San Diego School Spent $500K on Robot Tutors. Reading Science Disagrees.
Last updated:
When a school that primarily serves low-income, homeless, and disabled students spends half a million dollars on six-foot robots with silicone faces, it is worth asking who, exactly, that money was meant to help.
Altus Schools, a San Diego charter chain known for credit recovery, purchased two ChatGPT-powered Ameca humanoid robots for a combined $500,000 and installed them in its resource centers in January. School officials call the initiative the “first school in the world researching the use of physical AI as a teaching partner.” Three independent education researchers, from University College London, Monash University, and MIT, reviewed the same premise and used words like “parody,” “charade,” and “delirium.”
The research on what actually changes outcomes for struggling learners has been consistent for decades. It does not look like this.
TL;DR
Altus Schools in San Diego spent $500,000 on two ChatGPT-powered Ameca humanoid robots; students’ top descriptor is “creepy” and no performance data has been collected.
Three researchers from UCL, Monash University, and MIT independently described the initiative as unsupported by evidence, a spectacle, and a delirium with technology.
Reading science (Rayner, Seidenberg, IDA 2025, Shaywitz/Temple brain imaging) shows the interventions that rewire struggling readers’ brains are targeted, multi-system, and structured, not conversational AI.
The students Altus primarily serves are low-income, homeless, and disabled, the population that benefits most from targeted intervention and loses most when resources go to unproven technology.
$500,000 directed toward one-on-one tutoring from instructors trained in structured literacy would serve the same students with a far stronger evidence base.
A San Diego charter school spent $500,000 on two humanoid AI robots for students already behind. Three leading researchers called it theater. Here is what the reading science says should have happened instead.
Common questions
Are AI tutors ever useful for kids who struggle to learn?
Some AI tools support specific things, such as structured phonics or vocabulary practice within a well-designed program. But conversational AI robots are not the same as deficit-targeted instruction. For children who struggle with reading, the bottleneck is typically phonological processing, auditory processing, or working memory. Systematic practice addressing those specific systems is what the research shows produces change. Ask what processing gap an AI tool claims to address before trusting the personalized-learning label.
What does the research actually show works for struggling readers?
Decades of converging evidence points the same direction: explicit, systematic instruction targeting phonological awareness and the related processing systems that reading draws on. Rayner’s eye-tracking studies show skilled readers process nearly every letter. Seidenberg’s cognitive linguistics research identifies guessing-based approaches as characteristic of poor readers. The IDA’s 2025 definition confirms reading draws on multiple systems simultaneously. Brain imaging studies at Yale and Stanford (Shaywitz, Temple) show the right intervention physically changes reading pathways. Novelty and engagement alone do not produce those changes.
Should I be concerned if my child’s school is adopting AI learning tools?
Not necessarily, but asking better questions is warranted. What specific processing deficit does this tool target? Is there independent research on outcomes for students with my child’s profile? And would the same dollars fund something with a stronger evidence base? AI tools range from well-designed phonics apps to expensive spectacles with no peer-reviewed support. Those questions are the advocacy your child needs from you.
My child is behind in school. Where do we start to figure out what actually helps?
Start with the processing gap. A screener identifies which systems are falling short, whether phonological processing, auditory processing, working memory, or others, so you know what kind of targeted practice your child needs. A screener is a starting point, not a diagnosis. If your child might need formal accommodations such as an IEP or 504 plan, or you suspect a vision, hearing, or medical cause, a professional evaluation is the route to those supports. Consistent practice addressing the identified systems is what brain imaging research shows produces real change.
What a San Diego charter chain just spent $500,000 on
Altus Schools bought two Ameca robots from British firm Engineered Arts. Each stands six feet two inches tall, with a dull gray silicone face designed to simulate emotion and exposed motors in the elbows and shoulders. The robots run on ChatGPT and shift between four educational “personas”: Sage the Teacher, Remi the Wellness Coach, Ari the College and Career Planner, and Lexi the Translator.
Cathryn Rambo, Altus’s dean of academic studies, wrote to families that the program was an “innovative opportunity for your student to participate in a research-based learning experience.” In a subsequent interview, Rambo added: “We didn’t even know in the first six weeks what Ameca can do, and we still don’t know all the different possibilities.”
An observed lesson in which two middle schoolers tried to learn about Nikola Tesla through the robot was described by Rambo herself as “clunky.” The robot spoke too fast; students asked it to repeat its introduction three additional times. One robot has already logged technical problems since the January launch. The most common word students use to describe the machines is “creepy.” No performance data has been collected; officials say there is no set deadline for the pilot’s end. “As long as we keep learning and iterating, we’re going to continue,” Rambo said.
