Potato’s AI research assistant digs through publications for science experiment inspiration
Science can reward its practitioners with thrilling eureka moments — with an emphasis on “moments.” A lot of research is slogging through experiments that don’t always work, wading through research publications, and performing mental contortions in pursuit of the next hypothesis.
A new Seattle-area startup aims to change that. Co-founders Ryan Kosai and Nick Edwards have built Potato AI, which they’re pitching as an AI scientific research assistant.
Potato launched last year and today announced a $1 million pre-seed round. The startup also recently shared that it’s the first partner in an AI program created by Wiley, an academic and education publishing powerhouse.
The company’s tuberous name is a nod to the classic experiment where kids turn a potato into a battery using household items including pennies, copper wire and galvanized nails.
“That’s really what we’re doing,” Edwards said. “We’re helping people plan and run experiments.”
Edwards, the startup’s CEO, has firsthand experience toiling at the research bench. He has a PhD in neuroscience and did research at the National Institutes of Health and the University of California, San Diego. He recalled pouring through journal articles to put together research protocols.
But there can be inaccuracies in the articles and inconsistent approaches, which has created a reproducibility crisis where scientists can’t always replicate their own or others’ work and get the same results.
Potato uses chat-based, generative-AI tools built from large language models that are further refined using retrieval-augmented generation (RAG). RAG allows them to improve the accuracy of the results by incorporating scientific literature references that will now be expanded to include Wiley’s publications.
Potato’s platform is already being used in labs at biotech companies and universities including the University of Washington; Stanford; Harvard; Massachusetts Institute of Technology, University of California, San Diego; University of California, Berkeley; The Scripps Research Institute; and others.
“The goal is really to enable researchers to quickly distill scientific content and design repeatable, modified experiments so they can speed up the journey to medical breakthroughs,” said Kosai.
Kosai is Potato’s chief technology officer. He was previously a founder and CTO for the fintech startup Attunely, which has since shut down, and was CTO for the startup studio Pioneer Square Labs. He also worked for ExtraHop, Marchex and Athleon.
The tool can also critique journal articles, helping researchers write papers for publication, and assist in reviews of journal submissions. It can look for shortcomings in protocols, such as weak controls, and help brainstorm new lines of inquiry.
One hope is that it can find subtle, but potentially game changing tweaks to protocols. Edwards recalled his own neuroscience research and the realization that swapping one salt for another in a commonly used solution would keep the tissue he was studying alive for twice as long. Information like that can be easily missed in the traditional, informal methods for sharing protocols, he said, which include word-of-mouth and back-of-napkin exchanges.
There are other AI research tools available, though many focus on digesting and summarizing published research. Others working in the space include nonprofits and academic institutions.
A challenge for the co-founders is guarding against errors or hallucinations in the AI results. They’re doing their own fact checking and users will be pointed to source documents and can query the tool for more information.
While the current focus is on biology and medicine, the startup plans to expand its AI assistant to aid with research in materials science and chemistry.
Scientists can join the platform for free for limited use, and they can pay to upload and access additional information and expand their collaborations.
Participants in the pre-seed funding include Axial, Pioneer Square Labs, AI2 Incubator and Bow Capital, as well as angel investors such as Cloudera founder Jeff Hammerbacher and ExtraHop founder Jesse Rothstein.