Shinka Evolve Breaks the Mold: How AI Is Now Creating Its Own Challenges

Artificial intelligence has long been celebrated for its ability to solve problems that humans have already defined. From mastering board games to predicting protein structures, the most impressive achievements have been measured against a fixed set of rules and objectives. Yet a new wave of research is turning that paradigm on its head. Instead of merely optimizing within a predetermined space, a fresh generation of AI systems is learning to generate the very problems they will later solve. At the forefront of this shift is Shinka Evolve, a project from Sakana AI that promises to redefine what it means for a machine to think.

The Rise of Problem‑Generating AI

For decades, the AI community has celebrated milestones like AlphaGo, AlphaZero, and AlphaFold—systems that pushed the limits of what machines can achieve when given a clear goal. These successes, however, all share a common trait: they rely on human designers to craft the problem before the AI can even begin to solve it. The new question is not whether AI can solve a problem, but whether it can ask the right question in the first place.

Enter Shinka Evolve. Rather than being handed a puzzle, this system is tasked with discovering new challenges that are both solvable and valuable. By doing so, it mirrors the natural process of evolution, where organisms not only adapt to their environment but also shape it in turn. The result is an AI that can chart its own course, continually expanding the frontier of what it can tackle.

AlphaEvolve: Optimizing Within Boundaries

AlphaEvolve, a product of DeepMind’s research, exemplifies the traditional AI model. It excels at finding optimal solutions within a defined objective function. Think of a chess engine that can calculate millions of moves ahead, or a protein‑folding algorithm that predicts the most stable structure for a given amino acid sequence. These systems are powerful, but they hit a plateau once the objective is fully satisfied.

Because AlphaEvolve’s architecture is built around a fixed reward signal, it lacks an intrinsic motivation to explore beyond the limits of that signal. When the reward is maximized, the system has no incentive to look for new problems or to innovate beyond the existing framework. As Robert Lange, a founding researcher at Sakana AI, puts it, AlphaEvolve is “a brilliant student who can ace any exam but struggles to come up with the questions themselves.”

Shinka Evolve: A Self‑Directed Evolutionary Engine

Shinka Evolve flips this dynamic on its head. Instead of starting with a problem, it begins with a blank slate and a set of evolutionary principles—mutation, selection, and recombination. The system generates a population of candidate problems, evaluates them based on criteria such as novelty, solvability, and potential impact, and then iteratively refines both the problems and the solutions that address them.

Key features of Shinka Evolve include:

• Open‑Ended Exploration: The AI is not confined to a single objective; it can pursue multiple, sometimes conflicting, goals simultaneously.
• Self‑Generated Objectives: New challenges arise from the system’s own internal dynamics, reducing reliance on human input.
• Co‑Evolution of Problems and Solutions: As the AI discovers a new problem, it immediately begins to develop a solution, creating a feedback loop that accelerates learning.
• Scalable Complexity: The framework can handle problems ranging from simple puzzles to large‑scale scientific questions.

In practice, Shinka Evolve has already outpaced AlphaEvolve on several benchmark tasks. For instance, when tasked with optimizing a complex chemical reaction, Shinka Evolve not only found a more efficient pathway but also identified a previously unconsidered reaction mechanism that could open new avenues for drug discovery.

Implications for the Future of AI

The shift from problem‑solving to problem‑generating AI has profound implications across multiple domains:

1. Scientific Research: Autonomous discovery could accelerate breakthroughs in physics, chemistry, and biology by proposing novel experiments and hypotheses.