In this blog, I will explore the crucial role of ethics in the development and use of AI-powered systems. Drawing upon my extensive experience in various industrial sectors involving AI such as oil and gas industry, consumer energy solutions and cutting-edge applications of ChatGPT, I aim to highlight how age-old ethical principles can be relevant in guiding the advancement of AI applications.
Ethics, as a philosophical concept, is not limited to science or technology; instead, it is inherently interpretative and context dependent. It addresses fundamental questions concerning what is right v/s wrong, good v/s bad, and forms the basis of morality and religious beliefs.
It is rather interesting then that the notion of ethics or morality is being applied to modern technological innovations such as AI. It is certainly not the first-time technology has seen intervention from ethics and the precedents of it go back to the times of ancient philosophers. In his famous dialog Phaedrus, Plato discusses the impact of writing on basic human skills of memory and thinking. Writing is something we don’t even consider as technology in twenty-first century, but it was a part of leading technology at the times, and Plato had concerns that writing could weaken the human capacity to memorize or even think and ultimately leading to lessening of human intellect. Whether writing actually reduced pure human intellect can still be a matter of debate, but there is no doubt that it helped humans advance as a race. Writing made learnings from one individual available to the entire population, advancing an entire generation with it. Over time, we are getting more and more open minded towards the new technology as we went through the industrial revolution in eighteenth century, followed by information revolution in 20th century along with continuous advances in medical technology. However, each new piece of technology brings new questions that cannot be answered by technology alone and need continuous correction using the chisels of ethics and morality.
The emergence of Artificial Intelligence in the 21st century presents us with a similar situation, but this technology introduces conflicts between principles that were previously harmonious. For instance, efficiency may clash with bias, privacy and security with transparency, and control with responsibility. There is no universal rule to address all these contradictions, but foundational ethical principles can serve as a reference and must be addressed on a case-by-case basis. The goal of this blog is not to undermine AI's potential impact or provide comprehensive solutions to all problems but rather to raise awareness among developers and users of this technology.
During the development phase of any AI-powered solution (aka training), historical data is used to teach the AI model, like the way humans are trained before starting a new job. The selection of comprehensive and representative data for training is crucial. Using biased or incomplete data can lead the AI model to inherit these biases, just as humans might be influenced by biased training. Despite the temptation to use smaller, biased data for simplicity and cost efficiency, developers must strive for unbiased data to ensure fairness and accuracy in AI applications. For example, an AI facial recognition system trained on limited European and African faces will likely perform poorly when recognizing Asian faces, resulting in biased outcomes with severe consequences if used for sensitive purposes.
AI models may have access to personal historical data from the users that cannot be leaked. It is also important for the AI models to be able to explain how they came up with a recommendation. Imagine an AI system that is developed to diagnose medical illness and suggest OTC medications. Such a system must be trained on medical data from thousands of patients. Consider an unusual case of hypertension for which the only recommendation is going to ER. With such a strong recommendation, the system needs to back it up with some prior evidence, which may lead to giving out the intimate medical details of a single person, causing invasion of privacy. There may not be one ideal response in this case, but all such cases must be handled with care balancing the aspects of privacy and transparency.
Control and responsibility are significant aspects of AI-powered systems that often pose ethical questions. In cases like self-driving cars causing accidents, determining responsibility can be complex. Should it lie with the driver, the car manufacturer, or the AI system itself? These questions demand thoughtful ethical analysis on a case-by-case basis. As AI applications become more pervasive, they will impact various aspects of society, from insurance claim processing to content moderation and misinformation detection in social media, further highlighting the importance of ethical considerations.
The timeless ethical principles that have been the foundation of many religions and philosophical doctrines can serve as guiding principles for AI-powered systems in conflicting situations:
As AI continues to shape the future of technology, humans must continue to direct and moderate its use through the lens of universal ethics. Our coexistence with AI can lead to a better world if we remain mindful of the ethical implications and use AI responsibly.
Dr. Ameet Joshi received his PhD from Michigan State University in 2006. He has over 15 years of experience in developing machine learning algorithms in various different industrial settings. He is currently a Data Science Product Manager at Microsoft.