DORi the owl is the Colorado Department of Revenue’s chatbot for tax information. (Screenshot)
Ask a generative AI chatbot to describe what Gov. Jared Polis has accomplished for Colorado, and it might mention that he “worked to expand funding for early childhood education and kindergarten” and “pursued initiatives to lower the cost of prescription drugs,” as ChatGPT accurately reported recently, based on a prompt I entered.
The digital tool also stated that Polis “established a state office focused on fighting climate change and protecting the environment.” Hmm. I know of no such office.
The query demonstrates the extraordinary power, and potential pitfalls, of generative AI — the kind of artificial intelligence that can learn and create new content, such as the human-style words and images that sophisticated chatbots like ChatGPT can churn out.
AI already is being put to use in countless ways throughout society, such as for education, business, entertainment, health care and journalism. Governments are also finding applications for it, and this is where constituents should demand special care.
GET THE MORNING HEADLINES DELIVERED TO YOUR INBOX
Generative AI chatbots draw on vast stores of information and almost instantly assemble cogent passages of informative text, making them potentially attractive communication tools for government agencies. But chatbots are fallible and prone to abuse. They have been found sometimes to reflect bias and spread misinformation.
Government agencies should adopt clear policies around whether generative AI has a place in their operations and, if it does, how it may be used.
AI already shows up in some areas of Colorado state government. DORi is the Department of Revenue’s owl-character chatbot that answers Coloradans’ tax questions. The DMV has a public-facing chatbot. These are relatively primitive tools, with a limited set of responses, and resemble the chat features that banks, utilities and other businesses increasingly use to interact with customers. You know you’re interacting with a computer when you use them.
Generative AI chatbots, however, are capable of unlimited responses, which can be practically indistinguishable from human creations.
How might such a tool be put to use in government? It could be used to write reports. It could transform how government agencies interact with residents by “generating detailed responses that mimic human conversation.” It could help government staff write briefings, design communications campaigns and produce summaries of public meetings. It could help draft bills and speeches.
These and other applications might benefit constituents in some ways. But, because AI can produce inaccurate and discriminatory products, guardrails are necessary, as some lawmakers around the country acknowledge. For example, a bill in California, known as the “AI-Ware Act,” includes “a requirement that any state agency deploying AI in programs or services notify the public in a conspicuous way.”
The real balance is to not stifle innovation ... I think the big thing we have to look at is what controls the businesses are putting in place to review their decision-making processes. And are they adequately updating them?
– State Sen. Robert Rodriguez
State and local governments in Colorado appear so far not to have enacted robust policies around generative AI. Kevin Bommer, executive director of the Colorado Municipal League, which advocates on behalf of cities throughout the state, told me this week that AI is “a major discussion topic” among city leaders around the country. His organization does not currently have guidance for Colorado cities around generative AI.
“But that’s kind of one of those things that we do need to pay attention to and kind of find out not only what our members are talking about but whether or not there’s any space for us to provide any guidance,” he said.
I checked with officials at the Colorado House of Representatives, the Senate, the attorney general’s office, the governor’s office, the secretary of state’s office — they all indicated that they do not have concrete policies around AI.
“The Department of State does not have a formal written policy regarding official use of AI technology, though we are closely monitoring continuing developments to the technology including potential threats and uses. We do not use it for internal or external communication,” a secretary of state’s office spokesperson wrote in an email.
This could soon change.
State Sen. Robert Rodriguez is in the early stages of developing possible AI regulations that could be introduced in the next session of the Colorado General Assembly. Rodriguez has some related experience in this realm. Two years ago he helped spearhead Colorado’s data privacy law. AI helped him write introductory remarks for a gig-worker transparency bill he sponsored this year.
“I’m not a great writer or speaker,” he said. “I tried it, and it gave me a great base to start.”
Now he is thinking about the various ways generative AI, not just in government and not just in written communication, is emerging throughout society. Tech observers have cautioned that AI could decide whether students are accepted to college, how public benefits are distributed, and how health care is administered.
“HR,” Rodriguez adds, noting hiring decisions can be AI-based. “I mean, how many things do you do that’s probably decision-making-related? It’s a pretty broad industry.”
Republican U.S. Rep. Ken Buck of Windsor recently co-sponsored a bill that would ensure AI could not launch a nuclear weapon.
Rodriguez is consulting with lawmakers in other states, such as Connecticut, to establish some policy uniformity across the country. He wants to avoid excessive regulation of the private sector, because AI can be useful, he said.
“The real balance is to not stifle innovation,” he said. “I think the big thing we have to look at is what controls the businesses are putting in place to review their decision-making processes. And are they adequately updating them?”
Asked if the same controls should apply to government agencies, Rodriguez said, “I would hope so. That’s the heavy lift.”
It’s a lift that ChatGPT would support.
I asked ChatGPT this week if the state government should use AI language models like itself, and the response sounded eerily like a government spokesperson. The decision is “a complex one that requires careful consideration of several factors,” it said. “Ultimately, the decision to use AI language models for government communication should involve careful consideration of these factors, weighing the potential benefits and drawbacks.”
Couldn’t have said it better myself.
SUPPORT NEWS YOU TRUST.
Our stories may be republished online or in print under Creative Commons license CC BY-NC-ND 4.0. We ask that you edit only for style or to shorten, provide proper attribution and link to our web site. Please see our republishing guidelines for use of photos and graphics.