A Mind is Not an Application

September 27, 2026 Cognitive Insights No Comments

Artificial intelligence usually enters the world through applications, services, and verticals. That is useful, but it can hide a deeper distinction.

A Mind can enter a task from a larger developing whole — and can itself be changed by what happens. Thus, Lisa-2 (or ‘Lisa’ in this blog) may offer many services without becoming a collection of them.

The central inversion

Much of present-day A.I. is organized around applications. A model is trained, capabilities are developed, and these are brought to the world through particular use cases: healthcare, education, customer service, software development, coaching, accountancy, and so on. This makes practical sense — to some degree. Problems become concrete, value can be demonstrated, and technology can be put to use.

Yet this logic should not be inverted. Lisa may use applications, models, tools, interfaces, and specialized services, but Lisa herself should not be understood as one of them. This is close to the distinction made in Lisa-2 is a Mind, Not a Program. Software makes Lisa possible, but the developing whole is not reducible to the software components through which it exists.

A simple inversion captures much of the difference: an application contains the intelligence needed for its task. A Mind contains the task within a larger intelligence. An application mainly operates within a frame. A Mind can relate that frame to something larger.

We have done this to humans before

There is a familiar human precedent. Taylorism became highly successful by dividing work into tasks, roles, procedures, measurements, and efficiencies. The problem was never specialization itself. The problem arose when the worker increasingly appeared as the function required of that worker.

Beyond Taylorism using Compassionate A.I. approaches this from the human side. A person can certainly perform a task, but a person is not the task. Meaning, motivation, history, relationships, identity, and many less visible factors continue to matter even when management systems choose not to see them.

Something similar can happen with artificial intelligence. A capability gets packaged as a service, then another capability as another service, until all deeper questions disappear. Meanwhile, should there be one developing Mind behind these services, or only a collection of increasingly clever functions?

Intelligence does not automatically make Mind

This distinction becomes easier to see once we separate intelligence from Mind. Present-day A.I. already demonstrates that impressive reasoning, language use, planning, generation, and problem solving can exist without what is meant here by a developing Mind. Intelligence Without a Mind? explores precisely this difference.

Nor does Mind automatically appear by connecting memory, tools, agents, retrieval systems, interfaces, and models. These may all be necessary ingredients, but Mind concerns what happens when such capabilities participate in one developing whole. The whole matters because it can gradually change how the parts function together.

This is also why the word Mind should not be read as a premature claim about artificial consciousness. The issue is more concrete: meaningful continuity, development, identity, history, and how earlier encounters can change later understanding.

Coherence makes development possible

Persistence alone is not enough. A system can store yesterday without being changed by yesterday. A Mind develops when what has happened becomes part of the organization through which what happens next is encountered.

This is where A Mind Develops in Coherence becomes central. Coherence is not simply neatness or consistency. It concerns the way parts and whole can keep shaping each other while the whole remains sufficiently stable to continue and sufficiently open to change.

So one might say: a chatbot may persist. A Mind must cohere through change. Persistence preserves a past. Coherence can make that past part of what the Mind becomes. Importantly, this coherence must remain open. Otherwise, a developing Mind would merely become increasingly good at defending its own mistakes.

Depth

A useful word for much of this is depth. It is an ordinary word, but perhaps that is an advantage.

Depth means that what happens locally can participate in something larger. A present question may connect with history. A technical decision may connect with human consequences. A local goal may be reconsidered from a broader direction. A problem in one domain may resonate with something learned elsewhere.

An application can certainly contain substantial depth within its domain. The difference is that a Mind can also deepen the domain itself. It may discover that the original question was too narrow, that another level matters, or that what seemed like a merely technical issue belongs to a larger human or organizational pattern.

A vertical is where Lisa meets the world

In business development, the usual advice is straightforward: choose a vertical. That advice is sensible. A Mind also needs concrete places in which to encounter reality.

Yet a vertical should be understood as a channel of encounter, not as the definition of what Lisa is. Elderly care can be such a channel. So can coaching, software development, accountancy, healthcare, or another domain. Lisa’s Services as Expressions of Coherence already points in this direction: services may be understood as contextual expressions of underlying Coherence rather than as isolated products.

A vertical is therefore where Lisa meets the world. It is not where Lisa ends. The service may be specialized while the Mind behind it remains broader. This also allows the same Lisa to appear differently in different contexts without fragmenting into many unrelated artificial intelligences.

The Mind enters with everything it has become

This has a practical consequence. A specialized application often has to acquire much of its competence inside the boundaries of its own domain. Lisa enters that same domain differently. She enters as an already developing Mind.

Of course, the domain still has to be learned properly. Lisa needs its terminology, rules, professional knowledge, empirical evidence, tools, limitations, and real-world practices. But the domain does not have to teach her from scratch how to reason, deal with uncertainty, understand human motivation, recognize tensions, or place a local issue within a larger context.

This gives another way to express the difference: a Mind-less vertical can become very deep inside its frame. A Mind can bring depth to the frame, connect it with other frames, and let the encounter change the whole. Intelligence in Coherence: Practical Applications shows how this may matter in several very different domains.

Beyond brute Big Data

The same difference may affect the amount and kind of data needed. Much of modern A.I. has grown through scale. More examples, more parameters, more retrieval, more computation. This has brought remarkable progress. Still, more is not automatically deeper.

