Google

The inadequacy of outcome-only feedback for complex tasks can be understood through an analogy with a student learning math. Informing the student only whether their final answer is correct or incorrect does not help them identify their mistakes. Effective learning requires process-based guidance, such as a step-by-step explanation of the solution, which clarifies the underlying logic and concepts.

Learning Analogy: Outcome vs. Process Feedback

A development team is training an AI assistant to solve complex, multi-step word problems in mathematics. Their current training method works as follows: the AI generates a complete solution, and if the final numerical answer is correct, the entire solution is marked as 'good.' If the final answer is incorrect, the entire solution is marked as 'bad.' Despite extensive training, the AI struggles to generalize its problem-solving skills to new types of problems.

Critique of an AI Training Methodology

A team is training a robot to assemble a complex piece of furniture from a kit. The assembly requires 20 distinct steps. Which of the following training approaches would be most effective for teaching the robot to perform this task reliably and to correct its own mistakes in the future?

Imagine you are explaining to a colleague why a language model needs detailed, step-by-step feedback to learn complex problem-solving, rather than just being told if its final answer is correct. Create a simple, original analogy (different from a student learning math) to illustrate this point. Explain how your analogy demonstrates the superiority of process-based feedback over outcome-based feedback.

Learn Before

Related