September 4, 2025 | In a lively matchup between clinical microbiology experts at the recent Next-Generation Dx Summit, onlookers already predisposed to believing artificial intelligence (AI) would not be replacing conventional microbiology and infectious disease testing in the next 10 years were further swayed toward dismissing the possibility. It perhaps helped that the deciding vote on the debate came immediately after the person arguing mightily about the inevitability of an AI takeover himself switched from the affirmative to the negative position.
Khosrow Shotorbani, CEO and founder of Lab 2.0 Strategic Services, cannot really be termed a turncoat. As he pointed out in closing remarks, he’s been “throwing AI under the bus” for quite a few years now and his role in the point-counterpoint debate had given him an identity crisis.
Officially on the opposing side of the issue was Steven Dallas, Ph.D., professor at the University of Texas San Antonio, who serves as the microbiology laboratory director for three nearby health systems. He repeatedly referenced the inherent complexity of microbiology with three living systems—the offending microbe, human, and microbiome—featuring a practically unquantifiable number of variables and the “MALDI divide” between the haves (big labs) and have nots (smaller labs).
MALDI, which stands for matrix-assisted laser desorption ionization, is a mass spectrometry technique used to prepare samples for analysis. Many studies have shown it to be significantly faster and more accurate than conventional methods that it is supplanting, which rely on biochemical and visual characteristics. But it is expensive for most small labs.
Large labs are also significantly more likely to use total laboratory automation (TLA) than smaller labs, but overall adoption is still not that high, says Dallas. Consolidation has been reducing the number of both core microbiology labs and smaller independent ones, while the rise of point-of-care testing means there often is no lab at all. Meanwhile, taxonomists seem never to stop changing the names of different bugs.
In his opening statement, Shotorbani introduced the idea of a future state where microbiology labs are in the business of creating value rather than volume. Diagnostics is functioning as the “meteorologist” of diseases detected at their onset, measured quantitatively and with warning of pending adverse reactions.
This stands in stark contrast to the current situation, where the “massive migration” of lab data into electronic medical records (EMRs) has failed to translate into meaningful clinical actions, he says. Despite spending about $5 trillion on 4% of the global population, “we should all be offended that longevity of life is dropping. We’re not getting our bang for the buck.”
The old-school debate was facilitated by Susan Butler-Wu, Ph.D., director of clinical microbiology at LAC+USC Medical Center and an associate professor in the department of pathology at Keck School of Medicine of USC in Los Angeles, California. The goal, she says, was to prompt both the debaters and attendees to think “more deeply and differently about the [AI] topic.”
Dallas was unceremoniously declared winner of the debate by convincing a greater percentage of the audience to change their vote on the AI matter after hearing both sides of the argument. Three-quarters of people in the room started out believing AI wasn’t an all-out threat to the field, and even Shotorbani, in his amusing yield, says everything Dallas shared was “absolutely right.”
Speaking in favor of the motion, Shotorbani preached about “democratizing data,” which in the financial world has enabled credit card companies to detect fraud and abuse within seconds. “Yet we can’t detect a clinical risk a month out? Of course we can; we just don’t get paid for it.”
In the current healthcare business model focused on getting “heads in beds,” data and therefore care is fragmented, says Shotorbani. The shift to a sustainable, value-driven future, where biomarker-based diagnostics are the highest-yielding assets, will necessitate AI technology moving the field beyond the existing “order in, results out" mindset. This transition to population health will be marked by a different supply chain and “proactive prediction diagnostics—the right care at the right time, at the right place, and at the right cost.”
Under risk-based healthcare, where providers take on financial responsibility for the quality and cost of care, knowing the progression of a condition like chronic kidney disease matters more than whether an individual tests negative or normal, he continues. Faced with a high-prevalence condition, clinical labs of the future will be using domain knowledge of pathology and longitudinal data to “proactively risk-stratify the population without an order coming in ... looking for a needle in the haystack.”
Knowing what has and hasn’t been done for high-risk patients would inform the search for gaps in care, says Shotorbani. “The whole idea in diagnostic pathology is triaging of the care—who needs to be seen by whom and when, and the measurement is all about the outcome.” While the clinical focus is on prevention and intervention, the financial side emphasizes optimization of performance improvement tools (i.e., Healthcare Effectiveness Data and Information Set, or HEDIS), payment risk adjustment (i.e., the Centers for Medicare and Medicaid Services’ hierarchical condition category, or HCC), and total cost.
“But the naked eye misses this opportunity,” Shotorbani says. “The whole idea is to apply a machine learning process and connecting the dots and turning that data into ... proactive medicine.” The clinical lab could be imagined as “the Uber of the medicine,” risk-stratifying devastating conditions like sepsis using all available data on individuals, including their gut bacteria.