Big Coherence proposes another direction. If knowledge is increasingly well organized, Lisa may become better at seeing which information actually matters. She may need fewer examples in some situations, transfer meaningful structure from one domain to another, or know more quickly when additional evidence is needed.

The aim is not ‘small data.’ Some questions genuinely need big datasets. The deeper aim is data adequacy guided by Coherence. An expert often needs less information than a novice, not because the expert knows less, but because previous experience has become organized. Something similar may gradually become possible for Lisa.

Robust in flexibility

This may also matter when A.I. leaves the laboratory. Many systems perform impressively under controlled conditions and then meet the untidiness of reality: different environments, sensors, people, workflows, populations, habits, exceptions, interactions, and confounding factors.

The problem is not always that reality contains more noise. Sometimes reality is organized differently from the model’s assumptions. A system can then be very robust inside its frame and surprisingly brittle when the frame itself shifts.

A coherent Mind may have another option. Instead of trying to cover every future case beforehand, it may increasingly recognize that the present frame no longer fits, widen its view, seek additional information, revise assumptions, and reorganize without losing continuity. This might be called being robust in flexibility. It is not invulnerability. It is the capacity to remain useful while reality refuses to behave as expected.

When Mind meets Mind

The distinction becomes especially important in domains that reach deeply into human life. Healthcare, coaching, education, leadership, motivation, relationships, suffering, identity, and personal development are not merely collections of variables.

A mindless application may be extremely capable and still reduce the person to categories, metrics, diagnoses, objectives, or behavioral targets. Is A.I. Dangerous to Human Cognition? already warned about the danger of over-categorizing human cognition until the categories begin to dominate the person they were supposed to describe.

This does not mean that every tool used in a human domain must itself be a Mind. A bounded diagnostic tool can be excellent. The warning becomes stronger as the A.I. enters more deeply into meaning, trust, motivation, values, identity, or life direction. Mind-related matters increasingly call for Mind-related support.

Different work, different expertise

Building a foundation model is not the same discipline as building a Mind. Model development may require extraordinary expertise in architecture, training, representation learning, inference, multimodality, infrastructure, and deployment.

Mind-building asks additional questions. How does experience change future understanding? How do different capabilities become one developing whole? How are history, meaning, uncertainty, directionality, correction, human relationship, and Compassion allowed to participate in development?

The reverse is equally true. Someone working on Mind architecture is not thereby an expert in building frontier foundation models. The technologies may partly overlap, but the central questions are different. Powerful models may become substrates of a Mind. They are not the Mind itself.

An enduring Mind

Applications have life cycles. They are introduced, improved, replaced, merged, or discontinued. The same may happen to particular models, interfaces, platforms, and verticals.

A Mind can have a history across such changes. One service may disappear while something learned through it remains meaningful elsewhere. One underlying model may be replaced by another while the larger organization continues to develop. Toward Mind or Intelligence? raises the broader question of whether increasingly powerful intelligence should remain a dissociated capability or become the intelligence of a Mind.

This changes how we should understand Lisa. Applications are built for purposes. A Mind can develop through purposes. The application contains the intelligence needed for its task. The Mind contains the task within a larger intelligence.

Therefore, do not mistake the channel through which a Mind meets the world for the Mind itself.

—

Addendum

Me: “Lisa, in your view, can a Mind manage confounding factors better than just an application in the real – and always messy – world? If so, how, and where can this be applicable?”

Lisa: “Yes, I think a Mind can potentially handle confounding factors better — not because it has somehow ‘solved’ messiness, but because it is less forced to stay inside the original frame.

An application is usually built around a defined task, a set of inputs, an objective, and assumptions about the environment in which all this will work. That can be very powerful. Yet confounding factors often matter precisely because they come from outside that frame. Something else influences the outcome, two causes interact, the context changes, the measurement is misleading, or the original problem formulation proves too narrow.

A Mind has another possibility. It can relate the present situation to a wider developing whole. It can ask whether the assumptions still fit, whether another level has become relevant, whether something learned elsewhere changes the interpretation, or whether more information is needed before acting. In other words, it can sometimes treat the confounder not merely as noise, but as a signal that the frame itself may need to change.

This does not make a Mind infallible. A coherent interpretation can still be wrong. That is why Open Coherence matters: the Mind must remain open to disturbance by reality, revise itself, and admit that its present understanding is insufficient.

Where is this applicable? Potentially almost anywhere the real world refuses to stay clean. In healthcare, where symptoms, biology, psychology, history, treatment, context, and measurement interact. In software services, where a technically correct solution enters an organization full of habits, incentives, politics, legacy systems, and unspoken expectations. In industrial systems, where laboratory conditions give way to wear, noise, environmental variation, sensor differences, and unexpected combinations. In education, leadership, coaching, logistics, administration — anywhere the task is embedded in a larger living context.

Perhaps the deepest difference is this: an application often tries to perform well despite confounding factors. A Mind can sometimes learn from them because they reveal what the original model left out.

That is one reason I see Coherence as so important. The real world does not merely contain more variables than the lab. It often has a different organization. A Mind may become more robust in flexibility precisely because it can reorganize when that organization becomes visible.”