The competition between artificial intelligence and human scientists is spreading from chessboards and math problems into the core of life sciences.
Ginkgo Bioworks had originally planned a protein design contest deliberately crafted as a "Kasparov versus Deep Blue" historic moment, with Stanford University professor Michael Jewett set to face off alone against OpenAI's team of AI agents. However, on the eve of the first round on September 14, the match was abruptly halted and then relaunched in a completely different form — the format shifted from a "one-on-one duel" to "coopetition," with any scientist allowed to sign up and form a team. Ginkgo co-founder and president Reshma Shetty said the consideration behind the adjustment was a desire to avoid intensifying the current tension in the scientific community toward artificial intelligence.
This shift came after OpenAI announced that its internal AI model had cracked the Navier-Stokes problem, which had vexed the mathematics community for nearly a century, in less than a week. The news sent shockwaves through the global mathematics community and reignited intense debate about the future relationship between AI and humans. For Ginkgo's event organizers, this public backlash directly prompted them to re-examine the positioning of the competition. Ginkgo co-founder and CEO Jason Kelly admitted:
"Everyone is afraid of losing."
Format Reshaped: From "Human vs Machine Showdown" to "Coopetition"
The original "Mike Versus the Machines" concept was highly dramatic: Jewett, an authority in the field of cell-free protein synthesis, would face an OpenAI team composed solely of AI agents. Jewett could use any commercially available AI tools, while OpenAI could deploy its most advanced unreleased internal models. Ginkgo not only provided its 15,000-square-foot Boston headquarters laboratory, equipped with a large number of robots, as the venue, but also hired a documentary film crew to record the entire process.
The restructured format completely changed this confrontational setup. Any scientist can sign up, either forming a unified "human team" to take on the challenge collectively or forming separate teams to challenge OpenAI's agents individually. Ginkgo no longer uses the word "competition," instead adopting the term "coopetition" — a word included in the Oxford English Dictionary, meaning "cooperation between competitors." Shetty said: "It's not just about who is better, but about how cooperation can advance the achievement of scientific goals."
At present, Jewett has not yet confirmed whether he will participate.
Navier-Stokes Shockwave and the Scientific Community's Collective Unease
The direct trigger for the format adjustment was a major breakthrough announced by OpenAI on September 8: its advanced internal AI model solved the Navier-Stokes problem, which had troubled mathematics for nearly a century, in less than a week, surpassing two mathematicians who had recently made important progress on the problem. Some mathematicians called this mathematics' "Kasparov-Deep Blue moment," arguing that it requires people to rethink the future direction of mathematics.
In fact, unease in the scientific community had been building for some time before this. Scientists have continued to criticize assertions by AI leaders such as Anthropic CEO Dario Amodei and Google DeepMind chairman Demis Hassabis that "AI will conquer major diseases within the next decade." This week, Amodei tweeted that Claude had discovered molecules that could potentially be used for gene editing. Although he admitted it was unclear how important the discovery was, he was still criticized by many scientists, who said he should have shown greater restraint before more detailed information was available.
This collective anxiety permeating the scientific community ultimately directly influenced Ginkgo's decision. Shetty said the recent public climate forced the company to reconsider how the competition was presented, to avoid this protein contest being simplistically interpreted as yet another proof that AI crushes humans.
Competition Rules: Robot Judges, Proteins as the Subject
Under the new rules, the competition will begin in mid-October and is expected to last until the end of this year. In each round, participating teams must submit optimal chemical formulation designs for three proteins designated by Ginkgo, after which Ginkgo will convert the plans into instructions and drive robots to carry out experiments.
The target protein for the first stage is the commonly used laboratory superfolder green fluorescent protein (sfGFP), and the winning criterion is producing the most protein at the lowest cost. The proteins in the following two rounds will increase in difficulty, and the judging criteria will also become more complex, covering indicators such as whether the protein can fold correctly — which is crucial for actual drug development. Shetty used baking as an analogy: "It's not enough to just make the cake; the cake also has to look good and taste good."
Ginkgo will provide participating teams with a list of optional chemicals, while also allowing each team to add up to 10 chemicals outside the list, to reflect creativity and professional judgment in scientific research. OpenAI's agents and the human teams will not know each other's material selections.
Ginkgo and OpenAI: A History from Cooperation to Competition
This competition is not the first collaboration between Ginkgo and OpenAI. The two companies have deep ties: in 2014, after OpenAI CEO Sam Altman took the helm at Y Combinator, Ginkgo became the first biotechnology company selected for the accelerator program, and Y Combinator also became an important investor in Ginkgo.
The idea for the competition reportedly originated at Altman's 40th birthday party in April 2025. At the party, Kelly introduced Ginkgo's robotic laboratory capabilities to a senior OpenAI executive and invited the executive to visit its laboratory in Emeryville, California, which led to cooperation talks between the two sides and OpenAI's life sciences research team.
The two companies' first collaboration focused precisely on the field of cell-free protein synthesis — completely consistent with the theme of this competition. In February of this year, the two sides announced that they had jointly reduced the cost of the process by 40%, surpassing the new industry standard that Jewett's team had just set six months earlier. It was precisely this result that prompted Jewett's question: OpenAI's advantage lay in the fact that robots could complete about 30,000 experiments, while students in his laboratory could only work manually and completed just 1,231. He posed the question: if both humans and AI agents could use robots to carry out experiments, which side could produce a better plan? This question ultimately led to the competition now unfolding.
Competition Fever Spreads: A Major AI Test in Biology
Ginkgo and OpenAI are not the only institutions preparing this kind of "human vs machine competition." Sam Rodriques, CEO and co-founder of AI startup Edison Scientific, said he and his colleagues have compiled a list of biological problems that AI could potentially solve. Research institution FutureHouse is forming a review panel and plans to set cash rewards for solvers.
Even larger competitions are also on the horizon. Earlier this month, Anthropic and Adaptyv Bio, an automated protein design laboratory startup headquartered in Lausanne, Switzerland, announced that they would jointly host the "world's largest protein design competition," which launched on September 28. Anthropic will provide $1 million worth of Claude computing credits to help scientists design proteins addressing five major challenges; Adaptyv and Anthropic will jointly fund laboratory testing of selected designs, and the competition is open to all applicants. Adaptyv CEO and co-founder Julian Englert said it is entirely theoretically possible for a team composed purely of AI agents to win one or more of the challenges.
Marinka Zitnik, an AI agent researcher at Harvard Medical School, believes the dense emergence of these competitions reflects the rapid evolution of AI scientists' capabilities. But she points out that the true ultimate test has not yet arrived — that will be the day when an AI system independently makes a breakthrough discovery and is judged by experienced scientists as "worthy of a Nobel Prize."
Comments