Why Human Performance May Not Give a Useful Error Target
Question: Why can human performance be a poor guide for setting a target error rate in a system that reviews insurance claims or sorts customer support requests?
Sample answer: If people find the task difficult, their results may be too weak to serve as a useful reference point. In that case, human performance does not tell you what a strong system should achieve, so it becomes hard to estimate the best realistic error rate for the model.
Key points:
- Humans may struggle with tasks such as claim review or support-request sorting
- Weak human results are not a dependable baseline
- Without a dependable baseline, the best achievable error rate is hard to estimate
Rubric: The essay should explain that when humans perform poorly, their performance is not a reliable benchmark, which makes the target error rate difficult to estimate, and it should use one of the given task examples or a comparable one.
0
1
Tags
Machine Learning
Deep Learning
Supervised Learning
Dive into Deep Learning @ D2L
Data Science
Machine Learning Strategy
Machine Learning Yearning @ DeepLearning.AI
Related
Why Human Baselines Are Less Helpful on Hard-to-Judge Tasks
Which situation makes it hard to estimate the best achievable error rate for a task?
Recommending products to shoppers is an example of a task that people may find hard to evaluate well.
When human performance is poor, estimating the _____ error rate becomes difficult.
Match each concept to the description that best fits how its error rate can be estimated.
Order the steps for judging whether expert performance can set a target error rate.
Which task is given as an example of a problem that is hard for people to judge reliably?
Human performance is always a dependable estimate of the best achievable error rate, even when people find the task difficult.
Predicting Which _____ to Show a Shopper Can Be Hard for People Too
Match Each Situation to Its Effect on Estimating the Best Possible Error
Order the reasoning steps for a task that is difficult even for people
Why Human Performance May Not Give a Useful Error Target
Estimating a Benchmark for Ticket Routing
Examples where human judgment is a poor guide to optimal error