As the director of a busy clinical microbiology lab, Dallas says he’s preoccupied with what’s happening today. He also points to the many false hopes and misguided predictions from the past, including one on the cover of Popular Mechanics in 1951 forecasting the rise of personal helicopters and a 1984 prophecy that PCR was going to take away culture.
Most recently, it has been suggested that sequencing would likewise eliminate culture, Dallas adds. While traditional culture methods for stool, viral, and sexually transmitted disease cultures have mostly gone by the wayside, “80% of what we do is still culture.”
The problem, as he sees it, is that “infectious disease is complicated,” says Dallas. “If you get cancer, it is ... basically that you just went wild, but in infectious disease and in microbiology there are three living systems [microbe, human, and microbiome] ... and we’re supposed to only kill one of those, but most antibiotics do a good job of killing a lot more than that.”
Referencing a simple drawing of a “Doug to bug to drug triangle,” Dallas notes that susceptibility testing in the microbiology lab can only measure those interactions in two directions but not the other four (i.e., pharmacokinetics, pharmacodynamics, cytokines, and antibodies). “At best, we’re giving a prediction and that’s why microbiology is so complicated and will not lend itself well to AI systems as they exist now.”
Microbiology is evolving, but the big, MALDI-equipped labs have the advantage, he continues. Smaller labs “either get bad identifications or they shut down.”
Available TLA instruments have effectively “automated the 20th century,” which is a good match for much of what is still being done in the lab (culture) but they have rarely been deployed in practice relative to the total number of labs in the country. ChatGPT indicates an estimated 1,300 of Copan’s WASP systems have been deployed worldwide, while no information is available on BD Kiestra, the other leading TLA system.
In a side-by-side comparison of humans and AI, people have a significantly smaller footprint since they require only personal protective equipment, outerwear, and underwear while artificial intelligence requires hardware, middleware, and software. “For TLA to be effective, it is going to need to get small,” concludes Dallas. In 10 years, AI will not replace conventional microbiology testing but rather “augment, supplement, complement, enhance, [and] improve” it.
Technological Possibilities
In the clinical microbiology lab, a decade from now, it will all begin with patients providing consent to “liberate” their personal data from its current custodians and allow its aggregation to create longitudinal data to help clinicians know best how to treat them, argues Shotorbani. “Diseases don’t happen in a snapshot ... [but] in progression.”
Labs will need to develop a new business model accommodating different customers and demand alternative payment mechanisms that reward disease management instead of analytical precision, he adds. “We say we are in the business of diagnostics ... [but microbiology] rarely ever renders a diagnosis.”
AI would be simple enough to regulate in terms of what TLA is able to accomplish now, says Dallas, notably pattern recognition in Gram stains and parasite identification using a microscope and the ability to recognize MRSA (methicillin-resistant Staphylococcus aureus) on a plate. These are media products that can be approved by the Food and Drug Administration, with machines rather than human bench techs ultimately doing the work.
“If AI can get to the point where it can flip some plates over and some machine could look at it and ... [correctly identify] E. coli and strep and their variants” and from which body site, “a crazy kind of regulation” is going to be needed, Dallas says. “Honestly, I don’t know if that can be regulated.”
The timeline and trajectory for innovation in this space are, admittedly, hard to predict. But if 80% of infections are caused by only a subset of microbes, an argument might be made that this identification work would be a good starting point for AI, says Dallas. The problem is that a lot of data are nuanced, one example being a culture that contains only “a little” Streptococcus pneumoniae. Another important clue to render the diagnosis would be a high white blood cell count infection, especially if the patient also has symptoms, and all that information would need to be interconnected.
Technology “enables but does not disrupt,” interjects Shotorbani. “Uber did not become so big because it came up with a fancy app ... [but] because it was able to normalize, or democratize, the business of logistics.”
A Six Sigma process in the clinical lab reveals almost 3.4 errors per million opportunities, he says, which is nearly perfect. Yet without a new value-based business model, the treasure trove of data in EMRs cannot further optimize the performance of labs in ways that will keep them relevant.
Smaller labs “just can’t compete on any scale,” Dallas says. “The only competition they have is that they’re there and they have pride.” Here, he refers to a lab based in the last little hospital on the coast of North Carolina, the others being gobbled up by Novant or Atrium healthcare system. Their only choice, other than to be acquired or shut down, is to continue doing biochemical identifications and susceptibilities that “we know are not as good as MALDI.”
Unlike chemists, who can learn their job in days or weeks because of the heavy automation, it can easily take microbiologists months to years, Dallas adds. They might see one of everything in five years.
