A provocative question: whether modern AI, particularly large neural-network-based systems, may ultimately struggle not because of a shortage of computing power or energy, but because of fundamental software inefficiencies. The article draws inspiration from computer scientist Niklaus Wirth’s 1995 essay A Plea for Lean Software, which argued that software complexity often grows faster than hardware improvements, leading to increasingly inefficient systems.
The central argument is that today's AI models are becoming extraordinarily large and resource-intensive. Training and operating advanced AI systems requires vast amounts of data, computing infrastructure, and electricity. Critics suggest that continually scaling models may deliver diminishing returns, raising questions about whether the current approach is sustainable in the long term.
However, many researchers believe AI is far from doomed. Challenges such as software inefficiency, hallucinations, and the growing amount of AI-generated content online are real, but they are increasingly being addressed through improved architectures, better training methods, and stronger governance. Recent research has also shown that issues like "model collapse" can be mitigated by ensuring AI systems continue learning from authentic human-generated data.
Rather than signaling the end of AI, these concerns highlight a critical transition point for the industry. The future success of AI may depend less on making models bigger and more on making them smarter, more efficient, and better aligned with human needs. The debate reflects a broader realization that sustainable progress in AI will require innovation in software design, data quality, and governance—not just more computing power.
