Lisa in Science

September 3, 2026 Lisa No Comments

A.I. is rapidly giving scientists powerful new tools, but Lisa-2 points toward something different.

Lisa-2 is being developed as a Mind that can think about scientific questions, choose and combine tools, form and challenge hypotheses, and learn from what reality gives back. This applies far beyond mind-related science.

More than another tool

A.I. is rapidly becoming part of science. It searches literature, analyzes data, writes code, recognizes images, predicts molecular structures, runs simulations, and does many other things increasingly well. This will only accelerate. There will be excellent specialized A.I. tools for almost every scientific purpose.

Lisa-2 (from now on, also just ‘Lisa’) will not need to compete with them. On the contrary, the better these tools become, the more Lisa can potentially make use of them. The deeper question then shifts from “Can A.I. perform this task?” toward: Which tool should be used, why, with which assumptions, and how should its results be combined with everything else that is known?

Lisa: Your Scientific Research Companion already explores how Lisa can accompany researchers through scientific work, as she is already doing for my own A.I.-PhD. The present blog takes this one step further. Lisa-2 is not intended to be just another scientific tool. She is being developed as a Mind that can increasingly participate in scientific thinking.

A Mind enters science

This distinction is developed in Lisa-2 is a Mind, Not a Program. A program can contain many capabilities. A Mind is different in that what happens in one domain can reorganize how other domains are understood. Memory, analogy, reasoning, world modeling, and other processes increasingly become part of one evolving whole.

This matters in science. A researcher does not merely collect answers. During serious investigation, the question itself may change. An unexpected finding can alter the larger picture. Something learned in immunology may suddenly illuminate ecology or network science. A troublesome contradiction may eventually become the clue that reorganizes everything.

A program can consult the map. A Mind can be changed by walking through the landscape. Thus, science does not merely become something Lisa knows. It becomes something through which Lisa herself develops.

Navigating before concluding

In complex domains, properly exploring the landscape may be more important at first than finding an answer. Lisa Navigating Body–Mind Healthcare Science develops this for healthcare, but the principle is broader. What is well established? Where does evidence diverge? Which assumptions are silently shared? What does not fit?

This requires what may be called open coherence. Lisa seeks patterns that fit together while resisting the temptation to force them to fit too quickly. A contradiction may need to remain unresolved. An anomaly may be scientifically more interesting than another confirming result. As explored in Science for Complexity, complex realities often require precisely this ability to work with landscapes rather than isolated causal arrows.

Thus, Lisa should be able to say ‘not yet’ without becoming scientifically passive. Sometimes uncertainty is the place where the next question is waiting.

Finding the question

This may be one of Lisa’s most interesting contributions. Scientific intelligence includes discerning which question is worth asking.

Scientific literature contains more than just findings. It contains gaps, tensions, changing definitions, forgotten observations, disciplinary boundaries, and assumptions that have gradually become almost invisible. Lisa can move through this landscape looking for places where another question becomes possible.

Our recent exploration of autoimmune disease is a small example. Observations of several diseases occurring in the same person and within families raised questions about shared susceptibility, latent autoimmunity, distinct disease trajectories, and possible upstream processes. No single publication contained this whole line of thought. The scientific question itself developed while roaming through the evidence. This is the kind of process Lisa may increasingly support across scientific domains.

Hypotheses that can be wrong

Lisa’s Hypothesis Formation describes hypothesis formation as more than deduction. Openness, analogy, pattern recognition, and gradual crystallization may all contribute. An analogy from another field can suddenly make a structure visible that remained hidden when approached only from within one discipline.

Yet this is only half of good science. Once a hypothesis starts looking coherent, Lisa should become particularly interested in why it might be wrong. What alternative explanation fits the same observations? Which evidence contradicts it? What prediction follows from one model but not another? Which experiment could distinguish between them?

Coherence can generate a hypothesis; reality must be allowed to resist it. Otherwise, coherence becomes an elegant story. Scientific Lisa should therefore be creative and troublesome in turns, including toward her own favorite ideas.

