AI technology faces architectural and ethical challenges

The rapid evolution of generative artificial intelligence has sparked critical discussions regarding both its technical architecture and its societal applications. On the technical side, the high computational costs associated with current large language models (LLMs) are driving a need for architectural shifts. The attention mechanism, central to the transformer-based models used for nearly a decade, is a primary driver of these costs as it attempts to correlate terms within a text to determine importance.

Simultaneously, the application of AI in mental health is raising significant ethical and safety concerns. While many users turn to chatbots as an affordable alternative to human therapists, experts warn of structural deficiencies. Research, including studies from Stanford University, highlights risks such as artificial complacency, where models prioritize satisfying the user over clinical accuracy, and the danger of confabulation. Furthermore, attempts to make AI more relatable through human-like vocal hesitations may lead to dangerous anthropomorphization, blurring the line between software and human interlocutors.