Scientists published in Nature developed microchips through artificial intelligence which produced superior designs than human-made ones but the creators themselves cannot explain these designs. Princeton electrical engineer Kaushik Sengupta directed the team that used deep learning to produce chip layouts which seemed random and alien in appearance. According to Live Science Sengupta stated that humans lack the ability to comprehend these designs which resemble H.R. Giger paintings.
The new synthesis method begins with desired performance results that AI uses to develop functional structures through reverse engineering. The results? The new electromagnetic chip designs outperform all previous methods in their optimized state. The $4.5 billion millimeter-wave wireless chip industry stands to experience a massive impact because market projections indicate it will triple during the next six years.
Building conventional computer chips needs multiple experts who must perform multiple tests with their specialized knowledge through a trial and error method. The process of workflow becomes faster with AI because it generates unexpected effective solutions. The system produces imperfect results which force designers to step in and handle the failed designs. According to Sengupta the main goal is to boost productivity rather than eliminate human designers from the process.
The promising innovation of AI technology sparks doubts about whether we are building an era where we will use systems whose operation remains completely unclear. The transformations brought by AI technology in engineering create new doubts about our capacity to handle the tools we develop and how we generate creative work.