According to a report by MIT Technology Review, many AI models are “black boxes,” meaning their inner workings are unclear and unpredictable. This creates difficulties for researchers who want to explain, develop and control artificial intelligence systems.

Some experts attribute the lack of understanding to the field’s infancy, arguing that the nascent nature of AI means that researchers sometimes have to study experimental results and conclusions.

But some researchers warn that this approach is risky, especially as artificial intelligence becomes more integrated into human life and society. They argue that safeguards are important to ensure the security, reliability, and ethics of AI applications.

If you can do great things but can’t control them, then it’s not that surprising. It’s like driving a car at 300 miles per hour with a shaking steering wheel.
– said computer scientist Mykhailo Belkin from the University of California, San Diego.

Scientists agree that the AI ​​industry needs to develop more rigorous theoretical frameworks and methods for understanding and testing its models.

Scientists also call on their colleagues to collaborate and communicate more closely with artificial intelligence from different disciplines and with different experiences.

“These are exciting times. Many people working in this field often compare it to early 20th-century physics. We have a lot of experimental results that we don’t fully understand, and often when you do an experiment it surprises you.” Boaz Barak, a computer scientist from Harvard University, said: