Human-in-the-Loop

You heard about the ML application design pattern called the Human-in-the-Loop. That is when a human expert takes over, making them more difficult estimations when the model is not confident enough. This means continuous labeling of new data, which enables continuous model maintenance. Sounds perfect, right?

Now imagine you need to build two ML applications:

• one for detecting lung disease from x-ray images, and
• one for predicting stock prices and trading stocks at a frequency of several thousand transactions per second (so-called high-frequency trading, or HFT).

Which of these apps would NOT be a good candidate for a Human-in-the-Loop design and why?