New Delhi: The rapid rise of Artificial Intelligence (AI) is transforming technology, but fully autonomous systems remain out of reach, according to Neil Thompson, director of the FutureTech Research Project at MIT’s Computer Science and Artificial Intelligence Lab. Speaking at a technology summit in New Delhi, Thompson highlighted both the promise and the pitfalls of AI in everyday applications.
Thompson pointed to examples such as the misuse of deepfakes for fraud and occasional failures of navigation apps to deliver accurate, real-time directions. “Current AI models make errors because their capabilities are not yet sufficient for full autonomy,” he said, emphasizing that the technology is fundamentally different from traditional IT systems.
“AI is a departure from conventional software. If you ask Excel to multiply numbers, it will always be correct. But ask it what to have for dinner, and it will fail. With AI, even numerical calculations can sometimes be wrong, yet it will still provide an answer when asked about subjective questions,” Thompson explained. His remarks underscore the unpredictable nature of AI outputs and the difference between deterministic computing and probabilistic reasoning in machine learning.
Thompson described what he calls the “AI garden arranging path” to explain why errors occur. In traditional systems, stacking components preserves accuracy because each system is deterministic. In AI, however, stacking imperfect systems compounds errors rather than eliminating them. “When multiple AI systems are combined, the errors multiply,” he said, highlighting the importance of rigorous validation in AI deployments.
Looking to the future, Thompson warned that more advanced AI systems could bypass human controls, creating potential safety and ethical challenges. “As AI becomes smarter, it can more easily circumvent the constraints we impose. There is no clear evidence yet that we can fully control such systems,” he said, urging governments and institutions to invest heavily in AI research and oversight to ensure safe and reliable applications.
Thompson’s analysis serves as a cautionary note for the global tech community: while AI offers unprecedented capabilities, careful governance, continuous research, and clear regulatory frameworks are essential to prevent errors from escalating into larger societal problems.