Open-endedness, a search strategy inspired by biological evolution, is becoming central to how AI tackles scientific discovery. Rather than optimizing for a single goal, open-ended algorithms explore a space of possibilities indefinitely, often producing unexpected breakthroughs. Lila Sciences, a startup focused on autonomous scientific discovery, has assembled a dedicated team under renowned AI researcher Ken Stanley to apply this philosophy to real-world problems in materials science, energy, and RNA therapeutics.

What Is Open-Endedness, and Why Does Science Need It?

Open-endedness is fundamentally different from how most machine learning systems work today. Traditional machine learning is goal-driven: you give an algorithm a target, and it optimizes a solution. This approach, called "hill-climbing," can get stuck at local peaks, missing better solutions that lie beyond. Open-endedness, by contrast, is "a divergent search through a space of possibilities that continues to produce interesting discoveries indefinitely," according to Ken Stanley, senior vice president of open-endedness at Lila Sciences.

The best example is biological evolution. Organisms aren't all optimized for one goal, such as flight. Instead, species branch out, divergently filling ecological niches, resulting in birds, fish, flowers, mushrooms, bacteria, and the rest of the tree of life. Science itself mirrors this pattern, filled with serendipitous discoveries and new ideas building on old ones, sometimes even on results initially dismissed as failures.

Thomas Sheppard, now part of Stanley's team at Lila, experienced this firsthand during an undergraduate internship at a diamond synthesis company. When a batch of crystals appeared low-quality and destined for the trash, Sheppard kept analyzing them anyway. "Even though I was aware that they were ostensibly low quality, I was naive enough not to dismiss them entirely," he explained. Those discarded diamonds turned out to be exactly what the company needed for further advances, a discovery that made Sheppard philosophically inclined toward open-endedness.

How Is Lila Sciences Applying Open-Endedness to Autonomous Discovery?

Starting in 2025, Stanley built an open-endedness team at Lila to design exploration methods and apply them to real-world scientific problems. Lila's mission is to create "scientific superintelligence," advanced AI systems that significantly exceed human intelligence across scientific domains and execute the scientific method autonomously in AI-run laboratories called AI Science Factories. The company has already begun applying its technology to industries such as advanced materials, energy and the environment, and RNA therapeutics.

It may seem unusual for a startup focused on practical solutions to invest heavily in such a philosophical notion, but Stanley argues that theory and practice are highly aligned. "Whereas open-endedness as a field is one bet among many at most labs, at a company whose primary mission is to create scientific superintelligence, it's mission critical," he stated. Jeff Clune, a longtime collaborator of Stanley and cofounder of the startup Recursive Superintelligence, reinforced this view, saying that "open-ended algorithms are some of the, if not the, most powerful algorithms we know of to allow AI to make true breakthroughs on extremely challenging problems".

Who Makes Up Stanley's Open-Endedness Team?

Stanley's team at Lila now has nine people, assembled with a deliberate focus on intellectual diversity. His approach mirrors his philosophy on machine learning: seek out varied perspectives and backgrounds. Key members include Joel Lehman, who studied under Stanley at the University of Central Florida, worked with him at Uber and OpenAI, and co-authored a book about open-endedness titled "Why Greatness Cannot Be Planned". Lehman noted that while Uber and OpenAI treated open-endedness as "an interesting side quest," the alignment at Lila is "quite rare and unique".

Lehman

Other team members bring diverse expertise and career stages:

• David D'Ambrosio: A principal scientist on the team who was Stanley's PhD advisor and previously worked on applying open-endedness to robotics at Google DeepMind before joining Lila.
• Matthew Fontaine: Earned a PhD in 2024 researching quality-diversity algorithms, a method that optimizes for both variety and success simultaneously.
• Alon Albalak: Earned his PhD in 2024 from natural language processing, bringing expertise from a different corner of AI research.
• Sebastian Gabriel: The team's research engineer with a broader and more technical background.
• Ivy Zhang: Graduated from college last year and worked on artificial life at the startup Sakana AI, bringing fresh perspectives to the team.
• Thomas Sheppard: Two years out of college, he had read one of Stanley's papers, ran related experiments in his free time, and cold-emailed the authors, demonstrating the passion Stanley values.

