Back in March, the Los Angeles Unified School District was held up as a trailblazer for its embrace of artificial intelligence, when it unveiled a custom-designed chatbot.
Superintendent Alberto Carvalho a 鈥済ame changer鈥� that would 鈥渁ccelerate learning at a level never seen before.鈥�
But in just five months, LAUSD has gone from enviable AI pioneer to
The district has temporarily 鈥淓d.鈥� That decision appears to have been prompted by upheaval at AllHere, the company LAUSD hired to create the tool at a
AllHere has . Its CEO and founder, Joanna Smith-Griffin, is no longer with the company. And a company whistleblower has
LAUSD has now become the poster district for what not to do in harnessing AI for K-12 education.
For starters, the district didn鈥檛 appear to have tightly defined the problem it was trying to fix with the tool, experts said. Plus, the district selected an inexperienced vendor and set an overly ambitious timetable for the project without proper protections for student data鈥攁ll signs that its leaders bought uncritically into the AI hype.
鈥淭here鈥檚 a dream that AI is just more or less automatically going to solve all or many problems [of K-12],鈥� said Ashok Goel, a professor of computer science and human-centered computing in the School of Interactive Computing at Georgia Institute of Technology. 鈥淚t鈥檚 overhyped. That is not how learning and education works.鈥�
Even so, LAUSD has made it clear it is not giving up on 鈥淓d鈥� or AI. In a statement, the district said that the tool鈥斺�� 鈥渂elongs鈥� to LAUSD and that it will ensure whatever company acquires AllHere 鈥渨ill continue to provide this first-of-its-kind resource to our students and families,鈥� a district spokesperson said.
AllHere did not respond to 澳门跑狗论坛鈥檚 requests for comment.
This marks the second time the district has bungled a cutting-edge tech initiative. LAUSD, under different leadership, rolled out a 1-to-1 computing iPad program in 2013 that was a complete disaster.
It鈥檚 unclear what will happen next with AI in LAUSD. But the challenges the nation鈥檚 second largest school district faced in developing an AI tool offer important lessons for other school systems.
Those lessons include:
1. Be clear about what problem you鈥檙e trying to solve with AI
LAUSD had hoped 鈥淓d"鈥攁 chatbot shaped like the sun鈥攚ould serve as a one-stop shop for students and parents seeking information on everything from bus schedules to upcoming tests. The district also wanted the tool to boost students鈥� academic skills, enhance their social and emotional development, and improve attendance.
That鈥檚 a broad to-do list for a single tool, suggested Katy Knight, the executive director of the Siegel Family Endowment. (澳门跑狗论坛, the publisher of 澳门跑狗论坛, receives support from the endowment for its coverage of AI and other technology issues. The media organization retains sole editorial control over that coverage.)
鈥淭he notion you can get everything you need from a chatbot, it鈥檚 sexy,鈥� Knight said. 鈥淏ut I look at this and I think, OK, this is what happens when you let the technology lead you instead of figuring out what you actually need. And then finding technology to serve that need.鈥�
Before purchasing鈥攐r in LAUSD鈥檚 case, commissioning the design of an AI product鈥� districts must articulate what need or challenge they are addressing and decide if AI is really the best way to go.
鈥淵ou need a clear understanding of what you want,鈥� said Rick Gay, the executive director of business services for the Fort Bend Independent School District in Sugarland, Texas. Before even looking for a vendor, answer questions, such as: 鈥淗ow do you want [the tool] to look? How do you want it to perform? What does success look like?鈥�
And if a district decides to deploy AI, leaders need to explain the technology to anyone who will use the tool, including principals, teachers, students, and parents. The district must clearly communicate AI鈥檚 limitations, including bias, and tendency to spit out completely wrong information.
鈥淭here鈥檚 an eyes wide open component to this,鈥� said Bree Dusseault, the managing director for the Center for Reinventing Public Education, a research organization, . 鈥淥nce you choose to go this route, you have to build everyone鈥檚 capacity鈥� on AI.
2. Vet ed-tech companies carefully
Many ed-tech companies purporting to be AI experts are startups without much of a track record in general, much less in K-12 schools, Gay said.
鈥淭hey spring up out of somebody鈥檚 basement,鈥� he said. 鈥淵ou have to vet these guys pretty carefully.鈥� That means asking for references from school district officials who have worked with the company on similar projects and asking it to demonstrate similar tools it has built, Gay said.
LAUSD may have rolled the dice in choosing a startup like AllHere, experts said.
A more established company might not have had so much difficulty delivering the tool and adhering to the principles necessary for it to work in K-12 schools. But a big-name ed-tech business would also likely have been pricier, and less attentive to the district鈥檚 specific needs.
鈥淲ith a bigger player, the cost would have been higher, but the risks would have been lower,鈥� Goel said. But many districts 鈥渄on鈥檛 pay as much attention to risks鈥� when assessing a new technology, he added. 鈥淭hey pay most of the attention to benefits and cost analysis,鈥� Goel said.
To be sure, AllHere, a Boston-based company best known for creating a text-messaging system to help improve student attendance, appeared by some measures to be a thriving organization. AllHere was (ranking 74 out of 250). The company鈥檚 website says it has worked with more than 8,000 schools in 34 states.
That may speak to another AI challenge: The technology is so new that it鈥檚 hard to discern which vendors have the capacity and expertise to make a customized product like 鈥淓d鈥� work, Dusseault said.
