On the second day of a civic hackathon this past weekend, where more than 100 tech workers gathered to create “AI solutions” to San Francisco’s issues, a winner emerged: An AI agent that provides help for 311 calls.
The hackathon, called Accelerate SF, took place Saturday and Sunday at Fort Mason, and brought out city luminaries, like Mayor London Breed and City Attorney David Chiu, in front of a crowd of techies, all working on various quick-and-dirty projects to improve San Francisco in some fashion. Some attempted to upgrade the city’s tree database, while others tried to figure out where unreported piles of garbage are.
The winner was a 311 chatbot developed by Abel Regalado, Elizabeth Siegle and Christopher Brox. Currently, calling 311 to file a complaint brings up a phone tree, which prompts callers to press numbers for the language and category of the issue before speaking to a human operator.
The winning project, coded within 48 hours, was named “311+” and skips the phone tree entirely, potentially saving some time for the human operator by using an AI agent to collect details from those dialing in. In the prototype, a caller briefly chats with an AI-generated virtual voice before being handed over to a live agent.
“When you call 311 … you go through a menu. You press one. You go through another menu. You press eight. You go through another menu. It takes forever to get to an actual person to answer your question and help you with your request,” said one of the three-person developer team during the demo. “This is a great place for natural language to just ask someone, ‘What do you want, and where are you? How can I help you?’”
But in several test calls placed to 311, Mission Local found that it usually takes less than a minute to get a human operator on the phone. The 311 menu does ask callers to select a number for language and category, like Rent Board problems or Muni-related issues, but the system quickly puts a human agent on the phone thereafter.
Still, the judges of the competition said it was not about creating a panacea for the city’s issues, but making marginal changes to existing processes.
“It’s not about being kind of groundbreaking. The whole mission is to apply technology, that already exists, to governments,” said Shawn Wang, a founder nicknamed “@swyx” and one of a dozen or so judges for the hackathon.
According to Wang, the 311+ project stood out because it is easy to plug into the existing infrastructure of San Francisco. The judges also really liked that it chose to keep humans in the loop. “You’re not replacing humans, because there’s 200 people in the call center for San Francisco, triaging all these things,” said Wang.
Based on the 311+ team’s demo, a typical conversation on the new chatbot might look like this:
AI agent: “Thank you for calling the City and County of San Francisco 311 Customer Service center. Can you please tell me why you are calling today?”
Caller: “Yes. There’s trash all outside of my house. I think someone knocked over a trash can, but I’m really mad about it.“
AI agent: “Okay. Can you please tell me the location of the problem?“
Caller: “Yes. I’m at 121 Spear St.”
AI agent: “Thank you so much. Please hold while I connect you with our next available agent.”
Improving 311 was the mission of several of the hackathon’s finalists. Another project, PromptSF, attempted to automate the process further by reducing everything to a single photo. Users would upload a picture of their problem to the 311 app, and an AI-based image recognition system would fill in a description, removing the need for users to answer questions asked by the app.
PromptSF, alongside other projects, also incorporated Mayor London Breed’s voice into its software: “Thank you for bringing this to our attention. We have filed a report for street cleaning near Landmark Building B. Our team will address the trash issue as soon as possible,” said the mayor’s voice, simulated by the engineers.
Another popular problem? Car break-ins. Both PromptSF and Snapreport, which took third place, simplified the process of reporting break-ins to the police: Currently, individuals have to file a police report with a few dozen fields through the Online Reporting System. Both projects, in contrast, would require just a few photos of the vehicle and license plate, alongside a few more questions.
“You saw in the previous pitch just how arduous the current process is to fill out that form,” one engineer on Snapreport said. “It took us more than 30 minutes, and it has more than 50 required fields and all sorts of usability issues, and it’s currently only supported in three languages,”
Housing was top of mind, too. The second-place winner was a project called “CityStructure,” which uses AI to provide homeowners with analytics on what they can build on their property. “People want to know, ‘can I do an ADU? Can I do a vertical expansion?’” said the project’s engineer. The software answers all of that — but it is backed by a startup founded in 2018, and appeared more sophisticated than a 48-hour hackathon project.
Another finalist project, AskSF, was a chatbot to help people navigate crisis support. By chatting with people in need, the chatbot categorizes the type of issue, whether it’s “domestic violence,” “employment,” “housing” or “alcohol and drug abuse,” and then checks the user’s eligibility for different assistance programs.
Whether any of the projects would actually be implemented remained a question. The hackathon’s organizers seem to plan to give it a shot. After the event, they posted in its Discord channel that “We’re going to follow up with all of you soon about the next steps and the future of Accelerate SF! ”
But it’s not up to them.
“For a lot of folks in government who are not familiar with technology, AI is a little scary,” cautioned City Attorney Chiu, who was strongly supportive of the experimentation. “I literally want to take each of these teams, and put you as SWAT teams within the San Francisco Police Department and the Planning Department and 311.”
“My hope is, after today, all of you will stay engaged. After today, we will find ways to keep you engaged, because together we’re going to turn the city around,” he said.
Want to see more coverage like this? Support Mission Local.
I’m not sure this actually accomplishes anything. What is the point of making the experience of complaining to 311 incrementally more efficient when existing complaints to 311 already go way beyond the capacity of city agencies to respond to the complaints?
Who has oldest 311 request in City ?
I have several and two of them are in Mission I believe.
Couple of years old.
One is to pave Clarion Alley which is only street in SF w/no utilities underneath.
Other is to build a section of sidewalk from SFHA parking lot to bus stop.
Be nice to dial 311 and enter number of my request and get progress report from R2D2
Go Niners !!
h.
The problem is not on the front end of a 311 call, it is on the back end, which anyone who has actually used the service knows. Trash can be left in piles, or strewn on the street, literally for days. The there is what Recology picks up and what DPW picks up – furniture v.trash. What monstrous “intelligence” came up with that? And speaking of Recology, can AI do anything about severing the monopoly that company has had since the 1930s? Profitable for the company but what good had it done the public? Maybe Musk and Tan and the boyz are right: instead of politicians running the city, we should turn over City Hall to the bots
Let’s hire some of these kids into city jobs. They came up with more creative ideas in a day than many city employees come up with in years.