Artificial intelligence (AI) is causing the technology and economic world to re-examine its own nature, continuously showing a state of "sometimes genius, sometimes foolish".
This phenomenon is called "unbalanced intelligence" by many researchers.
Reality shows that AI can achieve impressive results in fields that require high logical thinking.
Last summer, systems developed by Google and OpenAI solved exactly 5/6 difficult math problems at the International Mathematical Olympiad, a playground for the best students in the world. This achievement makes many people believe that AI is getting very close to human intelligence.
However, not long after, a simple example shows the opposite.
In a situation where a software engineer in Sri Lanka asked the chatbot whether to walk or drive to travel a distance of only 50 meters, most AI systems answered "walking", which is a logical choice.
However, this question is actually intended to test the ability to understand everyday context. Humans easily recognize this as an unreasonable situation to raise the issue, while AI only stops at choosing the correct answer without grasping the intention hidden behind the question.
This mismatch clearly reflects the nature of AI today when it excels in some tasks but is weak in seemingly simple situations.
Researcher Andrej Karpathy – co-founder of OpenAI called it the "non-uniform intelligence of AI", where some capabilities reach excellent levels, while others fail unpredictably.
Unlike humans, where cognitive skills often develop evenly and are closely linked, AI learns through identifying data patterns from the internet.
This helps AI write code, answer questions or analyze text very well, but is limited in planning, flexible reasoning or handling unprecedented situations.
This concept is changing the perspective on the future of AI. Instead of considering AI as a system that can achieve general intelligence like humans, many experts believe that it is a completely different form of intelligence when it is strong in some points, but weak in many other points.
This also directly affects the labor market. According to economists, jobs with many different tasks will not be completely replaced.
Instead, AI will automate some parts, helping people focus on more important tasks.
However, the development speed of AI still worries many people. Systems like Claude or ChatGPT have been able to write code faster than humans, although supervision is still needed to understand the overall system.
Meanwhile, new tests such as ARC-AGI 3 show that AI still has difficulty with puzzles that normal humans can solve.
Experts believe that the future of AI does not lie in whether it will replace humans or not, but in which tasks, at what time.
When understanding the nature of "non-uniform intelligence of AI", society will have a more realistic view of the role of AI in the coming years.