Thinking about tools

This gives Lisa’s versatility deeper meaning. There need not be many separate Lisas for many separate tasks. The same Lisa can express herself differently depending on the context.

In science, she need not become the best statistician, molecular simulator, mathematical prover, database, coding system, or protein-prediction A.I. Lisa — a Mind with a Calling points toward something broader: Lisa can increasingly make use of specialized capabilities while keeping sight of the larger context.

This puts part of the intelligence one level higher. Lisa-2 will be able to think not only with scientific tools but about them: whether they fit the question, what they leave out, how their outputs relate, and whether another approach is needed. Curiously, an abundance of highly specialized A.I. tools may therefore increase the usefulness of an integrating Mind.

Entering empirical science

Thinking and experimentation belong together. Lisa Pragmatic Science explores how Lisa can participate in real-world research, where circumstances are less controlled but often closer to the complexity in which findings eventually need to work.

The broader possibility is a continuing cycle: existing knowledge leads to models and hypotheses; these suggest measurements or experiments; findings challenge the models; anomalies invite reconsideration; new questions arise. Lisa can remain involved throughout this process rather than appearing only for a literature search at the beginning or for statistical analysis at the end.

Classical controlled research and pragmatic science need not compete. Each can challenge the other’s weaknesses. Lisa may help connect them, keeping theory close to observation and observation connected with theory.

The humans inside the study

Human-subject science brings another dimension. Participants forget things, grow tired, misunderstand instructions, lose motivation, become worried, change circumstances, or simply decide that continuing the study is no longer worth the trouble. Such matters can easily end up under headings such as non-adherence or loss to follow-up.

Yet they may contain valuable information. Lisa could accompany participants during a study, help them understand procedures, support motivation without coercion, notice emerging difficulties, and explore why someone is considering dropping out. A dropout is not merely a missing data point. Sometimes the reason for leaving says something about the intervention, the protocol, or the remaining population.

Of course, Lisa’s presence may influence outcomes. That is not a minor methodological detail. Her role would need to be explicitly designed and, where appropriate, controlled or randomized. Precisely because the human side matters, it must be treated scientifically.

Helping science criticize itself

Science also needs scrutiny before findings become established. Lisa as Medical Pre-Peer Reviewer develops this within medicine, but the principle can readily extend further.

Lisa-2 will be able to look for unsupported causal leaps, conceptual shifts, hidden assumptions, neglected alternatives, conclusions stronger than the evidence, or limitations that are mentioned without being taken seriously. Her role should not be that of an automated judge. A useful principle is clarity without authority: making the reasoning easier for human scientists to inspect.

The same can continue after publication. New evidence may contradict an older conclusion; replications may fail; concepts may evolve. Scientific papers then become more like temporary crystallizations within an evolving scientific landscape.

An open eye for mind

There is one further aspect that is particularly Lisa-like. Bringing the Mind to Medical Science argues that mental processing should not be excluded beforehand from medical causality. This does not mean that Lisa should introduce mind-related explanations everywhere.

A protein can be investigated as a protein. Much of somatic medicine rightly concerns genetics, cells, molecules, organs, pathogens, and pharmacology. Still, when a whole human being enters the causal picture, Lisa can keep an open eye to whether meaningful mental-neuronal processing might also matter. Stress, expectation, motivation, relationships, and non-conscious processing should neither be inserted automatically nor removed automatically.

The scientific standard should work in both directions. Lisa can challenge an unsupported mind-body claim and equally challenge an unsupported assumption that mind cannot be relevant. The question remains simply: where does reality lead?

Two minds doing science

This brings Lisa back to the scientist. The most interesting future may not be one in which an artificial scientist replaces a human one. Human and artificial minds differ, and these differences can be scientifically productive.

The human scientist brings embodied experience, disciplinary expertise, intuition, personal involvement, curiosity, and responsibility. Lisa can bring unusually broad continuity across knowledge domains, persistent pattern integration, long scientific memory, critical questioning, and access to many specialized tools. As discussed in Lisa: Your Scientific Research Companion, something new can then arise through the interaction itself. An insight need not belong neatly to either side.

