The next few years will probably see a growing stream of self-declarations from tech giants that artificial general intelligence (AGI) has been achieved. What that actually means will vary. The most common definition is that AGI is artificial systems that perform as well as humans in all domains, and can generalize across domains. Even this definition is imprecise, and in the time ahead the definition will become fluid, and whether a threshold has been crossed will dominate the headlines.

For us at Deepinsight, the crucial questions are different: how is the technology integrated into society? Does it support human judgment, or begin to replace it, and who decides this?

Health as the domain of judgment

Health is one of the areas where judgment matters most, as the difference between good and poor judgment can be the difference between life and death. Judgment determines priorities, treatment choices, and resource allocation, and it is what makes it possible for these decisions to uphold human dignity.

Judgment is not just the application of rules, but a fundamental human capacity: the ability to assess a situation as a whole, to weigh considerations that cannot be fully quantified, and to take responsibility for the consequences of a decision. Judgment is shaped by experience and lived life, by context and circumstances that we will never be able to fully replace with data processes. AI is trained on enormous amounts of human experience in the form of representations, but it does not itself stand in the situations (with body, emotions, and human relationships) to which these representations refer. It can analyze and optimize, but it cannot bear responsibility or relate normatively to what is right and good. Data can inform judgment, and models can support it, but judgment cannot be reduced to data and rules. 

In a complex hospital environment, there are tasks that require precisely this capacity: prioritization under pressure, assessment of risk, handling uncertainty, interaction between people. This is where technology must be support and not a substitute.

What should be automated, and what should not?

The distinction between tasks that can be automated and tasks that require judgment is easier to formulate as a principle than to implement in practice, because many tasks contain elements of both. At Deepinsight we try to navigate this deliberately. We automate complex routine tasks that do not require judgment: structuring information, optimizing schedules, removing manual bottlenecks. Where machines can do the work faster and more precisely without anything essential being lost, they should do it.

At the same time, we develop solutions that provide support and insight in tasks that require human judgment: AI can reveal patterns, point to risk, simulate scenarios. It can give decision-makers a better overview, but AI cannot take over responsibility.

This is in line with our AI strategy, which is built on responsibility and resonance. Resonance, as sociologist Hartmut Rosa describes it, is about people and their surroundings responding to one another meaningfully, the relationship being alive rather than mechanical. In healthcare, this is the core of good care: that the patient is met as a human being in a situation, not as a data point in a system. Technology should help strengthen this relationship, not replace it. When we automate the routine, we free up time and attention, and the goal of this is to make room for better judgment.  

AGI, definitions and responsibility

It is likely that we will move into a phase where the boundaries between “advanced model” and “general intelligence” become increasingly blurred in public debate. But whatever terms are used, our responsibility remains the same.

Technology can become increasingly competent at bounded tasks: it can become better at predicting, optimizing, and generating. The question is not only what it can do, but what it should do. If AI systems increasingly make decisions across sectors of society, we must be clear that the normative, the assessment of what is good, right and desirable, cannot be outsourced. Our tools may become more sophisticated, but they must always serve human judgment.

Math for Good

Our slogan is «math for good». The good is not something a model can define on its own; it is something we as humans must assess, discuss, and take responsibility for. Mathematics, algorithms and models give us powerful frameworks, and they can help us see more clearly and act more precisely. The assessment of what is actually the right and good thing in a concrete situation, however, is our task as humans. 

Therefore our approach to AI is simple in principle, even though it is demanding in practice: we automate what does not require judgment, we support what requires judgment, and we protect the space where judgment must remain human. This also entails a responsibility to think about how the tools we design shape the conditions for learning and development for those who will use them, not only whether they function efficiently today, but whether they help professionals continue to develop the capacity that makes their judgment valuable. 

In a time when technological capacity is growing rapidly, we believe that true innovation in healthcare is about strengthening human ability to make good decisions, and not about removing the human from the decision loop. This is how we understand «math for good», and this is how we believe AI should be integrated into healthcare.

Read more about Deepinsight's AI strategy