As this week's National Association of State Chief Information Officers (NASCIO) conference got underway in Philadelphia, GT talked with California CIO Liana Bailey-Crimmins, who outlined the vast potential AI holds to improve the lives of state residents in several areas.
"The Department of Technology is responsible for all IT policy," Bailey-Crimmins said, "and so what we were required to do is not only look at the benefits [of AI] but also to look at the guardrails," including security policy that stipulates that third parties the state works with don't own government data.
"Any vendor partner that's going to work with us, the data is preserved and the state owns it no matter where the source of the information, the prompts, come from, from an engineering perspective."
California's Chief Technology Officer Jonathan Porat noted that the state was well prepared to confront generative AI (GenAI) given its prior work on artificial intelligence policy. But generative AI has changed some of the rules of engagement.
"Regular AI and machine learning, cloud computing — there's all types of technologies that we at the state have implemented for a long time. We know what the main risks are and we can streamline those to a couple of points," Porat explained.
"What we've realized with GenAI is we do have to innovate the way that we think about that because if we make individual restrictions or we make our governance too detailed, it can end up limiting what we're doing," he added.
Now, instead of focusing on inputs and outputs, Porat said the state uses an outcomes-based approach.
Video transcripts available below.
Video transcript, Liana Bailey-Crimmins: We’re first in the nation to do a government-managed sandbox where we were doing nine proof of concepts. We had five challenges and had nine separate instances of how we could address a challenge on the business level using generative AI. So thinking of in California we have over 200 dialects and languages spoken. So how do we break down the barrier of language and accessibility? Also looking at efficiencies, we know that surveyors go out and go into hospitals. How can we make it easier so that they could do more of those inspections to make hospitals safer?
And then when you think of the roadways, we want zero fatalities on our roads. And so just what we could be doing in relation to using generative AI to potentially look at vulnerable road users. If you're a cyclist, motorcyclist, walking, how can we use the technology potentially to then give safety engineers opportunities to look at the data differently so they can make decisions that would benefit Californians?
Video transcript, Jonathan Porat: When we think about AI use cases, we actually evaluate them against three different dimensions. One is the data. Are the data that we're using appropriate for a GenAI system? Are they properly being governed and secure? One is the technology itself. Has the technology been proven? Does it come from a place where the state has a lot of experience or is it something that might require some more information? And the last is the use case itself. Is the use case appropriate for GenAI? Are we using this to replace a state worker? Are we using it for something that we may not think is ethical or may hurt the inclusivity or the equity of the state's programs? So working on all of those together we have a really strong partnership with both the other state departments as well as the vendor community to make sure that we're responsibly and ethically implementing AI across the state.
And so we've had to kind of adapt our approach to become a lot more flexible and a lot more agile and really rely on those outcomes that we care about and try to partner up to the extent that we can with agency and department, chief information officers, chief information security officers. So what we've done is we've built a self -risk assessment for each department and built that into our policy. So CDT does not need to go in and actually approve every single use case. We don't have to go in and say yes or no, but we work with those departments and those individual tech leaders at the departments to help evaluate those different dimensions that we mentioned earlier, data, the use case and the technology to make sure before a department goes out, we can tailor governance so it's the right fit. So if someone is just looking for a tool in very highly used and commonly used enterprise software to generate images or to help with creating content, those are things that we can have a much more streamlined approach. If someone is looking for something that's going to be more mission-driven, is going to have a much larger public impact saying that we don't have a lot of experience with, we can take a more hands-on approach and be a little bit more deliberative and explorative.
