Autoscience spends $14 million to build automated research lab for machine learning models

Artificial intelligence startup Autoscience Institute has launched with $14 million in seed funding to automate research into new machine learning models.

AutoScience is developing an autonomous artificial intelligence research lab that can run experiments continuously, rather than just building another machine learning model. The platform “employs” non-human AI scientists and engineers who invent, validate, and deploy specialized cutting-edge models.

“We’ve reached a point where human intuition is no longer sufficient to overcome the complexity of algorithmic discovery,” CEO Elliott Cowan said.

Cowan said the company’s goal is to “compress 10 years of machine learning research into months” to give customers a competitive edge.

More than 2,000 machine learning papers are published every week in an equally large number of publications. AutoScience claims that human research teams, even human scientists, cannot keep up with the enormous amount of research being produced. Valuation is no longer out of reach and it’s time to automate it.

The company’s initial deployments will automate high-stakes financial applications, manufacturing, fraud detection, and enable businesses to benefit from AI-driven research without the need for human resources.

Autoscience first gained attention when its autonomous lab AI system produced a peer-reviewed scientific research paper. The company said its AI agent, Karl, produced the work that was accepted into the workshop track of the 2025 International Conference on Learning and Representation, limited to citations and formatting, and required only minor human editing.

This paper was based on the first workshop submission, which was turned into a full-length paper titled “Investigating alignment signals in early token representations.”

The scientific community has raised questions about papers authored by AI during peer review. Most ethical questions focus on transparency, accountability, and preventing fraud. By 2025, there was a trend towards AI-related language appearing in scientific papers, especially given the already growing popularity of chatbots such as OpenAI Group PBC’s ChatGPT, and the fact that some scientists had already begun to use AI models for peer review, even against policy.

In many cases, generalized models such as ChatGPT misunderstand scientific concepts. The further you progress into a particular field, the more complex and specific the terminology becomes. At the same time, nuance and specificity are important. Off-the-shelf conversational training on very large datasets works for OpenAI’s models and more.

Arguably, Autoscience has solved this problem by developing automated AI lab models that match machine learning science as closely as possible.

Last year, Tokyo-based startup Sakana AI also built “AI Scientist” to automate scientific discoveries. We also submitted a paper to the ICLR 2025 workshop and passed peer review.

The round, announced Wednesday, was led by venture capital firm General Catalyst with participation from Toyota Ventures, Perplexity Fund, MaC Ventures and S32.

AutoScience said it will use the new funding to expand its current capabilities for a select group of Fortune 500 companies and large private companies training specialized models in high-stakes environments. The company deploys managed services that automate AI research scientists to continuously generate and deliver improvements to machine learning models in parallel, enabling companies to discover, test, and deliver better models.

This capital will enable the company to support a large engineering team to accelerate human-driven AI research.

Image: SiliconANGLE/Microsoft Designer