Is this the same as interactive evolutionary computation?

The areas overlap, but human-based approaches can emphasize human generation and recombination in addition to human evaluation.

Must all operators be performed by people?

No. The allocation between people and software depends on the particular system.

Does the method guarantee a globally best solution?

No. It is a search process whose effectiveness depends on representation, variation, selection, participation and evaluation.

AI Glossary / AI and machine learning / Intermediate

Human-based evolutionary computation

Definition

Human-based evolutionary computation uses people's contributions within an evolutionary search process to create, modify, combine or evaluate candidate solutions. It organizes human judgment and innovation around repeated variation and selection rather than relying entirely on automated operators.

What it means

Evolutionary computation searches a population of candidate solutions. New variants are produced, candidates are evaluated, and selected material influences later generations. In a fully computational version, algorithms carry out those operations.

Human-based approaches assign some of that work to people. Participants may invent initial candidates, revise existing ones, combine useful elements or judge which alternatives are promising. The division of work varies between systems.

This is broader than using a person only to score machine-generated results. Human-based genetic-algorithm research explicitly considers human innovation and recombination as well as evaluation, which can matter when solutions are difficult to encode or modify automatically.

The process still needs structure. A suggestion box with unrelated submissions is not automatically an evolutionary system. There should be a defined way for earlier candidates and judgments to shape the next round of search.

How it works

Define the problem, candidate representation and evaluation criteria. Candidates might be designs, written proposals or other artifacts, but participants need enough shared context to make meaningful comparisons.

Create an initial population through human contributions, automated generation or both. Record candidate identities and lineage so that later modifications and combinations can be understood.

Run variation and selection rounds. People may improve a candidate, combine features from several parents or evaluate alternatives. The platform coordinates these contributions and determines which candidates remain available for further development.

Evaluate the process as well as the final results. Participant fatigue, inconsistent preferences and premature agreement can narrow the search. Clear criteria, diverse participation and independent checks help reveal whether the process is producing useful solutions rather than merely popular ones.

Three practical examples

1. A design workshop evolves packaging concepts over several rounds. Participants combine useful features from earlier designs and evaluate new candidates against a shared brief.

2. A service-improvement platform lets staff revise and merge workflow proposals. Selected proposals become starting points for later variants rather than disappearing into a one-time vote.

3. A research study compares human-generated mutations with automated changes for a creative task. It measures contribution effort and result quality instead of assuming that either source of variation is always superior.

Think of it this way

Think of a collaborative recipe workshop where cooks taste versions, keep useful ideas and combine promising ingredients into the next round. The analogy is limited because preference alone may not satisfy the real objective: a popular recipe could still fail cost, accessibility or production constraints, just as a popular candidate solution can fail a project requirement.

Worked example

Suppose a workshop begins with 8 service proposals. Participants select 3 for further development, create 2 revised versions of each, and add 2 entirely new proposals. The next round contains 6 revisions plus 2 new candidates, again totaling 8.

One revised proposal combines the short intake form from candidate A with the clearer confirmation message from candidate C. Recording that lineage helps the team understand what changed and which ideas are being reused.

The arithmetic describes population management, not proof of improvement. If every evaluator favors only familiar designs, repeated selection may discard useful novelty. A separate assessment against the project brief is needed to determine whether the later generation is actually better.

Illustrative case study

Fictional case: Linden Services invites frontline employees to improve an appointment workflow. Its platform allows contributors to revise a proposal or combine clearly identified elements from existing proposals.

Each round uses the same criteria: clarity, effort and feasibility. The team limits review workload and keeps a small number of unconventional candidates available so that early popularity does not eliminate every alternative.

Before adoption, operations staff test the shortlisted workflows with realistic cases. The example illustrates structured collaborative search, not a claim that voting alone establishes effectiveness or that a real company achieved a particular saving.

Common misconceptions and mistakes

Mistake: Calling any crowdsourcing activity evolutionary computation. Iterative variation, selection and inheritance of candidate material are important to the concept.

Mistake: Restricting the human role to scoring. Some human-based approaches also use people to initialize, mutate and recombine solutions.

Mistake: Equating collective preference with objective quality. Evaluators may share blind spots, and the chosen criteria may omit constraints that matter in practice.

Business use

The approach can be useful when human creativity or judgment is difficult to replace with a reliable automatic operator. It may support design exploration, proposal development and other tasks where a solution's quality is hard to capture in a single formula.

The business case should include participant time, incentives, contribution ownership and the cost of evaluating finalists. Preserve traceability and make selection rules understandable. A process that produces many variations can still waste effort if contributors do not share a workable definition of success.

Check your understanding

A platform collects ideas once and ranks them by votes, without revision or recombination. What is missing from a strong evolutionary interpretation?

Answer: A defined iterative process in which selected candidate material produces new variants for further evaluation.

Related terms

Learn next

Learn population, variation and selection in a genetic algorithm. Design a two-round collaborative exercise that records candidate lineage and uses an independent final check beyond participant popularity.

Sources

Alex Kosorukoff, Human Based Genetic Algorithm: https://www.researchgate.net/publication/2361218_Human_Based_Genetic_Algorithm. Author-uploaded primary paper record consulted. Related primary research: https://gpbib.pmacs.upenn.edu/gecco2002/MPP262.pdf. Workshop examples and fictional case are original.

Previous: Human - AI interaction Next: Human-centered AI