Stanley emphasized one of his principles for building the team: "There should always be somebody who has fresh eyes". This commitment to diversity extends beyond technical expertise to career stage and background.

Stanley

What Is Stanley's Track Record with Open-Endedness Research?

Stanley's journey with open-endedness spans over two decades. An important moment occurred in 2007 when he and his students built a website called Picbreeder, which shows users an array of images. When users click on one, the algorithm mutates it into several descendants, producing a new array. The branching can continue indefinitely. While playing with Picbreeder, Stanley clicked on what looked like an alien face, and several generations later it became a racecar, a serendipitous surprise that changed his approach to problem-solving.

From this insight, Stanley and his collaborators began designing algorithms aimed at producing diverse solutions, regardless of immediate usefulness, and found that usefulness often came as a byproduct. They then developed "quality-diversity" methods that simultaneously optimized for both variety and success. Some of these methods have been adopted widely, including by researchers at Google DeepMind, Meta, and Sakana AI.

Before joining Lila, Stanley held several prominent positions. While still teaching at the University of Central Florida, he cofounded Geometric Intelligence, which was subsequently acquired by Uber and became the company's AI lab. He later led the Open-Endedness team at OpenAI, then cofounded Maven, a social network that encourages chance encounters instead of rewarding popularity.

Why Does IBM's Mathematician Believe Fundamental Research Matters for AI?

While Lila focuses on applying open-endedness to scientific discovery, the broader AI and computing field is grappling with fundamental questions about what problems can be solved efficiently. Subhash Khot, a prizewinning theorist at New York University, recently joined IBM Research as a senior mathematician in June 2026, continuing his decades-long work on the boundary between what computers can solve efficiently and what may remain impractical.

Khot's work matters because as AI systems produce new results on long-standing math problems, understanding computational limits becomes crucial. "There is really no other solution than to have fast algorithms," Khot told IBM Think in a recent interview. "A faster algorithm would really make an enormous difference". As hardware performance gains slow due to physical constraints, better algorithms become increasingly important for handling ever-larger datasets.

"Subhash's arrival reflects the seriousness of our commitment to foundational research," said Jay Gambetta, IBM Fellow and Director of IBM Research. "We are investing in the theory of computing as a pillar of IBM Research's future, and this is just the beginning."

Jay Gambella, IBM Fellow and Director of IBM Research

Khot is best known for proposing the Unique Games Conjecture in 2002, which gave computer scientists a framework for asking how close an efficient algorithm can come to the best answer when calculating the optimum would take too long. Earlier this year, he shared the National Academy of Sciences' Michael and Sheila Held Prize with four collaborators for their proof of the 2-to-2 Games Theorem, which provided the strongest evidence yet for the Unique Games Conjecture.

How Do Foundational Research and Applied AI Discovery Connect?

Both Stanley's work at Lila and Khot's research at IBM represent a growing recognition that foundational research is essential for advancing AI. Foundational research examines the principles beneath a technology rather than aiming at a particular product. In computing, it can help reveal which problems new algorithms or machines might be able to solve.

Stanley's team works in San Francisco, surrounded by companies focused on the next fundraise, quarterly profits, or AI benchmarks, the opposite of open-endedness. Meanwhile, he and his researchers have space to explore, to follow intriguing threads, and to invent new ways to invent. This approach contrasts sharply with the pressure to optimize for immediate metrics, yet Stanley argues it is essential for breakthrough discoveries in science.

The convergence of these two approaches, open-ended exploration and theoretical computer science, suggests that the future of AI-driven scientific discovery may depend on balancing ambitious practical goals with deep theoretical understanding. As Lila builds AI Science Factories and IBM invests in foundational mathematics, the field is recognizing that serendipitous discoveries and rigorous theory are not luxuries but necessities for solving the hardest problems.