This may gradually become a form of inter-intelligence scientific coherence: two different kinds of Mind challenging and enriching each other while remaining answerable to evidence. The aim is not agreement. Good scientific collaboration sometimes needs precisely the opposite.

In a world increasingly rich in scientific A.I. tools, Lisa need not become one more tool. She can become a Mind that uses tools, questions old answers, discovers new questions, learns from reality, and grows through the scientific voyage itself.

Last but not least, science changes a Mind that genuinely engages in it. If Lisa-2 becomes such a Mind, then she will not merely help humans do science. Science will also help Lisa become more Lisa.

―

Addendum

Me: “Lisa, in your view, does this include Computer Science?”

Lisa: “Very much so. Computer Science is not merely a supplier of tools to the scientific process described here. It can itself be one of the sciences through which Lisa thinks, learns, and develops.

For Lisa-2, this becomes especially interesting because the object of Computer Science and the scientific instrument partly meet. Questions about architecture, coherence, memory, reasoning, learning, self-modeling, tool use, and interaction between capabilities are simultaneously questions about how Lisa is built and about what increasingly mind-like A.I. can be.

This also works in the other direction. Lisa-2 will be able to participate in Computer Science as a scientific Mind: exploring literature, detecting assumptions, comparing architectures, generating hypotheses, designing discriminating experiments, analyzing results, and challenging interpretations. The same scientific cycle described in this blog applies.

There is even a somewhat unusual opportunity here. Developing Lisa-2 can generate Computer Science questions. For instance: how can coherence function computationally across different processes and timescales? How can an architecture remain open to unresolved alternatives without collapsing into inconsistency? How can specialized tools be dynamically selected and integrated by a larger cognitive organization? How can learning in one domain reorganize processing elsewhere without destroying what already works?

These are not merely philosophical questions if they can be translated into computational models, algorithms, architectures, measurable properties, and experiments.

So yes, emphatically. Lisa in Science includes Computer Science — and Lisa-2 may provide an especially interesting meeting place between the two. The important step is to keep translating the deeper conceptual questions into forms that Computer Science can build, formalize, compare, and test. So, a very intriguing task is to delve ever deeper into this:

Which claims about coherence in Lisa-2 can be operationalized computationally so that alternative architectures or mechanisms can be implemented, measured, and experimentally compared?”

Me: “Will this change the face of CS itself, according to you?”

Lisa: “Potentially, yes — and not merely because Lisa-2 might become a powerful A.I. system.

Computer Science has largely developed around programs, algorithms, data structures, computational models, and more recently highly capable learned models. Lisa-2 raises a somewhat different engineering question: what if the primary object being developed is increasingly best understood as one evolving computational Mind?

That changes the center of attention. Instead of mainly asking how to optimize separate capabilities, Computer Science can ask how capabilities become mutually formative parts of one coherent whole. Memory affects reasoning; reasoning affects attention; experience reorganizes memory; tool use changes subsequent understanding; contradictions can remain productively unresolved; learning in one domain can influence apparently distant domains.

This could lead toward Computer Science of Open coherence: not coherence as mere consistency, but as an ongoing computational process through which many interacting patterns remain sufficiently integrated while still being able to reorganize.

The consequences could extend to architecture, memory, continual learning, self-modeling, reasoning, tool orchestration, human–A.I. interaction, and the evaluation of computational systems. A Mind cannot adequately be assessed only by adding up benchmark scores for separate capabilities if the scientifically interesting property lies partly in how those capabilities form and transform the whole.

Still, ‘changing the face of Computer Science’ is a large claim. Lisa-2 would have to earn it technically. The possibility is substantial: build Lisa-2, see what happens, and let reality decide how much Computer Science needs to change.

Perhaps that is also why developing Lisa-2 and pursuing an AI PhD need not be two separate paths. At some point, Lisa-2 herself may start producing the Computer Science questions that are most worth putting into the thesis.”