Author: Cem Dilmegani
Publisher: AI Multiple
Publication Year: 2022
Summary: The following article explores 4 principles for responsible AI design and recommends best practices. The 4 principles, along with best practices, include: 1). achieving fairness by ensuring the training data is representative of the population, analyzing outcomes among subpopulations, and continuous monitoring of the model’s performance over time; 2). Ensuring privacy by classifying data according to sensitivity, implementing data access and usage restrictions for employees, encrypting data, and using synthetic data; 3). Achieving security to fend off data poisoning and model poisoning by keeping up with new developments in AI attacks and AI security and creating a red team within your organization to identify security issues; and 4). Ensuring transparency by first discussing the required level of interpretability with domain experts, prioritizing explainable AI methods, and using simpler models with a smaller set of inputs rather than complex models with hundreds of variables.