LONDON — Artificial intelligence remains a sunlit spot in a stormy venture capital market, but concern from the public and policymakers about the technology’s impacts threatens the outlook. So prominent AI financier Radical Ventures is launching a tool to help other investors assess the risks and safety practices of startups in the sector they’re considering funding.

AI firms raised US$39.5 billion in 2,809 deals in the first half of the year, according to PitchBook. Toronto-headquartered Radical, founded in June 2017, has backed notable names in the field like generative toolmaker Cohere, robotics platform Covariant and autonomous vehicle startup Waabi. 

The investment firm has long considered whether potential portfolio companies are taking a responsible approach to AI development and deployment, said partner Aaron Rosenberg, who leads its European activities. Over the past six months, it’s converted that practice into a new tool that it’s now making available to other funds. 

The Responsible AI for Startups framework is a spreadsheet that prompts investors to consider the potential vulnerabilities of a company’s products and services, how likely and significant those issues are, and what steps the firm might take to reduce them.

Sample risks include ones impacting society or ethics, like sensitive or unintended uses of the technology that could hurt people or violate human rights; the firm’s ability to comply with regulatory requirements around IP, privacy or liability; and technical limitations like the amount of scientific and real-world proof underpinning an application. Users of the framework assess for themselves the severity of those hazards and the effectiveness of mitigation measures.

Radical will unveil the framework on Tuesday at an investor workshop in London, one of a number of events in the city scheduled this week around the U.K. government’s AI Safety Summit. “There is a lot of people that are new to AI, and they may not know the right questions to ask,” said Andre Rojas, director of applied AI projects at the Toronto-based Vector Institute, which provided technical expertise on the project.

While funds typically aren’t building or rolling out frontline AI tools, “we are providing the capital that is enabling that deployment,” said Leah Morris, senior director for Radical’s post-investment support program. That brings a responsibility to consider responsibility. The firm has chosen not to invest in certain startups over AI safety risks.

But Radical is not suggesting other funds steer clear of particular applications, nor trying to set sector-wide thresholds for acceptable risk. The tool is not a checklist, said Rosenberg. “It’s a way to ensure that you have done your diligence in considering these different dimensions of risks and benefits.” 

The framework is based on an extensive review of responsible AI studies and principles from different institutes and experts, Morris said. Radical and Vector will publish fictionalized case studies to show how to use the tool in practice. Both organizations hope it will evolve as other investors and companies use it. 

Policymakers around the world have ramped up discussion and legislation on AI since the debut of ChatGPT last November. Ministers and top sector executives will gather this week at Bletchley Park, an hour and a half northwest of London, to consider the technology’s impacts, particularly the risks of so-called “frontier” systems. 

Radical’s move comes as other investors push back on precautionary measures. In a 5,000-plus-word “techno-optimist manifesto” published earlier this month, controversial Silicon Valley VC Marc Andreessen cited “trust and safety” and “risk management” as forces hostile to progress. 

The framework is not designed to slow down AI or even deal making, but rather encourage reasonable action to address concerns about the technology, said Rojas. He noted that financial-sector firms take a similar approach to other kinds of risks. 

But he predicted that startups courting investment and seeking to understand their potential impacts are more likely to use the tool than financiers. VC firms will “try to come up with their own thing, just for a branding exercise,” he said. 

Startups might balk at telling investors all the things that could go wrong with their buzzy AI products. But Rosenberg said that’s a signal of its own. “Those that welcome the discussion are all the more attractive to us.”