The Human Threshold

When we talk about artificial intelligence, sooner or later two concepts come up: Artificial General Intelligence, or AGI, and superintelligence. At first, these terms seem fairly straightforward. AGI describes an artificial intelligence that reaches at least human-level performance across a broad range of cognitive tasks. Superintelligence goes beyond that, significantly surpassing human capabilities.

But even this seemingly simple definition raises a problem: Which human are we actually talking about?

People differ considerably. We do not all possess the same abilities, the same knowledge, or the same ways of thinking. A mathematician may solve complex abstract problems that others cannot. An experienced physician may recognize connections that remain invisible to a layperson. A musician may perceive structures that someone without musical training might not even notice. At the same time, a person can be exceptionally capable in one area and entirely average in another.

Of course, we do try to make intelligence measurable and comparable. IQ is the best-known example. It condenses certain cognitive abilities into a single score, allowing for standardized comparison. But this is precisely where the difficulty becomes apparent: an IQ score does not represent the full spectrum of human capabilities. It captures certain forms of cognitive performance, but not everything that shapes how a person thinks, acts, and solves problems. Expertise, creativity, social abilities, experience, or context-dependent judgment cannot simply be reduced to the same number.

IQ therefore does not fully solve the problem of finding a human reference point. If anything, it illustrates just how difficult it is to reduce something as multifaceted as human intelligence to a single measure.

So if AGI means that an artificial intelligence is “as intelligent as a human,” the immediate question is which human sets the benchmark.

This is where it might make sense to think of AGI not exclusively as a fixed, universal threshold, but at least partly as a relative concept.

Intelligence Has No Single Human Reference Point

Imagine an artificial intelligence that was better than me at practically every cognitive task. It could learn faster, understand more complex relationships, write better texts, make more precise decisions, and familiarize itself with new subjects more easily. From my personal perspective, this artificial intelligence would be superior to my own general cognitive abilities.

For someone else, however, the situation might be different.

A highly specialized scientist might still outperform the AI in their field. Someone with exceptional social abilities might be better at interpreting interpersonal situations. Others, meanwhile, might already be surpassed by the very same AI in nearly every area relevant to them.

This leads to an interesting possibility: The point at which we subjectively perceive an artificial intelligence as generally intelligent could differ from person to person.

This perspective does not begin by asking which abstract technical threshold a system has crossed. It begins with the human being and asks what the capabilities of that system mean in relation to their own abilities.

From this perspective, AGI would not merely be a property of the system itself. It would, at least in part, describe a relationship between the capabilities of a system and those of the human being facing it.

We might therefore say:

An AI can achieve general superiority relative to a particular person before it would qualify as AGI relative to humanity as a whole.

This does not necessarily mean that the conventional definition of AGI is wrong. Rather, it suggests that the term may be blending two different levels: an individual one and a societal one.

Personal AGI and General AGI

To make this distinction clearer, we could think in terms of a personal threshold and a general threshold.

A “personal AGI” would be reached when a system matches or exceeds a particular person across nearly all relevant cognitive abilities. For different people, that threshold would be reached at different times.

A societal or general AGI, by contrast, would have to be defined much more rigorously. It would need capabilities comparable not merely to those of a single individual, but to a broad spectrum of human abilities.

That changes the question.

Instead of asking:

“Is this AI as intelligent as a human?”

we should really be asking:

“Against what distribution of human abilities are we comparing it?”

Perhaps a formal definition of AGI should therefore not rely on a single human reference value at all, but on something more like a capability space. A system would then qualify as generally intelligent if it reached a defined level of human performance across many different cognitive domains.

Even then, however, the question would remain which level to choose: the population average, particularly capable people, or even the best specialists in each respective domain?

Depending on the benchmark we choose, the point at which AGI is reached could shift dramatically.

Superintelligence Works Differently

At first, superintelligence appears to face a similar problem. After all, we still have to define what “superhuman” actually means.

But the concept introduces a crucial shift.

Superintelligence does not normally describe a system that merely surpasses individual people. Rather, it refers to an intelligence that leaves human capabilities as a whole far behind.

The reference point is therefore no longer the individual human being, but the limits of human performance.

An artificial intelligence would not be superintelligent simply because it outperformed me in mathematics, writing, planning, and analysis. Many current systems could already do so in individual domains.

Superintelligence would mean that a system is superior even where humans achieve their highest known levels of performance.

It would have to compete not with the average, but with the boundaries of human capability.

That makes superintelligence considerably less relative.

There can still be debate about which capabilities should count and how great the superiority would need to be. But fundamentally, the comparison is no longer centered on the individual.

We might therefore summarize it this way:

AGI can be relative. Superintelligence must be defined in much more absolute terms.

