The Human-Centred AI development process

Building Human-Centered AI products and services requires hard questions across different steps of the AI development process.

Here are some helpful questions. 

Step 1. Planning

• What problems will the model solve?
• Who’s the intended user?
• What are the risks associated with the use-case?

Step 2. Data Collection, Training, and Evaluation

• How was the training data collected, sampled, and labelled?
• Is it representative of the real-world?
• Should we use existing training data to develop our own?
• Is the training data skewed?
• Is the data privacy-protecting?

Step 3. Model Design and Evaluation

• How was the model trained?
• Who trained it? When?
• How was the model debugged and improved?
• What were the model’s limitations?

Step 4. Feedback and Decision Making

• How was the model tested?
• What test datasets were used?
• Is the model stable, high-performing and safe?
• Is the model trustworthy?

Step 5. Deployment and Monitor

• Is the model behaving as expected?
• Why did the model fail in certain cases?