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

Instead of simply building yet another machine learning model, Autoscience is developing an autonomous artificial intelligence research lab that can run experiments continuously. The platform “employs” nonhuman AI scientists and engineers that invent, validate and deploy specialized state-of-the-art models.

“We’ve reached a point where human intuition is no longer enough to navigate the complexity of algorithmic discovery,” said Chief Executive Eliot Cowan.

Cowan said the company’s objective is to “compress a decade of machine learning research into months” and gain a competitive edge for customers.

More than 2,000 machine learning papers are published every week across equally numerous publications. Autoscience argues that no human research team can keep up with the sheer volume of research being produced, even by human scientists. Evaluation is no longer within reach – it’s time to automate.

The company’s first deployment with automate high-stakes financial applications, manufacturing and fraud detection and enabling companies’ benefit from AI-driven research without needing headcount.

Autoscience first gained recognition when its autonomous lab AI system to produce a peer-reviewed scientific research paper. The company said its AI agent, Carl, produced work accepted to the International Conference on Learning Representations 2025 workshop track, needing only minor human edits, limited to citations and formatting.

The paper was based on initial workshop submissions and became a full-length paper named “Investigating Alignment Signals in Initial Token Representations.”

The scientific community has raised questions about AI-written papers in peer review. Most ethical questions center on transparency, accountability and prevention of fraud. By 2025, it was noted that a urge of AI-related language had appeared in scientific papers, especially given the already present popularity of chatbots such as OpenAI Group PBC’s ChatGPT and some scientists had already begun using AI models for peer review, even against policy.

In many cases, generalized models, such as ChatGPT, get scientific concepts wrong. The further into a specific discipline that they go, the more complex and specific the jargon becomes. At the same time, the nuance and particularities become important. The off-the-shelf conversational training of extremely large datasets works against models such as those from OpenAI.

Arguably, Autoscience has resolved this by developing its automated AI laboratory models to align with machine learning science as accurately as possible.

Tokyo-based startup Sakana AI also built an “AI scientist,” to automate scientific discovery last year. It also submitted a paper to the ICLR 2025 workshop that passed peer review.

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

Autoscience said it will use the new funding to scale its current capabilities for a select group of Fortune 500 and large private companies who are training specialized models in high-stakes environments. It’s deploying a managed service to automate AI research scientists that continuously generate and ship improvements to machine learning models in parallel, which will allow enterprise companies to discover, test and serve better models.

The capital will also allow the company to support a larger engineering team to accelerate human-driven AI research.

Image: SiliconANGLE/Microsoft Designer