Overview
The landscape of neurology is being transformed by what some call the "Diagnostic Oracle(opens in new window)." Companies like Neuropacs(opens in new window) (recently FDA-cleared(opens in new window) in April 2026) are using AI to differentiate between Parkinson’s(opens in new window) and much more aggressive syndromes like Multiple System Atrophy(opens in new window) (MSA) with near-perfect accuracy using MRI(opens in new window) data. Meanwhile, tools like Rune Labs’ StrivePD(opens in new window) use wearable data from Apple Watches to provide real-time AI guidance to Parkinson's patients, effectively acting as a digital care partner that adjusts medication schedules based on algorithmic off-time predictions. The promise is precision medicine(opens in new window); moving away from wait and see, reactive treatments to proactive, data-driven interventions that may prevent the onset of these diseases (or at least delay them and soften the impacts).
However, the Pierian spring(opens in new window) of medical AI is fraught with the intoxication of early detection and perfect treatment outcomes. New AI agents are now capable of predicting Alzheimer’s onset up to a decade before the first memory slip by analyzing subtle linguistic patterns or retinal scans. This creates a harrowing ethical black box: if an algorithm identifies a 92% risk of a terminal, neurodegenerative decline in a 40-year-old asymptomatic individual, has the AI ceased to be just a tool and instead become a partner in the patient's identity? Because these models are often opaque, the physician cannot explain how the AI knows; they can only deliver the machine's verdict.
The reality for the junior software engineer is that you are often the one setting the certainty threshold in code for these verdicts. In the case of IBM Watson Health(opens in new window), high-profile failures in oncology demonstrated that algorithms trained on codified versions of one elite hospital’s methods often failed(opens in new window) when applied to the messy, variable reality of global patient data. In neurology, where symptoms are deeply subjective and progression is non-linear, a shallow draught of data can lead to devastating false positives, e.g., labeling a healthy person as pre-symptomatic, or false negatives that delay critical end-of-life planning.
==============================================================================================According to the US CDC National ALS Registry Dashboard(opens in new window), Amyotrophic Lateral Sclerosis (ALS) affects about 35,000 persons in the US in 2026, or about 0.01 % of the population. This low prevalence puts ALS in the category of "rare diseases," those that affect less than 200,000 people in the US. Life expectancy is generally limited, according to the Muscular Dystrophy Association(opens in new window):
From the time of diagnosis, most individuals will live for three to five years. Around 30% of ALS patients are alive beyond five years after diagnosis, and 10-20% of patients survive 10 years or more. Survival beyond 20 years is possible but rare.
Eighteen months after graduating from NC State, you are a software engineer working for a healthcare startup that is developing an AI system that analyzes speech patterns(opens in new window) to predict ALS. This system is not intended to be used in a broad, general-purpose fashion, but as one of several different types of diagnostic tools that physicians use to evaluate an asymptomatic patient's likelihood of developing ALS. You are working on the team that tunes and refines the model's inferences from training and user data to generate predictions of the potential to develop ALS at some time in the future. Your company has access to a vast amount of highly diverse, anonymous data from over 10 million patients with detailed healthcare records from the US and several other countries, covering an average of 7 years per patient with ALS and 26 years per patient without a diagnosis of ALS.
Your company has been approached by one of the largest health insurance companies in the US who would like to use this system as one of many components making up a "Wellness Screen" on their patient portal. They envision this Wellness Screen as a means of presenting a holistic view of a patient's health that will combine information from the patient's doctors and other providers with lab results, pharmacological information, clinician notes, and published research results.
Even with the quantity and quality of data your company has used to train the ALS prediction model, you know that its false-positive rate is 4%, meaning 4 out of every 100 otherwise healthy people who are evaluated with this screening tool will be told they are likely to develop a terminal, paralyzing disease for which there is no cure. This also means that 96 out of 100 asymptomatic patients who are screened will receive results sooner than other forms of testing. While their prognosis may not change, they will have the opportunity to take advantage of additional time before the disease progresses to a more severe and debilitating stage. Despite the volume of training data available, your team has not been able to improve on the accuracy of the model, but you all believe that it is a tool that should be available to physicians and their patients.
Your task for this forum is to respond to this question:
Do you, as a software engineer working on this AI system, have an obligation to prevent the harm of providing this information altogether, or does the right to know for the 96% who will receive an accurate result outweigh the potential harm to the 4% who will receive a false positive result?
For this forum, clearly state your response to this question. Your response must argue your position from one of the following two ethical perspectives:
- A Rights-based view that compares the moral rights of individuals who are likely to develop ALS with the moral rights of individuals who are not likely to develop the disease.
- A Duty-based view that compares the moral duties that you, as a software engineer, have to individuals who are likely to develop ALS with your moral duties to individuals who are not likely to develop the disease.
Your argument from either perspective must be supported by concrete, verifiable evidence from appropriately referenced external source(s). Note that there is a wealth of information on ALS available from reliable sources. When you reply to your colleagues, look for responses that take a different position from your own and build your response to compare your points of view.