Accountability is a critical piece of AI ethics, and it is relevant in all stages of the lifecycle of an AI system, including during design, after…
Accountability is a critical piece of AI ethics, and it is relevant in all stages of the lifecycle of an AI system, including during design, after deployment, as well as when any harm occurs because of the system. In these lessons on Accountability in AI, you’ll learn how AI systems can be assessed using principles of accountability, along with regulatory reforms and legal frameworks and their intersection with accountability.
Complex systems, although more difficult to assess, are especially important to analyze to determine potential impacts and harms. You will learn how AI systems can be assessed, including Human Rights Impact Assessments (HRIAs), and why it is especially important to conduct preventative impact evaluations, so any potential harm can be mitigated.
There are several additional, specific principles that are important to accountability, including validation, evaluation, and auditing. These lessons will cover accountability and several important principles: the ability to appeal a decision made by an AI and the ability to challenge the use of the AI system to make that decision.
Finally, some areas are seeing increasing regulatory reforms and more robust legal framework in terms of accountability. But it is unclear which current regulations are appropriate for AI systems. We will discuss why regulatory compliance is an extremely important consideration for organizations, and liability and legal considerations around AI system design, development, and deployment.
Learning Objectives:
- Describe how AI systems can be assessed
- Identify issues around regulatory reforms
- Explain how accountability is important to AI ethics
Skills you’ll gain
AccountabilityApplied EthicsData EthicsEngineering EthicsReferential IntegritySocial AccountabilityWhat You'll Learn
- Describe how AI systems can be assessed using principles of accountability, including Human Rights Impact Assessments (HRIAs)
- Identify issues around regulatory reforms and legal frameworks relevant to AI accountability
- Explain how accountability is important to AI ethics across the AI system lifecycle
- Apply principles of validation, evaluation, and auditing when assessing AI systems
- Evaluate the ability to appeal and challenge decisions made by an AI system
- Examine liability and legal considerations around AI system design, development, and deployment
Key Takeaways
- Accountability is relevant in all stages of an AI system's lifecycle, including design, after deployment, and when harm occurs.
- Complex systems are more difficult to assess but especially important to analyze for potential impacts and harms.
- Preventative impact evaluations, such as Human Rights Impact Assessments, help mitigate potential harm before it happens.
- Accountability includes the ability to appeal an AI decision and to challenge the use of an AI system to make that decision.
- Regulatory compliance is an important consideration for organizations, though it is unclear which current regulations are appropriate for AI systems.
Frequently Asked Questions
What does this course cover?
It covers accountability in AI, including how AI systems can be assessed using principles of accountability, Human Rights Impact Assessments (HRIAs), validation, evaluation and auditing, the ability to appeal and challenge AI decisions, and regulatory reforms, legal frameworks, and liability considerations.
What will I learn about assessing AI systems?
You will learn how AI systems can be assessed, including the use of Human Rights Impact Assessments (HRIAs), and why it is important to conduct preventative impact evaluations so that potential harm can be mitigated.
Which topics are included in the lessons?
The lessons cover Validation, Evaluation, and Auditing; Assessing Impact of AI Systems; Ability to Appeal; and Liability and Legal Considerations.
What does the course say about regulation and accountability?
Some areas are seeing increasing regulatory reforms and more robust legal frameworks for accountability, but it is unclear which current regulations are appropriate for AI systems. The course discusses why regulatory compliance is an extremely important consideration for organizations, along with liability and legal considerations around AI system design, development, and deployment.
What skills does this course relate to?
It relates to Accountability, Applied Ethics, Data Ethics, Engineering Ethics, Referential Integrity, and Social Accountability.
Transcript
Show transcript (free preview lesson)
Transcript of the free preview lesson. Remaining lessons unlock with the full course.
(bright music) At its core, accountability is a fundamental piece of AI ethics. Accountability is relevant in all stages of the lifecycle of an AI system, including during design and after deployment. As well as when any harm occurs as a result of the system. It's easy to imagine an AI system for which no human takes responsibility for the decisions made by the system and thus no one is held accountable. In fact, there are many such systems that have been created already. But the principle of accountability says that there must always be a governing body of humans who are accountable for the AI system at all of its stages. Having a clear policy for accountability is especially important when technologies have the potential to impact humans on in enormous scale. And there are several additional specific principles that are important to accountability. Validation, evaluation, and auditing. Model validation is an important part of building any AI system. Design decisions that are made by the team building the model. How, when, and what data is collected. Decisions on data cleaning and excluding outliers. Combining and linking data sources. Treating missing variables. Have a demonstrable impact on the final model. Even which modeling method is used and why can make a major difference on the actual impact of the model upon deployment. Validation and evaluation may include technical checks to see how accurate or robust a predictive model might be, including the identification of false positives and false negatives. It might also include the evaluation and tracking that is performed after the AI is deployed, along with decisions made based on this performance. For this reason, organizations are looking for ways to externally validate, evaluate, and audit their systems. This suggests that in some cases, there could be an external monitoring body. There is debate on whether this monitoring body would be internal to the organization or an external group. Both possibilities provide challenges to organizations seeking external validation. Some countries have suggested national standards or groups that may take on this role. Validation, evaluation, and auditing are important components within a proper AI ethics framework. But organizations should be proactive in ensuring that these principles are systematized and created explicitly as part of processes. This would guarantee accountability in the design, deployment, and assessment of the AI systems they built.
Learn on the Go
Take your learning anywhere — the KnowledgeCity mobile app lets you watch lessons on the go.