From Humans to Machines

<aside>

Summary

Decisions can be deemed “well-made” if the triad of accountability, fairness, and transparency can be assessed against success conditions and failure conditions. Humans decisions can satisfy the success conditions of the triad where machines fall short

</aside>

<aside>

Reading

Previously…

Autonomous vehicles

Automated decision making systems

Value-sensitive Design

Algorithmic Bias (eg 2007 by researchers at Stanford University and Princeton University, Compass)

Trust, Transparency, and Explainablility

</aside>

<aside>

Keywords -

• Explanation - any reason, generally NOT subject to judgement

•Justification - has a right or wrong; assessed on the mode of justice • Contingent fact: It can happen or not “the rose is red”

Conditional - “if, Then Statement”

Counterfactual: special type of conditional. Contrary to fact conditional; relating to or expressing what has not happened or is not the case.

describes an outcome by proposing a "what-if" scenario, explaining that if a specific input had been different, the result would have changed

Substantive: having a firm basis in reality and so important, meaningful, or considerable.

Epistemic: relating to knowledge

Ontological: relating to the branch of meta physics dealing with the nature of being

Asymmetric overdetermination & Symmetric overdetermination:

</aside>

excessive epistemic deficits can be the root of conspiracy theories.

<aside> 📌

Side notes:

Business of -

Science: provide explanation

Philosophy: distinguish true and false

Psychology: reveal what goes on in the brain given an input

Engineering: Build stuff

</aside>

<aside> 👌🏻

(Other) Models of Explanation/Causation:

Pragmatic Model: Use whatever explanation (when there exists multiple possible explanations) that works for you in the context.

Sufficiency Model: Cause is sufficient for the effects

McDermit Model of causation: Causation is real but causes are neither necessary nor sufficient for the effects.

Deductive nomological

hybrid

statistical

</aside>

Sometimes we say that some shit is related to stuff that happens purely to soothe the very human need to make sense of the world we live in; we make it make sense so we don’t go crazy

Intentional logic: Very hard math, will need to convince Seb to explain but lets not

<aside> ✔️

Study Questions:

  1. Is a desire for value neutrality value neutral?
  2. What happens if we replace the causal model of explanation with a statistical model of explanation? </aside>

<aside>

Introduction

Salient Question - Is technology value neutral?

depends on what you mean by value and what you mean by neutrality

Definition of Artificial Intelligence has changed much over time - current term being coined is “Agentic AI”

There are two claims to be discussed relating to the triad of accountability, fairness, and transparency in the contents of humans and in the context of (automated) machines

</aside>

<aside> 📌

Well-made decisions

Decision Making ←→ Reasoning

When we are reasoning, we are doing something

For a human or machine-based decision to be well-made, they must be:

For this triad to be substantive - there needs to be a criteria to assess the success of each member

What would it look like for this thing to fail

</aside>

<aside> 📌

Claim One - AFT Triad and HDM

The triad of accountability, fairness, and transparency is satisfiable for human based decision making

  1. Epistemic actions are the sort of things we do to relieve an epistemic deficit; Actions intended to help in the discovery of information. We correlate decisions with actions based in the psychological domain
  2. Humans identify accountability with the provision of an explanation.

Being accountable is to accept responsibility → responsibility is accepting that you MUST respond.

ie. to be accountable means to give a clear and honest explanation for the actions and decisions you make

  1. We adopt a causal model of explanation

  2. We specify causal relations via counterfactuals.

  3. We believe that beliefs and desires are both necessary on its own, and jointly sufficient to explain behaviour

    eg. If Sarah takes an umbrella when leaving the house:

    • She believes it’s going to rain.
    • She desires to stay dry.

    Both are needed to explain her action — if she didn’t believe it would rain, or didn’t care about staying dry, she wouldn’t take the umbrella.

  4. Our explanation of accountability helps create a clear and workable connection between fairness and transparency

  5. We support claims about transparency by relying on what people say and declare

  6. We argue that what people believe and want is enough to explain not just why they acted, but also the moral reasons behind their actions.

  7. We think these moral reasons are enough to determine that a decision fair. </aside>

<aside> 📌

</aside>

<aside> 📌

</aside>

<aside> 📌

</aside>

<aside> 🔥

</aside>

<aside> 📌

</aside>

<aside> 🔥

</aside>

<aside> 📌

</aside>

<aside> 🔥

</aside>

<aside>

</aside>