Select Amazon Web Services customers will gain access to The Biological Computing Company on Tuesday. The startup builds an AI model using rat brain cells for a limited preview. The technology aims to improve AI systems for generating videos. Both Amazon and the startup expect a wider rollout to all AWS enterprise customers soon.
Co-founder Alexander Ksendzovsky explains the core process. The team codes information like images to biological material. They observe how the biology processes the data. Then they build a tool that mimics that process.
Deap Ubhi is global director of technology for startups at Amazon Web Services. He notes that Amazon already works with another biological computing company named Cortical Labs. Cortical Labs combines lab-grown neurons with silicon chips to process data. Ubhi praises the startup for taking a pragmatic approach. The company works within existing generative AI standards instead of trying to reinvent the transformer architecture.
The startup started four years ago in Baltimore, Maryland. Two neuroscientists and neurosurgeons founded the company. Alexander Ksendzovsky serves as CEO. Jon Pomeraniec serves as president and COO. The company raised $25 million earlier this year from investors led by Primary Venture Partners. A subsequent funding round brought its total funding to more than $50 million.
Last year the startup opened a research and development lab in San Francisco. Thirty-five employees work directly with rat brain cells and human stem cells there. Researchers place living cells on multi-electrode silicon arrays made by Swiss biotech company 3Brain. They send electrical stimulation patterns into the neurons and record the responses. TBC analyzes this neural activity to create software that improves video generation models.
Ksendzovsky states that focusing on video was a scientific and practical choice. The physical layout of the silicon arrays fits visual information easily. Images map onto the grid much better than text or language. Prominent AI researcher and investor Jeff Dean suggested fine-tuning video generation models first. This approach allows the startup to test performance against accepted industry benchmarks.
Previously, the technology was only available through Bluesky Compute. TBC claims its AI model is up to five times faster at video generation than an unnamed open-source frontier model. The company also claims the technology significantly lowers inference costs.
Ubhi points out potential scaling questions for the new technology. He wonders if fidelity will drop when customers attempt to generate longer videos. TBC will test whether its technology holds up as new customers push the system to its limits.
