Amid the philosophical and ethical issues swirling around the booming artificial intelligence (AI) revolution鈥攆rom questions about privacy, safety risks and surveillance to generative AI deepfakes, AI companionship and AI systems going rogue鈥�51爆料网 professor Johannes Himmelreich views its use with guarded optimism. 鈥淚 think there is a good way of using AI and I want to contribute to finding it,鈥� he says. 鈥淲e should be using AI, but the question is how should we use it?鈥�
Himmelreich has long been fascinated by the role of philosophy and ethics in technology. And as the landscape of AI rapidly expands, he is investigating its impact on society. 鈥淚 have interests across a whole range of topics in philosophy, and what binds many of them together is their association with AI,鈥� says the associate professor of public administration and international affairs in the . 鈥淚鈥檓 drawn to issues that intersect with emerging social topics or use new methods. That鈥檚 why computation and AI are an attractive topic for me to work on鈥攕o many changes coming into philosophy today are from these areas.鈥�
Himmelreich is focused on AI鈥檚 use in government and its role in decision-making, as well as AI regulations and policy. He鈥檚 co-editor of (Oxford University Press, 2024), which examines the challenges and opportunities at the intersection of AI and governance. His interest in politics, social justice, computation and the digital economy often spark his interdisciplinary approach to research. He鈥檚 delved into such topics as killer robots and self-driving cars. With the support of a two-year grant from the , Himmelreich is working on a book about the philosophy and ethics of data science and good decision-making. 鈥淚f data science is about supporting decision-making, then you want to make sure the decisions are fair and don鈥檛 harm people,鈥� he says.
Deliberating Data鈥檚 Ethical Dilemmas
Data-driven decision-making is deeply embedded in today鈥檚 world, offering informed choices through analysis designed to improve performance. The process is routinely accepted as factual and accurate鈥攁 fail-safe counter to 鈥済ut instinct鈥� decisions. But for all the successes, if data is flawed, it can lead to unforeseen problems. And since massive datasets fuel AI, Himmelreich sees inherent ethical dilemmas associated with data鈥攂ias and privacy risks, for instance. His current focus is on how data science methods that researchers use to collect, clean, analyze, model and present data can unintentionally distort the truth.
鈥淭he question鈥�What is good data?鈥攊s not as easy to answer as you might think, because good data is sometimes made up, generated by what鈥檚 called synthetic data,鈥� he says.
Synthetic data is artificially generated information that resembles real data, mimicking its patterns, relationships, and structures. Although the information isn鈥檛 collected from actual people or events, it is used to test and train AI models.
Often this type of data drives AI鈥檚 progress鈥攊n applications such as ChatGPT, facial recognition and tabular data. 鈥淎I becomes smarter with more data, and the more data you have, the better your AI is,鈥� Himmelreich says. 鈥淭hat scaling law holds also from made-up data. The algorithm doesn鈥檛 care if the data is real or not. It just wants more.鈥�
If the original data is distorted, the synthetic data can inherit that distortion. 鈥淭he sinister part is not that synthetic data is made up,鈥� he says. 鈥淵ou might not even know it鈥檚 happening because to show distortion you need to have data, and oftentimes you don鈥檛 have the data to show there鈥檚 a distortion.鈥�
Employing AI in Government Decision-Making
Maxwell School professor Johannes Himmelreich is interested in examining how data science methods used by researchers can unintentionally distort the truth.
Himmelreich highlights one routine example of AI employed in the public sector: The Social Security Administration uses AI to help decide who qualifies for disability benefits鈥攂ringing automation into some of the most critical, human-centered decisions in government. AI systems are trained to detect fraud, seemingly making it simpler to process claims and decisions, but they produce both false positives and negatives.
This raises the ethical issue of which mistake is worse: denying an individual鈥檚 legitimate claim, which could jeopardize their livelihood, or missing a fraudulent case that will cost taxpayers. 鈥淭he challenge is to figure out which mistakes are you more willing to accept given that you will make mistakes,鈥� he says.
An appropriate use of AI, Himmelreich believes, is in sorting cases based on their difficulty, prompting decision-makers to grant more attention to challenging cases. 鈥淭hat鈥檚 not just true for disability insurance or unemployment insurance, it鈥檚 also true in the medical sector,鈥� he says. 鈥淲e are already using AI for cancer diagnosis and assisting radiologists in reading medical imaging.鈥�
Himmelreich believes that problems in AI projects often happen at the point where humans and machines interact. To prevent these breakdowns, he says it鈥檚 important to have a clear goal, a well-defined step-by-step process and strong communication. 鈥淕ood data science is exactly at this interface and not necessarily in the technical analysis,鈥� he says. 鈥淭he really important skill for data scientists is to understand the situation of the decision-makers they鈥檙e supporting and produce something that augments their work and helps them make better decisions.鈥�
Balancing Benefits and Harms
In his Philosophy and Ethics of Data Science graduate course, Himmelreich emphasizes that data science involves countless dilemmas, trade-offs and value conflicts. 鈥淚 want my students to have the knowledge and methods to solve those dilemmas and to have confidence in their ability to solve them,鈥� he says.
From the early beginnings, safety concerns have gone hand in hand with AI. Himmelreich cautions that the more prevalent AI鈥檚 use becomes in government and other areas, the more risk is introduced. 鈥淭he harms that can come from AI can be much more nefarious because they鈥檙e harder to detect,鈥� he says. 鈥淭he question of how we control AI is really relevant.鈥�
Himmelreich is drawn to how technologies are designed and built and understanding how they鈥檙e implemented鈥攆or better or worse. Ultimately, he hopes AI enhances society, providing support where it鈥檚 needed, rather than replacing us. This will enable us to excel in our strengths and reap the rewards of AI while acknowledging its limitations and guarding against erroneous behavior. 鈥淭o me,鈥� he says, 鈥渢he ethical questions are super important because over the next few years we will make important decisions about how we鈥檙e going to use AI.鈥�
