  4. Upstream prevention vs downstream mitigation

Rethinking automation design

Breaking down the monolith

  • Concierge prototypes (human-as-bot)
  • Automations can be temporary or time-boxed
  • Dashboard? Notification? Escalation? Action?
  • Do usability testing (duh)
"The first principle is that you must not fool yourself – and you are the easiest person to fool."

— Richard Feynman

The real game

Trying to catch fraud,
yet not defraud ourselves

Thank you

@summerscope