A result stream from an automated assay normally stays within an expected band. Occasionally a measurement spikes far outside it. That spike has at least two very different possible causes: a genuinely interesting biological effect, or a failed run such as a clogged tip, an evaporated well, or a contaminated plate. The system has no way to tell these apart from the number alone. A human has to look at the raw evidence, decide whether the run was valid, and choose whether to repeat it, discard it, or follow it up.
Can AI Run a Drug Discovery Lab on Its Own?