<aside>

Summary

The whole goal of innovation and engineering is to benefit human performance/productivity. There needs to be a balance between automation and human control if we want to develop safe, trustworthy, and reliable products that advances human progress. This content focuses on the HCAI framework

</aside>

<aside>

Resources

Readings

Readings 2

What is explainable AI? - IBM

What is Black Box AI? - IBM

</aside>

<aside>

Keywords -

• Accountability

• Privacy

• Transparency

• Explainability

• explicability: it can be explained and understood because it is logical or sensible

• RST: Reliable, Safe, Trustworthy • Human Mastery: High human control, Low automation

• Requires rapid action: High automation, low human control

• XAI: Explainable Artificial Intelligence allows human users to comprehend and trust the results and output created by machine learning algorithms.

</aside>

<aside> 📌

Measures for Privacy

1. Secrecy: concerns information that others may gather about us
2. Anonymity: how much of us is in the public gaze
3. Solitude: degree to which others have physical access </aside>

<aside> 📌

CONSENT CRITERIA:

• Adequately informed
• voluntary consent
• has the capacity to understand and communicate their consent
• THE RIGHT TO BE FORGOTTEN (Data deleted) </aside>

<aside> 📌

OCCAM’s RAZOR

A problem-solving principle that states, when faced with competing explanations for the same phenomenon, the simplest one with the fewest assumptions is usually the correct one.

opposite: Hickam’s Dictum

</aside>

<aside> 📌

Trust in Humans vs Trust in Technology

The criteria for trust/ Nature of expectation for people vs technology is different

</aside>

<aside> 📌

ACM Principles for Algorithmic Transparency and Accountability

There are 4 main barriers to Accountability:

1. Many people involved (the problem of many hands)
2. The problem of bugs
3. Blaming the computer
4. Software ownership without liability

Transparency

Black-box and white-box describe algorithms based on their level of internal transparency. They are difference in terms of balance between interpretability and transparency.

Black Box Algorithm

• internal logic is hidden; Lack transparency
• only input and output is seen by end users
  • eg OpenAI’s ChatGPT
• AI models are trained on massive data sets through complex deep learning processes, and even their own creators do not fully understand how they work.
• The opacity of a black box model can hide cybersecurity vulnerabilities, biases, privacy violations and other problems.

White Box Algorithm

• internal logic is fully visible and understandable
• XAI
  • eg autonomous vehicles clarifying their braking decisions, financial fraud detection explaining why a transaction was flagged
• White box AI models make it easier to trust and validate outcomes, as well as tweak models to correct errors and adjust performance. But it isn’t easy to turn every AI into a white box.

other explicability measures (e.g. traceability, auditability and transparent communication on system capabilities) are used if explanation for a particular output is not possible

Ali et al, “Explainable Artificial Intelligence (XAI): What we know and what is left to attain Trustworthy Artificial Intelligence”, Information Fusion, 99, Nov 2023

</aside>

<aside> 📌

Algorithmic Biases and Privacy

A user is exposed to certain ads → user interacts with ad → LLMs construct a detailed and sensitive demographic profile of the user = ⚠️ Privacy Issue

</aside>

<aside> 💡

Human-Centred Artificial Intelligence (HCAI) framework

A two-dimensional framework that separates the levels of automation from levels of human control

1. Design for high levels of human control and high levels of computer automation so as to increase human performance
2. Understand the situations in which full human control or full computer control are necessary
3. Avoid the dangers of excessive human control ot excessive computer control

TL;DR - Balance human vs computer control in varying situations by assessing the benefit to humans and the risks of too much of one side

Methods of HCAI tend to produce designs that are Reliable, Safe, and Trustworthy

Previous one-dimensional list that suggested that increase in automation comes at the cost of lowering human control

Self driving car trade off Table (US Soc of Automotive Engineers)

Robin Murphy’s Law of autonomous robots: Any deployment of robotic systems will fall short of the target level of autonomy, creating or exacerbating a shortfall in mechanisms for coordination with human problem holders

Eg. Disaster from excess autonomy - Patriot missile sustem shooting down two friendly aircrafts during Iraq War

</aside>

<aside> 💡

</aside>

<aside>

</aside>

<aside>

</aside>

<aside> 💡

</aside>