AI: We must learn to recognize AI's errors

We cannot rely on artificial intelligence outputs if we do not understand how and why AI arrives at its conclusions. This perspective, shared by reader Tuomas Peurakoski, argues against blindly trusting AI, highlighting a significant danger in assuming AI is what we hope it to be, rather than what it currently is.

Peurakoski agrees with Anna Lahtinen's earlier point (HS Opinion, Aug. 10) that AI should not be trusted blindly. He sees a major threat in the use of AI and large language models, where users might mistakenly believe AI performs as desired, overlooking its current limitations.

Evaluating AI outputs requires expertise that is difficult to cultivate if AI handles simpler tasks for beginners. This evaluation process differs significantly from the act of creation itself. Minor "hallucinations" or small assessment errors by AI could go unnoticed and potentially escalate into substantial problems when scaled up.

Therefore, it is essential to discuss how future generations will learn the skills needed to assess AI-generated results. If we do not understand the reasoning behind an AI's solution, its output may not be safe to use. Peurakoski, an IT professional from Helsinki, stresses the importance of understanding the 'why' behind AI's actions.