PicbreederをサカナAIがVLMで再現、人間との創造性の差を検証

"Users simply selected images they found interesting, allowing unexpected forms such as faces, animals, vehicles, and skulls to emerge gradually across many generations and many different people." — 公式ブログの英語欄より（英日併記ページ・ロケール別表示）

"images published to the Picbreeder website between its launch in 2008 and its death around 2016"／"is the canonical example, in which users collaboratively create interesting images via interactive evolution" — 論文「3.2. Historical Picbreeder data」と「1. Introduction」より

"an initially small, protean neural network that takes as input some coordinates in pixel space, and outputs that pixel's color"／"these networks effectively encode infinite-resolution images"／"These individuals (each little brain containing a single picture) are then bred together and mutated to produce a new set of candidate images, and the process repeats."／"it's nearly impossible to plan for a particular long-term goal"／"Here, the artifacts materialize by accident, appearing to the user as objectives-in-retrospect." — 技術ブログのCPPNと育種の説明より

"かえって真に偉大な発見を遠ざけてしまうという逆説を論じた書籍です。その議論の中核にあったのが「PicBreeder」の実験でした。"／"目標をあらかじめ定めない探索こそが人間の創造性の根幹にあると考えたのです。" — 公式ブログの日本語欄より（英日併記ページ・ロケール別表示）

"They are trained, evaluated, and orchestrated as goal-following entities." — 技術ブログの冒頭より

"in collaboration with MIT and NYU, we revisit Picbreeder"／"We recreated this process using vision-language model agents."／"There is no target image and no explicit definition of what counts as progress." — 公式ブログの英語欄より（英日併記ページ・ロケール別表示）

"last revised 27 May 2026 (this version, v2)"／"to be published at GECCO 2026"／"Genetic and Evolutionary Computation Conference (GECCO '26), July 13-17, 2026, San José, Costa Rica" — arXivの論文ページより

"In practice, we run 10 breeding agents in parallel continuously, branching from and publishing to the shared archive, and launch a critic agent whenever the archive grows by 5 publications."／"Future samples are drawn partially at random, and partially according to recency of publication, mean ratings from past agents, and branching popularity." — 技術ブログの再現ループの説明より

"The session is the core unit of the Picbreeder loop."／"Each population comprises 15 CPPN-images; exact copies of parent(s) are always included among their offspring."／"the agent is asked to select an image for publication to the archive and give it a title"／"We initially allowed agents more freedom over when to publish, but agents chose to publish rapidly and end their session, hence we enforced longer sessions."／"allowing us to reconstruct the full phylogenetic tree of published images for comparison against that of VLM-driven Picbreeder" — 論文「3.2. Historical Picbreeder data」と「3.3. Playing Picbreeder with VLMs」より

"In particular, we inject random noise into the agents' decision making process, we play with their memories"／"we seed them with simple personality traits intended to indirectly influence their interaction with the system"／"we try to measure how visually different images in the archive look, how much they look similar to a predefined set of objects"／"We additionally measure the tree balance of generated phylogenies"／"The random baseline provides a lower bound."／"over 6 seeds; 2,000 sessions each" — 技術ブログの条件と指標の説明、表「Quantitative results」の注記より

"However, introducing a diverse population of agent personalities substantially improves exploration. In some runs, diverse agent populations approached or matched the human archive on measures of semantic diversity and produced more balanced evolutionary trees." — 公式ブログの英語欄より（英日併記ページ・ロケール別表示）

"The human archive emerges as a strong upper bound relative to our VLM-driven experiments"／"Among VLM experiments, agent diversity via the injection of subtle breeder system prompts is particularly effective in closing the VLM-human gap."／"The Human archive consists of a single historical"／"Adding agents also tends to steadily increase the degree of balance in the tree." — 技術ブログの本文と表「Quantitative results」の注記より

"They repeatedly select similar parents, make smaller conceptual leaps, and often refine an existing idea rather than abandoning it in search of something genuinely unexpected." — 公式ブログの英語欄より（英日併記ページ・ロケール別表示）

"Humans seem to take larger steps in semantic space (from face, to eye, to frog, to car) than VLMs (from abstract bird, to car seat, to dashboard, to hood ornament, to car)."／"Humans tend to take bigger leaps between publications and land on sharper, more refined images." — 論文の図5の説明と「4. Metrics & Interventions」より

"They often get sucked into their own attractors." — 技術ブログの質的な観察より

"When we remove agents' memories, and they only have context of the current generation, they're prone to reproducing the same image ad nauseam."／"Even one previous generation of history is enough to break this pathology, making the agent less likely to knowingly repeat a selection."／"Too much memory seems to overload the agent's context and degrade its capabilities, with archives at longer memory-lengths often filled with overly abstract or repetitive, simple forms."／"push the agents out of problematic basins, at the risk of filling the archive with a greater proportion of intermittent stepping-stone slop"／"A bit of noise in the selection process can also encourage exploration." — 技術ブログの本文と表「Quantitative results」の注記より

"small amounts of exploratory noise can increase diversity of generated archives, but at the cost of the quality of images therein"／"surprisingly little history is necessary for optimal performance, with greater context lengths leading to pathological behavior" — 論文「1. Introduction」より

"We also find intriguing evidence that open-ended evolution can produce more robust representations. A skull evolved by the agents changes smoothly when its underlying neural representation is perturbed, less fractured than a skull directly optimized with gradient descent, although still less cleanly disentangled than one evolved collectively by humans." — 公式ブログの英語欄より（英日併記ページ・ロケール別表示）

"But perhaps the most interesting result is the gap that remains."／"We still do not fully understand what enables humans to navigate open-ended search in this way, or what ingredient(s) current AI systems are missing. For now, the results suggest that there remains something important about human creativity that AI agents have not yet learned to reproduce."／"当社が探求するAI駆動型科学研究にも通じる大きな問いが残っています。"／"Sakana AIは今後も、オープンエンドな知性の探求を深めていきます。" — 公式ブログの英語欄と日本語欄より（英日併記ページ・ロケール別表示）

"Humans appear better at turning fortunate accidents into sustained creative discoveries: recognizing when something unexpected is worth pursuing, refining it, and then making a larger conceptual leap. The AI agents often notice interesting patterns too, but are more likely to become trapped in them." — 公式ブログの英語欄より（英日併記ページ・ロケール別表示）

"the human archive has a certain je ne sais quoi"／"as if the AI agents never got over the initial spark of resemblance to find something better yet" — 技術ブログの質的な観察より

"This paper will be presented at GECCO 2026 and is nominated for a best paper award" — 公式ブログの英語欄より（英日併記ページ・ロケール別表示）

"Best Paper (Complex Systems), GECCO 2026"／"that conventional benchmarks can never teach us"／"access an homage to the original site, populated with aggregate results over multiple replays of AI-driven reproductions of the collective human event" — 技術ブログの題名の下と図の説明より

"We make our code available at this https URL" — arXivの論文要旨より