AI and Ethics: Building a Framework

Artificial Intelligence (AI) is top of mind for every organization. As they consider their AI strategy, the discussions should include how to implement AI responsibly. Organizations should have a steering or governance committee developing AI guidelines and reviewing and auditing adherence, yet according to an article by VentureBeat, 63% of organizations surveyed by Wakefield Research and Juniper are on track for their AI implementation goals even though only 9% have fully developed governance policies in place.

Ethics is defined as a set of moral principles or standards of right and wrong that govern a person’s or institution’s behavior. Moral principles are a code of conduct for governing behavior that, given specified conditions, would be put forward by all rational people. As you develop your own AI policies, here are some AI ethics guidelines to consider.

As it relates to AI, there are several generally accepted key pillars of ethics that should be included when creating AI governance policies or frameworks.

Integrity (Beneficence and Non-Maleficence), Fairness and Non-bias

AI should be employed in a manner that is beneficial for humans and society. The AI should do no harm and respect human dignity, rights, and freedoms. The AI should also comply with all laws, rules, and regulations applicable to the industry or purpose for which it was created.

Accountability and Governance

The AI must be trustworthy and dependable and function correctly. AI must complete the tasks correctly and be evaluated continuously to ensure that business processes operate correctly. There must be mechanisms to provide accountability for the AI’s use.

Reliability

AI must also be reliable. As new sources of data are added, the outputs must be monitored and validated. The reliability of the algorithms can change over time due to concept drift, defined as situational changes making the relationships between system inputs and outputs unstable over time.

Explainability or Transparency

AI models should be clearly understood and explained across departments and organizations to ensure the models are clear, everyone understands the training data set origins and rights of use, and that there are no inherent biases. In addition, there should be transparency around AI’s methods, applications, and uses.

Security and Control

AI models must be developed using a security-first approach focusing on safety and cyber resiliency. AI models should be developed with the ability for a human to take control of any process, e.g., with self-driving vehicles, a human should always be able to override the operation and control of the vehicle.

Privacy

Protecting the data used to train the AI and created by the AI is paramount. Ownership of the data should be understood at the inception of the project.

Governments have already turned their attention toward AI regulation. The EU recently passed a draft of the AI Act, and the US White House Office of Science and Technology Policy has published the Blueprint for an AI Bill of Rights. Organizations and institutions should prepare for forthcoming AI legislation by putting AI governance practices together before implementing AI in order to protect against risk and litigation.

Our next issue will examine the AI Act and the Blueprint for an AI Bill of Rights in greater detail.

#IvonneOnTech #AI #EthicsinAI

Sources:

?Engati Team. (2023, January 20). 5 pillars of responsible and ethical AI. Tech Corner. Engati. Retrieved on April 27, 2023, from https://www.engati.com/blog/responsible-and-ethical-ai

?Friday, B. (2023, January 16). Building an AI governance strategy that works. VentureBeat. Retrieved on April 27, 2023, from https://venturebeat.com/vb-in-conversation-reimagining-the-data-center-in-todays-environment-promiti-dutta-citi/

?Gupta, V. (2022, November 4). AI ethics: 5 key pillars. The Enterprisers Project. Red Hat. Retrieved on April 27, 2023, from https://enterprisersproject.com/article/2022/11/ai-ethics-5-key-pillars