鈥淚 was struck by the fact that this organization was on lists and was recommended,鈥� Dusseault said. That means it鈥檚 possible that 鈥渢here鈥檚 no standard of quality for vendors right now. It鈥檚 really unpredictable who is truly equipped to succeed.鈥�
In fact, Jeremy Roschelle, the co-executive director of learning science research for Digital Promise, a nonprofit organization that works on equity and technology issues in schools, estimates only about a third of the new ed-tech startups marketing AI-powered products will be around a year from now.
鈥淚 say that with respect, because everything鈥檚 changing fast. The failure rate of new companies is high鈥� even when investors and developers aren鈥檛 trying to capitalize on a hyped-up new technology, Roschelle added.
3. Consider starting small and work on a reasonable timetable
LAUSD had big ambitions for 鈥淓d鈥� from the start鈥攁nd may have been unrealistic about the amount of time and testing it would take to achieve them, experts said.
鈥淣o district has done this type of work before, at the level that I know LAUSD wanted to do it,鈥� Dusseault said. 鈥淚 think that there is some element of biting off more than you can chew quickly.鈥�
The district put out a request soliciting a company to build 鈥淓d鈥� in February 2023. It piloted the tool with about 1,000 students before unveiling it with great fanfare in March. The district then made the bot available to an additional 55,000 students at select schools for the remainder of the school year.
And LAUSD planned to extend the tool to all students and teachers in the district by the beginning of the 2024-25 school year, that does not appear to have been revised in light of AllHere鈥檚 turmoil.
Compare LAUSD鈥檚 experience with 鈥淓d鈥� to a similar chatbot, , which Goel is developing and testing to assist adult learners at Georgia Tech.
That project began in 2016. About eight years later, the bot is still being put through its paces. It鈥檚 only been used in about 60 or 70 of the institution鈥檚 roughly 3,000 classes. Instead of going big quickly, Goel and his team have methodically used teacher feedback to improve the tool.
Although LAUSD probably had more resources to direct to a project like this than a researcher like Goel, the timetable was likely too aggressive, experts say.
Companies and school districts alike need to be careful about 鈥渟etting expectations around how fast and how much time鈥� it can take to create and properly test new AI tools, Roschelle said.
But some vendors are reluctant to acknowledge the long game, he added. When Roschelle listened to companies pitch AI tools at a recent ed-tech conference, 鈥淚 wasn鈥檛 [hearing] 鈥榯his is going to take a while,鈥欌�� he said.
That desire to set a speedy pace without considering the consequences prompted LAUSD to roll out 鈥淓d鈥� to all students. But experts say the district likely would have been better off targeting a limited number of schools.
鈥淚鈥檓 not saying don鈥檛 use the technology, but don鈥檛 jump all in with something which then you have egg on your face when things don鈥檛 go right,鈥� said Punya Mishra, a professor at Mary Lou Fulton Teachers College at Arizona State University. 鈥淭hink small pockets of innovation. Support those. Promote those and I think that you鈥檇 get more bang for the buck.鈥�
4. Make data privacy a top priority
LAUSD has laid aside 鈥淓d鈥� for now. But that move may not put to rest questions about how
A former AllHere software engineer warned district and state officials that the chatbot鈥攚hich required access to troves of student information to function鈥攚as violating the district鈥檚 privacy rules and putting some student data at risk,
In a statement to 澳门跑狗论坛, LAUSD said it takes data concerns seriously and will continue to protect students鈥� data.
鈥淭hroughout the development of the 鈥淓d鈥� platform, Los Angeles Unified has closely reviewed it to ensure compliance with applicable privacy laws and regulations, as well as Los Angeles Unified鈥檚 own data security and privacy policies, and AllHere is contractually obligated to do the same,鈥� a district spokesperson said in an emailed statement. 鈥淎ny student data belonging to the district and residing in the Ed platform will continue to be subject to the same privacy and data security protections, regardless of what happens to AllHere as a company.鈥�
Goel hopes districts will take this aspect of LAUSD鈥檚 experience to heart.
鈥淵ou need to start worrying about data privacy from Day 1,鈥� Goel said. 鈥淥ne of the big problems right now is who owns the data? Where is it stored? Who has access to it?鈥�
Districts need to make the answer to those questions crystal clear in their contracts with vendors, 鈥� said Mishra of ASU. And they need to spell out 鈥減enalties if we find out that our data has been [misused],鈥� he added.
Otherwise, vendors 鈥渃ould play loose and fast with鈥� student information, he said. And if a district is building a new AI learning tool, it shouldn鈥檛 just rely on the company鈥檚 stated privacy policy, even if it鈥檚 comprehensive.
鈥淐ompanies shift ownership,鈥� Mishra said. 鈥淭erms of agreement suddenly vanish.鈥�
5. Don鈥檛 let LAUSD鈥檚 experience completely sour you on AI
Back in 2013, LAUSD was among the first districts to embrace a 1-to-1 computing initiative, distributing iPads to students across a broad swath of schools. The effort was plagued by poor implementation and possible conflicts of interest鈥攍eaving districts around the country wary of going 1-to-1.
But districts interested in exploring AI shouldn鈥檛 expect that their own experience will mirror LAUSD鈥檚鈥攑articularly if they proceed cautiously.
鈥淚 hope that LAUSD moving so fast, in such a big way, and so publicly when there was not much there [doesn鈥檛] discourage districts from ever thinking that they should do this,鈥� Dusseault said.
The technology is here to stay. Understanding and using it properly will be important in the future workforce, she said.
鈥淟AUSD is a cautionary tale on how to not address AI,鈥� Dusseault said. 鈥淏ut it does not mean that districts shouldn鈥檛 be contemplating how to support their students and staff with AI-enabled strategies.鈥�
