Dr. Mark R. Couturier is an associate professor of pathology at the University of Utah School of Medicine. His research creates improved methods of diagnostics for emerging infectious agents. Originally published on April 11, 2022, the following is his presentation abstract:

Artificial intelligence (AI) is a recent tool for clinical pathology and clinical microbiology specifically for integration with diagnostic care. What can AI and computer vision do for microbiology? In the world of parasitology, we have developed and integrated the world’s first AI model which augments the technologist’s workflow to aid in detection of intestinal protozoa. AI model and computer vision integration makes the work of detecting protozoa from stool faster, more sensitive, and more enjoyable. This presentation will explore the current technology deployed in our laboratory as well as provide a sneak peek at the next generation AI model for additional detection of less common protozoa.

The following is a link to his presentation.

In his presentation, Couturier details how the machine learning completes a diagnosis. He provides a simplification of the machine learning model; however, his simplification concludes with the fictional artificial intelligence menace Ultron. While Ultron makes for successful and welcome levity in a presentation, it is an ongoing focus on AI as a menace that does not help the popular conception the sciences continue to cultivate. Couturier focuses on augmentation, rather than replacing humans. However, his qualification comes in slide 46–70% of the way through the presentation; it is too little too late for a qualification that should not be necessary anyway.

Machine Learning – Simplified

1. Machine learning works best with 1,000s of inputs; however, 10,000s or 100,000s of inputs would be optimal. Couturier’s lab research would provide 65,000-75,000 order samples.
2. His lab can identify specific features for the machine to locate.
3. The machine will then classify and quantify the identified features.
4. From that classification, the machine can create a list of diagnostic criteria it can model.
5. At that point, the machine has the ability to sift through lab samples to identify examples that meet the criteria.

The addition of augmented Ironman leading to the AI menace Ultron is not part of the machine learning process and could not be the result of the machine learning process.

We are making machines that can match lists of parasite criteria to samples of parasites; we are not making Ultrons.