Presentation via Zoom By Dmitry Semenov, National MagLab

Title: What Is Wrong with AI

Host: Science Council's MagX Presentation

Discussion Leader: Steve McGill

Abstract: Artificial Intelligence has achieved remarkable progress—machines can recognize faces, write essays, and even generate original art. Yet behind these successes lie fundamental problems that challenge both science and society. This presentation explores what is wrong with AI—not only the well-known ethical and social concerns, but also the technical, mathematical, and conceptual flaws that limit how machines learn and reason. We will unpack how machine learning systems depend on imperfect data, why neural networks sometimes make absurd mistakes, and how large language models (LLMs) like ChatGPT can appear intelligent while still lacking true understanding. Through simple visual examples and short code demonstrations, the talk illustrates how small changes in data can lead to large and unpredictable outcomes, revealing the sources of error, bias, and uncertainty in AI models. Rather than condemning AI, this session invites a balanced look at its limitations and potential. By understanding what is wrong—from overfitting to hallucination—we gain insight into how to make AI more reliable, interpretable, and aligned with human values. The goal is to encourage both scientists and the public to see AI not as magic, but as an evolving tool shaped by mathematics, data, and human judgment.