Ethical AI in a Rapidly Evolving Landscape - deep dive

[bacharakis]

Session description

Have you ever wondered if AI Ethics is an Engineering or a Product issue? Which software development life cycle phase should you start addressing ethics in your AI system? How do we ensure that ethics are integrated into the solutions by design and not only implemented on top as an afterthought?

Addressing the need for fairness and accountability in AI, this talk will explore our efforts to navigate cutting-edge technology while ensuring the ethical development of AI solutions. We will examine the requisite shift in perspective and straightforward, actionable measures that empower enterprises to seamlessly infuse ethical AI into their core values, steering clear of prevalent pitfalls.

Session goal

Understand how to navigate cutting-edge technology and build an open web while ensuring the ethical development of AI solutions.

Additional session chairs (Optional)

@humeranoor

IRC channel (Optional)

ethical-ai

Who can attend

Anyone may attend (Default)

Session duration

60 minutes (Default)

Other sessions where we should avoid scheduling conflicts (Optional)

No response

Estimated number of in-person attendees

Don't know (Default)

Instructions for meeting planners (Optional)

No response

Agenda, minutes, slides, etc. (Optional)

• Agenda: https://docs.google.com/document/d/1od0BwLAHxemYRtJnsR7cf8fL7t9z-9RoNmuMVPTdl3o/edit
• Slides: https://docs.google.com/presentation/d/1kOQppEX-SNDLPMEe13Mg-yIQuym7MKl7Eb2CJvaDBbQ/edit#slide=id.p1
• Minutes
• Live Google Doc Minutes
• Calendar

[humeranoor]

This is a sequel of our lightning talk at the AC meeting. Here we want to show case the concrete steps that we're taking at eyeo for putting AI ethics in practice, and looking forward to kick-off a discussion and knowledge-sharing session on how everyone else is doing the same. It's great to see that the standards have been defined, now what about the implementation?

[bacharakis]

Notes from the session

[bacharakis]

Session Recap: The session had two parts:

1. the theoretical for setting the note and
2. the practical/workshop style for addressing the question: Are you thinking or applying any ethic-related values in your product development process regarding AI? What are we doing now, and what are we willing to do?

In an effort to summarize the session, here are some risks identified and some opportunities for the space:

Risks:

• Unacceptable Use Cases: There's a risk that AI developers may create use cases or prototypes that are ethically unacceptable, potentially causing harm or controversy.
• Data Bias: Data bias in AI systems poses a risk of perpetuating unfairness and discrimination, which can negatively affect marginalized groups.
• Privacy and Bias Protection: Failing to safeguard user privacy and prevent biases can lead to ethical violations and potential harm to individuals, particularly vulnerable populations.
• Loss of User Agency: AI systems pre-emptively anticipating user needs may risk eroding user agency and limiting their active participation in decision-making.
• Ethical Variations: Differences in ethical values and principles globally can pose a risk to establishing universal ethical standards for AI, potentially leading to ethical conflicts.
• Educational Gaps: Lack of education and awareness about AI ethics can result in users' inability to validate information or make informed decisions, which may lead to misinformation.

Opportunities:

• Operationalizing Ethics: Organizations like Intel are adopting legal principles to integrate ethics into AI practices, providing an opportunity to promote ethical AI development.
• Open-Source LLMs: Encouraging the development of decentralized and open-source LLMs, as suggested by Taiwan's Ministry of Digital Affairs, can enhance transparency and inclusivity in AI technology.
• Public Engagement: Initiating deliberative processes involving the public to determine AI ethics creates an opportunity for more inclusive, informed, and balanced ethical standards.
• AI Explainability: The need for AI explainability provides an opportunity to make AI systems more transparent and comprehensible, allowing users to understand AI decision-making processes.
• Continuous Testing: Recognizing the need for ongoing security and ethics testing creates an opportunity to maintain and enhance the ethical integrity of AI systems.
• Reexamining Ethics: The call to reexamine ethics in politics and business due to AI advancements presents an opportunity to reshape and align these domains with more ethical practices.

These identified risks and opportunities reflect the multifaceted nature of ethical considerations in AI and the potential for both positive and negative outcomes in AI development and implementation.