Personal recursive self-improvement in economic reality

As borrowed fire grows stronger, where should a finite person invest their time? Personal recursive self-improvement makes the methods producing improvement themselves improvable; economic reality also calls for choosing which capabilities are worth developing. Learning takes time, bodies need practice, and tools need maintenance. Even a learnable capability competes with other uses of those resources: mastering it personally, accessing it through a model, or cultivating a complementary capability can produce different returns.

The direction chosen here is to develop embodied, multimodal reasoning in real situations, use capabilities that models can already provide reliably as tools, and revise the division of work as technology and society change. This personal RSI measures its gains in actual life, turning finite investment into understanding, judgment, and action that remain available for further use.

Replicable capabilities and worthwhile cultivation

The easier it is to represent a task's inputs, operations, and success conditions in a computer, and to generate and check candidates cheaply, the easier it becomes to organize large-scale training, search, and replication. Well-specified digital tasks permit many attempts in little time; improvements can then reach many users. Investing all one's learning time in reproducing operations already cheaply available carries an opportunity cost as tools continue improving.

This relation concerns task structure. Formalizability makes automation more feasible; economic substitution also depends on reliability, deployment costs, responsibility, demand, and the recombination of work. Autor's analysis of workplace automation emphasizes that replacing some tasks can raise the value of complementary human contributions, while occupations typically combine multiple tasks. Research entry

Comparative advantage arises within such combinations. Someone still better at a digital operation may gain more by calling on a model and spending time on more valuable observation, selection, or creation. Foundational knowledge remains part of this arrangement: understanding a model's contribution, recognizing consequential errors, and connecting results to reality can require personal concepts and practice. Those actual uses determine the depth worth cultivating.

Society responds to models of itself

Once social predictions enter society, participants can use, resist, and adapt to them. Hiring criteria change preparation, recommendations change attention, and evaluation metrics change organizational behavior. These responses alter later modeling data. Formalization has two roles here: it describes selected relations and, through institutions, tools, and adoption, participates in their transformation.

Perdomo and colleagues' work on performative prediction represents this feedback as a predictive model changing its target distribution, studying conditions under which repeated training can become stable. Paper It supplies formal tools for local analysis and clarifies the question: accuracy on past data and accuracy after intervention are distinct outcomes to examine.

Physiological and psychological states, mutual understanding, cultural meanings, resources, and institutions interact across different rhythms. The same words can have different consequences in different relationships; people also create activities that earlier categories did not capture. Data retain traces, models explain and simulate selected relations, and new participation continues producing conditions requiring renewed observation.

This essay therefore adopts an open conception of modeling whose use can change its own conditions: formalize within a stated scope, check through activity, and let changes in the object revise categories and methods. Formalization's participation in change gives a reason for this continuing return. A stronger assertion that every social process is in principle uncomputable would require a separate computability argument. A person's practical advantage can arise from relationships and opportunities for calibration gained through ongoing situated perception and action.

The body connects forms of understanding within one action

Human embodied reasoning connects vision, sound, touch, movement, affect, memory, and language. Picking up a cup involves weight, posture, and purpose together; listening involves words, tone, relationship history, and the immediate situation. Participants actively change their viewpoint, try an action, and adjust to the response. Multimodality here is coordination between perception and action within a continuing encounter between body and environment.

Robotics research already connects vision-language models to action and whole-body control. DeepMind's 2026 Gemini Robotics 2 report presents progress alongside remaining difficulties in multifinger manipulation, speed, and other capabilities. Research-team report This makes the essay's near-term judgment concrete: sustaining broad, reliable embodied participation affordably in changing real environments remains a capability to realize and test task by task.

The brain's energy use offers another line of inquiry. Levy and Calvert distinguish computation, communication, and other metabolic expenditure, showing why the frequently cited roughly 20 watts is not simply computational power. Paper abstract Low metabolic power coexists with rich perception and action and deserves investigation. Translating neural activity into machine operations, and determining which simulation reproduces relevant functions, are different questions. Completely reconstructing a brain and achieving comparable performance on a particular task also impose different implementation requirements.

A person already lives within this organization of body and world. Cultivation can develop existing perception, movement, and understanding: drawing to distinguish spatial relations, music to organize hearing and action, making to encounter material resistance, and reading and actual encounters to develop judgment about social meaning. Models expand explanations, provide comparisons, build tools, and organize experience, returning attention to relations worth personally perceiving and judging.

Make complementarity a testable division of work

Consider someone designing an interface for use while moving. They walk, hold the device, and notice environmental sound and visual interference, discovering an interaction difficulty absent from a static screenshot. A model develops that observation into candidate layouts, working prototypes, and checks. Actual use then determines which changes help. Situated participation and design judgment contribute on the human side; model support makes alternatives easier to generate and realize.

This arrangement can fail. Intuition can misjudge, models can omit conditions, and practice can cost much while yielding little. Compare the complete activity: its result, what the person can subsequently do, the recovery, supervision, and maintenance that support requires, and costs borne by others.

Being harder to replace in the near term is consequently a judgment to update with conditions. Difficulty automating an embodied activity still needs to be considered alongside demand, resources, and accessible opportunities when assessing economic value. New model capabilities can invite new complementarities. What is worthwhile in life also includes expression, understanding, relationships, and freely available time.

Improve how investment is selected

Selection itself becomes an object of recursion. A project reveals that understanding the use situation limits its results, changing observation and source selection next time. A better tool releases implementation time, changing practice and work allocation. A failure reveals that evaluation counted only output, bringing continuing judgment and maintenance burden into the comparison.

The four capacities thereby acquire further content: modeling explains contributions from people and tools under current economic and social conditions; simulation develops consequences of alternative investments and allocations; evaluation compares actual gains and opportunity costs; modification changes subsequent learning and cooperation. Later activity tests whether these selection methods improve.

Use model capabilities to extend embodiment and judgment, and let participation in reality continue changing how models are used. A person can thereby develop a contribution within a changing economy while retaining room to reconsider what is worth developing. A life worth sustaining gives this recursion a continuing, revisable direction.