🦸🏻#1: Open-endedness and AI Agents – A Path from Generative to Creative AI?
Read the full version on Turing Post
Intro
AI agents. Agentic workflows. Autonomous agents. Intelligent agents. Digital agents. Task-oriented agents. Smart agents. Copilots. AI personas. AI assistants. Embodied agents etc.
The topic of agents is so hot right now that no one even knows what the correct term for them is. What's more, in machine learning, the concept of agents originally meant something different than what it means in AI today. In this series on agentic systems and workflows, we will clarify these terms and set the record straight. There is indeed a lot of confusion and questions surrounding this topic. From theoretical frameworks to practical applications, current innovations to future potential, policies and roadmaps, we will cover it all.
We don’t know if AGI is at the end of this path but we will be open to it. And to emphasize that we will start our series … with open-endedness ;) That was one of the topics suggested by our readers here
AI agents are usually discussed through their most visible abilities: planning, tool use, memory, reasoning, reflection, and autonomy. But underneath all of this sits a deeper question: can AI systems move beyond executing tasks and start generating genuinely new directions, behaviors, strategies, and solutions?
That is where open-endedness becomes the foundation. Open-ended systems are not limited to one fixed goal or one predefined set of outputs. They can keep exploring, adapting, and producing novelty as they interact with their environment. In the context of AI agents, this matters because many real-world problems do not have a final checklist. Research, design, science, education, engineering, and creative work all require systems that can keep searching through possibility space instead of stopping once they find the first acceptable answer.
This is also why open-endedness feels like a bridge between generative AI and creative AI. Generative AI can produce text, images, code, music, video, and plans. But creation is not only production. Creation involves exploration, selection, surprise, and sometimes the discovery of goals that were not fully specified at the beginning. If agents are going to become more than high-level automation tools, they need some ability to explore beyond narrow instructions.
However, open-endedness did not suddenly appear with modern LLM agents. Its roots go back to cybernetics, self-regulating systems, evolutionary computation, reinforcement learning breakthroughs such as AlphaGo and many other concepts. Each stage added something important like feedback loops, adaptation or unexpected strategy discovery. The modern version of the question is different, though. Today we are asking whether AI agents can use open-ended exploration inside useful workflows like scientific discovery, engineering design, education, etc.:
- An open-ended agent would not only complete the task but also explore adjacent possibilities, discover new subgoals, build reusable skills, and improve its own approach over time.
- Another key idea is that open-endedness is not the same as randomness. For open-endedness to matter, the outputs must be novel, useful, and learnable from the perspective of an observer.
- Open-ended agents may become harder to predict, harder to evaluate, and harder to control. They may produce outcomes that were not anticipated when the benchmark was designed.
The biggest promise is that open-ended AI could help in areas where human creativity is slow, expensive, or bottlenecked by expertise. In science, it could explore hypotheses. In engineering, it could generate designs. In education, it could personalize learning paths. In software, it could discover better workflows. In agentic systems, it could turn static automation into continuous adaptation.
Recent agentic research makes this question more concrete. Systems like Voyager in Minecraft showed how an LLM-powered agent can explore, acquire skills, write code, and improve through interaction. The AI Scientist pushed the idea further into scientific workflows: generating ideas, running experiments, analyzing results, and writing reports. But these systems are still limited.
The full article unpacks this broader story: why open-endedness matters for AI agents, how the idea evolved historically, what it means today, what limitations it could address, and why the path from generative AI to creative AI depends on more than just bigger models.
*Originally published on Turing Post.
If you want to receive our articles straight to your inbox, please subscribe here