Author Quote"
“There is no independent evidence at scale that the use of these tools is either effective or safe, or even have a positive impact on the classroom. What we are increasingly hearing are bits of evidence that demonstrate the opposite.” — Wayne Holmes, professor of critical studies of AI and education, University College London
"
What the coverage gets wrong
Most reporting on the Altus robot story stops at the price tag and the expert skepticism. The harder question, rarely asked, is what this particular student population actually needs. The students Altus serves are disproportionately low-income, homeless, and disabled, exactly the learners for whom targeted, evidence-based instruction moves outcomes most dramatically. Research by Keith Rayner, Mark Seidenberg, and the IDA’s 2025 multi-system definition shows that closing reading gaps requires systematic, deficit-specific instruction, not conversational AI. The $500,000 question is not “is the robot impressive?” but “does it address the processing gap?” On that question, the coverage has been largely silent.
Why the reading science points somewhere else entirely
The central problem here is not that the robots are unattractive or expensive. It is that engagement and instruction are not the same thing, and the gap between them is exactly where struggling learners get lost.
Cognitive scientist Mark Seidenberg’s research characterizes guessing-based approaches as descriptive of how poor readers read, not strong ones. Keith Rayner’s decades of eye-tracking research show that skilled readers process nearly every letter on a page; they build precision through systematic, targeted practice rather than filling gaps from context. The International Dyslexia Association’s 2025 definition made explicit that reading difficulties draw on multiple systems simultaneously: phonological processing, auditory processing, working memory, and processing speed. Shaywitz and Temple’s brain imaging studies at Yale and Stanford showed that the right intervention physically rewires reading pathways. “Right” means targeted at the specific system that is falling short. A ChatGPT conversation does not do that.
Wayne Holmes, a professor of critical studies of artificial intelligence and education at University College London, put it plainly: “There is no independent evidence at scale that the use of these tools is either effective or safe, or even have a positive impact on the classroom. What we are increasingly hearing are bits of evidence that demonstrate the opposite.” MIT’s Sherry Turkle was direct about what the hype conceals: “It’s not hallucinating, we are—by imagining that it’s going to solve problems that it can’t possibly solve. And that is about to do a lot of damage.” Neil Selwyn of Monash University, who spent decades studying technology in classrooms, called the physical robot “a show, a charade, a spectacle.”
Altus serves students who are disproportionately low-income, homeless, or living with disabilities. These are the students whose outcomes the right intervention would move most. The differential boost in special-education research is real: targeted support at the right moment lifts a struggling learner more than it lifts anyone else. That is a scaffold doing its job. A six-foot robot speaking too fast is not a scaffold. It is a spectacle funded by the students who needed a scaffold.
Key Takeaways:
1
Engagement is not instruction: Rayner’s eye-tracking and Seidenberg’s cognitive linguistics research show what moves struggling readers is systematic practice targeting specific processing gaps, not novel conversational experiences.
2
The evidence on AI in classrooms is thin: UCL professor Wayne Holmes says there is no independent evidence at scale that AI classroom tools are effective or safe, and early signals “demonstrate the opposite.”
3
Ask the right question: When schools introduce AI learning tools, parents should ask what specific processing deficit the tool targets and what independent research backs the outcome claim, not just whether students find it engaging.
What this means for your child and what to ask your school
When a school announces an AI tool for students who learn differently, that announcement is an invitation to ask better questions. Three are worth putting directly to any administrator: What specific processing gap does this tool target? What does the independent research show about its effectiveness for students with this profile? And what would the same dollars have funded if spent on approaches that already have an evidence base?
Sherry Turkle noted that $500,000 funds tutors or community mentors who deliver the kind of responsive, targeted instruction that brain research shows actually changes outcomes. For children who struggle with reading or other learning differences, the most powerful lever remains an informed adult who knows which specific system to work on and how. A parent-facing screener identifies which processing systems are falling short before any program is chosen. A screener is a starting point, not a diagnosis. If your child needs formal accommodations such as an IEP or 504 plan, or you suspect a vision, hearing, or medical cause, a professional evaluation is the route to those supports.
The encouraging reality: the brain imaging research Shaywitz and Temple conducted confirms that children with reading difficulties who receive the right intervention develop the same reading pathways as typical readers. “Right” has specific ingredients. It does not require an impressive-looking machine.
Author Quote"
“Generative AI is a marvel. But the fact that it’s a marvel means that we’ve gotten into a kind of delirium with it. It’s not hallucinating, we are—by imagining that it’s going to solve problems that it can’t possibly solve. And that is about to do a lot of damage.” — Sherry Turkle, professor of social studies of science and technology, MIT
"
Your child’s brain is not waiting for a more impressive gadget. The research on what actually rewires a struggling learner’s reading pathways has existed for decades; the gap is not in the technology, it is in the resources directed toward what the science already proved. The real obstacle is a system that keeps betting on engagement over instruction, spectacle over structure. If you want to know which processing systems your child needs support with and what targeted practice for those systems looks like, Learning Success All Access gives you that map in plain language built around your child’s specific profile. Start here.
Is Your Child Struggling in School?
Get Your FREE Personalized Learning Roadmap
Comprehensive assessment + instant access to research-backed strategies
