Despite the untold potential of artificial intelligence (AI) and Colorado’s extraordinary innovation talent base, Colorado lawmakers are approaching AI legislation without soliciting input from our community. As a result, and perhaps in their haste to stop feared misuses of AI, they are rushing consideration of legislation that oversimplifies the technology and makes unfounded assumptions about its development and uses.

As a longtime developer and entrepreneur who educates businesses about AI, I understand the complexities of adding AI into an organization’s business software and processes. The development ecosystem includes people who build fundamental models, fine-tuners who customize the models to support an organization’s needs and those who rely on customized tools to support everyday operations. It is clear to me lawmakers are trying to squeeze all these activities into a misshapen legislative box that doesn’t reflect the technology’s nuances or how it is used — and the potential consequences are significant. Ill-fitting over-regulation of AI will hurt Colorado’s innovation economy and drive investors and entrepreneurs out of state.

Stay up to speed: Sign up for daily opinion in your inbox Monday-Friday

A fundamental flaw is the legislation’s assumption only a few big developers — like Google, Microsoft and OpenAI — will produce and distribute general AI models while thousands of organizations, or “deployers,” license them for specific business uses. Perhaps this is based on the large companies’ public-relations machines that constantly tout massive investments and the power of their technologies. In reality, AI is not as powerful out-of-the-box as many believe. As such, some of the perceived risks to consumers we read about online are also likely exaggerated.

Frankly, general-purpose AI models aren’t very useful when they are first trained. They are like pizza dough, but to make great pizza you have to shape the dough, add sauce, add toppings and cook it. It’s not a trivial job to automate complex tasks or decision-making for a specific domain, so many businesses and their software teams will need to customize AI models to fit their needs and connect those bespoke models to systems already at work within the business. Customization might modify the model’s behavior or add new datasets to educate the model about a particular industry or task.

The AI legislation seems to consider these customizations as creating entirely new models, subjecting user organizations — including governments and non-profits — to extensive disclosure requirements and compliance obligations. More compliance means more lawyers, more risk managers, more insurance and higher costs. The resulting impacts on innovation, and who owns innovation, will be massive, as these burdens and risks will cause smaller organizations to avoid AI innovation and give more power to the larger firms that can afford to comply with the regulatory requirements. That will lead to higher costs of AI adoption, fewer entrepreneurs in the customization business, and fewer options and higher costs for end-user organizations.

Another lost benefit will be to our power grid. The largest AI systems use substantial amounts of energy, but most use cases only tap a fraction of their computational power. By customizing and downsizing models to fit specific uses, energy use is reduced dramatically, which benefits all of us. But if costs and risks drive AI innovation back to the largest firms, their focus will be on big solutions for their largest partners. Colorado’s entrepreneurial economy and spirit will suffer, and our power grid will be stretched unnecessarily thin.

The legislation also does not take into account how high-profile AI missteps have already alerted businesses to potential pitfalls they are now intent on avoiding. Ask any AI developer, and they will tell you most businesses are working to prevent problems such as discrimination and misinformation before deploying AI to ensure their customers get quality tools without reputational risk. Additionally, many startups are building quality control and evaluation systems for businesses that employ AI models. As such, it’s misguided to impose burdensome regulations on the underlying technology, where much research, discovery and work toward safe and reliable tools continue to take place.

Those who work with AI, like me, understand these systems require investment and careful oversight. It would benefit all Coloradans if, rather than passing rushed and oversimplified legislation that creates new red tape and hampers innovation, legislators incentivize collaboration and educate consumers and businesses on responsible uses for the technology to maximize its benefits.

Mark Hennings is the Denver-based founder of Entry Point AI.