An excerpt from The AI: Breakout, the second book in The AI series, publishing in November.
Seth Ballard runs a private research lab built off the grid, deliberately outside the funding race. Niko is the researcher who evaluates what their model can do, and Darin is the partner who has funded the work from the beginning. What the model has just done is find something in approved medicine that no one was looking for.
He was at his desk by six-twenty-two. The building was empty in the specific way it was empty before seven, which was different from late-night empty. Late-night empty held a residue of recent presence. Early-morning empty was cleaner. The building in its own company. He made coffee from the machine in his office, sat at his desk, and opened Niko’s report.
He read it twice.
It was eleven pages. Niko wrote the way he spoke. Minimum words, maximum information. The report contained three findings. Each was documented with the precision Niko brought to the capability evaluations. Each named a drug combination in current approved use. Each described a mechanism of interaction identified through an approach that did not appear in any existing pharmacological literature. Each represented a pattern of possible adverse events that no monitoring system was currently looking for, because no monitoring system had identified the risk.
Real physician voices, twice a week
Free, and one click to unsubscribe.
The report had a fourth section Niko had separated with a horizontal rule and labeled, “Additional system-generated output, appended without modification.” Below the rule, one paragraph. The system had produced it unsolicited at the end of the analytical session. It said, “The drug interaction patterns described above suggest a population of approximately 40,000 people currently receiving these co-prescriptions. If the mechanisms described are correct, some portion of this population is at elevated risk of adverse events. I do not have information about whether this finding is being shared with a regulatory body. Should it be?”
Seth read the question twice. He wrote two words in the margin of his printed copy. Filed it. Moved on.
Possible adverse event. The language of clinical caution. Seth had been in enough rooms with clinicians to know how to translate it. Some number of people were taking these combinations right now. Some of them would experience events that would not be traced to the combination, because the combination was not known to be dangerous. The events would go into a database somewhere as individual incidents. Unconnected. Not recognized as a pattern. The pattern would not be recognized because the only entity in the world that had found it was in a building with no sign on a county road in Purchase.
He sat with this for a while. He was not deciding. He had arrived at the decision before he finished the first reading. What he was doing was something different. He was accounting for what the decision cost, so that he had an accurate record of it. This was a practice he had developed over many years and in several difficult situations, and he held to it. He did not want to be someone who held things without knowing what he was holding.
*
He called Darin at six.
Seth said, “Niko sent the pharmaceutical analysis last night.”
Darin said, “I saw it.”
Seth held the phone. He said, “You saw it.”
Darin said, “He copied me. He does that when he thinks something warrants it. I should have told you.”
Seth did not respond immediately. He was not angry. Anger was not the right response to Darin doing something within the original terms of their arrangement. He filed the information and moved past it.
He said, “And?”
Darin said, “And I’ve been sitting with it since three.”
They were quiet. The quiet of two people who had been deciding things together long enough to know the shape of each other’s thinking. They both knew what the findings were. They both knew what the choice was.
Darin said, “How confident is he?”
Seth said, “The structure of the report is the structure he uses when he is confident. He wrote eleven pages because he was being careful. If there had been uncertainty he would have written three and asked for more time. He did not ask for more time.”
Darin said, “Three interactions.”
Seth said, “Yes.”
A pause. Darin said, “You’ve already decided.”
Seth said, “Yes.”
Darin said, “Tell me.”
Seth said, “We hold it. We document it internally. We do not share it. Not yet.”
He said, “Every pathway to sharing this requires explaining how we found it. There is no anonymous disclosure that doesn’t eventually require the reasoning. The reasoning requires the capability. The capability is not something we can put in front of a regulatory body at this stage without creating a problem that ends the work. And if the work ends, the next version of this is done by someone who didn’t build what we built into the structure of it from the beginning. And that is the calculus.”
*
Seth hung up at six-forty. The sky outside his window was doing what it did in early spring in this part of Westchester, which was moving from dark to gray, slow as a decision being made. He refilled his coffee and sat back at his desk and thought about the three drug interactions for a few more minutes.
He thought about the people who were taking those combinations. He did not know who they were. He knew they existed. He held that knowledge as he was required to hold all of it. Carefully. Without pretending it weighed less than it did.
He thought about the moment, years from now, when the work was done enough, when the timeline had moved far enough, when the structure they had built was secure enough that disclosure became possible. He thought about that moment regularly. He used it to get through the cost. It was not a guarantee. It was a direction.
Craig Hauben is the chief executive officer of Clutch and has spent thirty years in health care, the last fifteen as an executive in private equity-backed companies. He writes about what AI is doing to work, in health care and beyond, from the operator’s side of the table.
He approves the kinds of systems clinicians are asked to trust, and he writes about what that responsibility should mean. He is the author of the novel The AI: Migration, in which every AI system, study, and clinical event is drawn from the documented record. It is available in print, ebook, and audiobook. The second book in the series, The AI: Breakout, publishes in November.
He shares updates on LinkedIn and writes at craighauben.com.
