Caltrans delves into GenAI testing to reduce bottlenecks, keep roads safe

Why does I-80 always clog right before University Avenue in Berkeley on Saturday afternoons? What’s causing the nighttime crashes on I-280 from Meridian Avenue to McLaughlin Avenue in San Jose? And what’s the reason I-5 near Del Paso Road in Sacramento continually ranks as one of the most dangerous stretches of highway in the state?

Caltrans wants to know the answers to these and countless other road-related questions and they are getting some much-needed help from generative artificial intelligence, better known as GenAI.

The pilot program is part of Gov. Gavin Newsom’s efforts to use the power of new technologies to improve the state’s public services.

Newsom touts California as a world leader in GenAI innovation with 35 of the world’s top 50 AI companies located in the state, including many right here in Silicon Valley. He signed an executive order last year to study the development, use, and risks of AI technology and to establish a responsible process for evaluation and deployment of AI within state government.

“From our perspective, GenAI is a new opportunity,’’ said David Man, division chief of traffic operations for Caltrans. “It’s unknown for state governments as a whole. We wanted to learn about its potential uses, bring in new ideas and innovations, study it and, at our option, consider procuring the technology.”

Caltrans says the sheer volume of data generated from traffic sensors and cameras — coupled with the continuous velocity of data generation and the variety of videos, images, log files, third-party data streams and guidance — poses significant challenges for human beings to digest and utilize.

To help with that, the state is focused on transportation-related contracts in two areas:

In the first, Deloitte Consulting and INRIX Inc., a transportation analysis company, will develop GenAI tools that would investigate near misses of injuries and deaths on California roads to identify hazardous areas and then allow workers to brainstorm ideas to better protect those outside vehicles — pedestrians and bikers.

In the second, Deloitte and Accenture, a global professional services company, will develop tools to improve traffic analysis, reduce bottlenecks, and bolster the state’s ability to reduce crashes and monitor the often mind-boggling traffic in both the south and north areas of the state.

Each company will be paid $1 to develop and test potential AI tools in a secure environment during a six-month pilot program. The companies will work with Caltrans during the testing window to evaluate whether the tools are working or make changes. Under state law Caltrans must pay at least $1 to enter into contracts with vendors.

Gen AI is a branch of artificial intelligence that can create text, audio and photos in response to prompts. It’s the technology behind ChatGPT, the controversial and often inaccurate writing tool started by Microsoft-backed OpenAI.

The companies developing the GenAI tools for Caltrans will use technologies developed by OpenAI as well as Google- and Amazon-backed Anthropic.

Right now the state transportation agency relies on a continuous data stream from more than 39,000 in-ground detectors, thousands of digital message signs and vehicle detection stations and at least 3,000 cameras along California’s highway system. That information is meshed with more data about infrastructure projects, weather, and transit and traffic engineering.

Collecting and housing all that data is “quite complex,” and Caltrans is often strapped by human limitations and time constraints in its efforts to get a clear picture about what’s happening on the roads and why, Caltrans officials say.

That’s where AI comes in.

“We really want to use and explore the potential benefits of AI, but it’s not intended to be a replacement for our own judgment,’’ said Caltrans’ Chief Safety Officer Rachel Carpenter.

Caltrans officials say the big benefit of the technology is that it has the potential to assist state workers in more quickly and accurately analyzing millions of data points to make more immediate decisions about roadway safety. Officials say AI tools won’t interact with confidential data or personal information.

In addition to the Caltrans project, the California Department of Tax and Fee Administration, which administers 40 programs, is also looking to create new ways to help Californians understand their business tax requirements by swiftly searching expansive reference materials and providing responses for staff to share with taxpayers by telephone and live chat.

AI tools could also help the state’s Health and Human Services Agency assist non-English speakers in obtaining timely access to information about public benefits and public programs. And the California Department of Public Health is exploring using AI to improve health care facility inspections.

But some GenAI experts are worried that having California become such a high-tech global hub comes with big risks.

“Generative AI generates sounds, text or images, including for good and for bad. You can make your own art and make deep fakes, you can generate music or songs and deep fake someone’s voice,’’ said Irina Raicu, who directs the Internet Ethics program at the Markkula Center for Applied Ethics at Santa Clara University.

Raicu, who said she has worked with companies developing “responsible AI policies and processes,” said the immediate need to safeguard GenAI can’t be overstated.

“My concern is that they could give totally wrong information,” she said, citing an example in New York City in which an AI-powered chatbot created by the city to help small businesses doled out false guidance and advised companies to violate the law. The rapidly growing technology has also raised concerns about job losses, misinformation, privacy and automation bias.

Caltrans says its framework allows potential GenAI solutions to be developed and evaluated in a calculated method with close oversight from qualified engineers and data experts.

“We know that GenAI tools generally lack the ability for engineering judgment or to consider any factual inconsistencies of the data used to train itself,’’ said Caltrans’ Carpenter in an email. “The Caltrans’ pilots are testing in a closed ‘sandbox,’ a controlled environment, using trusted and validated data sources so that any potential solution can be tested first under regulatory supervision.”