The Transition Would Not Be a Single Moment

This perspective has another consequence.

In public discussions, AGI is often treated like a technological milestone: on a particular day, or with a particular generation of models, AGI is achieved.

Perhaps that idea is too simple.

From a human perspective, a technological threshold does not necessarily coincide with the moment when that technology takes on a fundamentally new significance for an individual.

If intelligence is multidimensional and people differ in their abilities, AGI might emerge less as a single threshold and more as a transition zone.

First, AI systems become superior to individual people in an increasing number of domains. Then they reach average human performance across a wide range of tasks. Later, they may reach exceptionally capable humans. And eventually, perhaps even the best human specialists.

Depending on the definition being used, one person might already speak of AGI while another still considers the term inappropriate.

Within their respective definitions, both might even be right.

AGI would then be less like a distinct point on a timeline and more like an increasing overlap between human and artificial capabilities.

A Possible Counterargument

There is an important objection to this relative view.

Scientific concepts should, as far as possible, be defined independently of the observer. If AGI means something different for every person, the term risks losing its usefulness.

That is a valid objection.

A technical definition of AGI ultimately needs objective criteria. Otherwise, two people could classify the very same system in completely different ways.

The individual perspective should therefore probably not replace a formal definition.

But it can explain something else: our perception of AGI.

For one person, technological change may feel fundamental relatively early because AI surpasses a large portion of their own capabilities. Another person may experience the exact same technological state as far less transformative because their particular abilities still lie outside what the system can do.

This creates a distinction between technical and personal AGI.

Technical AGI would be a societally defined performance standard.

Personal AGI would be the point at which a system matches or surpasses one’s own cognitive capabilities across a broad range of domains.

Both concepts describe something real, but they do not describe the same thing.

And this distinction allows us to look at technology from the perspective of the human being without abandoning the need for objective technical definitions. The question then becomes not only what an AI can do, but also what those capabilities mean for different people.

Why This Distinction Might Matter

This distinction is not merely of theoretical interest.

It also changes the question of when artificial intelligence becomes societally relevant.

Perhaps we do not need to wait until a generally accepted definition of AGI has been fulfilled.

For a large part of the population, the practical effects of AGI could arrive earlier.

If a system can perform most of the cognitive tasks that a particular person carries out in their everyday life or profession, it may matter very little to that person whether the system is officially classified as AGI.

From their perspective, the decisive threshold has already been crossed.

This creates an apparent AGI paradox:

A society could still be debating whether AGI has been achieved while millions of people are already working with systems that surpass their own cognitive capabilities across a broad range of domains.

At first glance, this seems contradictory. How can AGI already be a reality for millions of people while we are simultaneously still debating whether AGI even exists?

The contradiction disappears once we distinguish between two different reference levels.

On one side is the technical or societal AGI threshold: the attempt to establish an objective and generally applicable standard for when a system can be considered generally intelligent.

On the other side is the personal threshold: the point at which that same system matches or exceeds the abilities of a particular person across a large part of their relevant cognitive spectrum.

What does not yet qualify as AGI technologically can therefore already have the effect of AGI for an individual.

The apparent paradox is not a logical contradiction. It arises because we are using the same concept to describe two different levels: the performance of a technology at the societal level and its impact on the individual human being.

The technical debate and the human reality could therefore drift apart in time.

Perhaps this is one of the most important consequences of a human-centered perspective: Technological progress does not become relevant only once we give it a name. It becomes relevant when it changes people’s abilities, opportunities, and roles.

Conclusion

Perhaps one of the fundamental problems in the discussion about AGI is that we are trying to define a fixed threshold for something whose reference point is itself not fixed.

“Human intelligence” is not a single value. Even when we try to make it comparable through metrics such as IQ, we capture only fragments of a much larger spectrum of abilities. Human intelligence remains an enormous range of different capabilities, experiences, and specializations.

It therefore seems reasonable to think of AGI, at least in part, as a relative concept.

An artificial intelligence could already have achieved general superiority relative to one person while remaining far from doing so relative to another. Only at the societal level do we then attempt to construct a shared threshold from those differences.

With superintelligence, the benchmark changes.

Here, surpassing individual or average humans is no longer enough. The comparison increasingly shifts toward the greatest capabilities humans are able to achieve at all.

This creates a conceptual asymmetry:

AGI can be understood relative to the human being. Superintelligence must be measured against humanity.

And perhaps that is the crucial shift in perspective: rather than asking exclusively from the standpoint of technology when it crosses a particular threshold, we can ask from the standpoint of the human being when their relationship with that technology fundamentally changes.

Perhaps there is no single moment when AGI suddenly comes into existence.

Perhaps instead, we cross millions of individual thresholds — one by one — before eventually looking back and realizing that the collective threshold had already fallen.