His 10-year outlook is not bright and includes little labs giving up as well as a surge in point-of-care testing with samples sent to central labs for processing, creating delays in care due related to transportation delays and specimens that need to be recollected. “The big people with technology are only going to get bigger,” he says.
“The future product is really not a test,” Shotorbani counters. “It’s a way to detect and mitigate [future] risk. If you are going at risk for someone ... you need insurance.” Conventional practices aren’t going to disappear because microbiology testing needs to continue, he adds.
But AI won’t help save smaller labs, says Dallas, unless it is an “equalizer like the cell phone.” For most of the major innovations of our time, including Uber and Airbnb, no one—taxi drivers and hotels included—saw them coming. “So, maybe it’s something that we don’t see coming ... that revolutionizes medicine.”
“Let’s not forget what happened to Kodak,” Shotorbani says. The 100-year-old company was the first to put the digital camera on the market but “forgot” what business it was in, costing 120,000 people their jobs. “You have to recognize when you are in a strategic inflection point; the past no longer represents the future.”
Getting to the point of abandoning some traditional microbiology tests will require more than automating a lab, and Karius might suggest its genomic diagnostic test is the answer, says Dallas. The test is “essentially the liquid biopsy for microbes” and looks for nucleic acids and pathogens while detecting some resistance genes. But it is reserved for highly unique cases, with lab director approval, and is “not a perfect test.”
AI might help enable development of lab automation by connecting test results to patient records, he adds. If it is documented that an individual has been bitten by fleas from a cat, for example, the medical pathologist might thereby consider a diagnosis of murine typhus that otherwise could be easily missed.
In the short term, Shotorbani offers, health systems could start harnessing metadata, diagnostic classification codes, and longitudinal data to “orchestrate their risk.” The opportunities around AI for data analytics are greater than the potential with biomarkers, and the prospect of personalizing test results by better understanding the microbiome is still a decade or more away.
Theoretically, AI could synthesize all that data to find a director-level role in the lab of the future, says Dallas, but that would require sharing protected health information with an outside entity. It will “probably be more than 10 years” before susceptibility testing will aid the prediction of disease outcomes.
Shotorbani again counters with his proposal to democratize data “above and beyond one single health system,” since people receive their care in different places. Informed consumers will demand that this happens by providing the necessary permissions. The question becomes who the aggregator will ultimately be, since the major EMR systems aren’t doing that work. “We need to have a business model around it, just like what happened with e-commerce and banking.”
This assumes patients are “even able to express their rights to sign permission,” says Dallas. “I work at a hospital where at any given time we have 10 [acutely ill] John Does ... people who are just scooped up off the street [and] maybe don’t speak English.” These unfortunate souls have “no patient history at all.”
With or without an AI takeover, the microbiology lab is universally accountable for the accuracy of its results, Shotorbani says. “AI just makes it overall more efficient ... regulation doesn’t just go away because AI has appeared.”
“The lab director will be held accountable for any result that comes out of the micro lab,” agrees Dallas, in addressing a question about patient consent in the context of bankruptcy as happened recently with 23andMe. Signing up for a DNA test can turn up surprising findings, including for Dallas that he has a half-brother 30 years his junior. AI also has a dark side and, much like a firearms and drones, “will be weaponized.”
In the hypothetical scenario where AI has replaced conventional microbiology and infectious disease testing, the “backup” is to revert to traditional approaches as currently happens with MALDI, he says.
“The good thing right now is that medical lab sciences programs don’t have MALDI, so they have to teach traditional biochemicals.”
What makes microbiology so hard is having to know how to do it both the old and new way, says Dallas. “We’re like hoarders; we never get rid of our old stuff. Anybody who is a microbiology lab director knows that.”
The wet lab is “here to stay until who knows how long,” Shotorbani says. AI is just the means for turning wet lab data into action.
What could theoretically go away in the microbiology lab are urine and blood cultures, two of the highest-volume tests that are the financial foundation of the lab business, says Dallas. People will always be needed for the “weird 20% on the edges” involving fungus, bacteria, and parasites that AI has never seen.
It is only a matter of time before AI is doing pattern recognition work in the microbiology lab as it is already doing in digital pathology with pixel-based data, says Shotorbani. But what’s going to be particularly disruptive is capitation where “there is no longer economics for a single unit of a test,” which has only encouraged gaming of the system purely for monetary gain.
Despite his no-confidence argument for AI, Dallas ends by pointing to a “brave new future” that includes legitimate concerns about potentially huge layoffs facing bench-level scientists in the lab. “Maybe we can all lose our jobs because AI can replace us,” he says with a degree of resignation.