Voice of San Diego
Investigative journalism for a better San Diego
Sign up for The Morning Report with all your must-read news for the day.
In a strip mall in the sleepy San Diego neighborhood of Tierrasanta is one of the newest additions to San Diego’s education system.
It stands 6 foot 2 inches, with a dull gray silicone face. Under its bald, transparent skull, purple lights pulse. Its bright blue eyes blink and flit from side to side. Its brow scrunches and its lips purse, creating a disconcerting simulacrum of emotion. All the while, the exposed motors in its elbows and shoulders whirr.
Its name is Ameca, and it’s been touted by school officials as the “world’s most advanced AI-powered humanoid robot.” It’s one of two such ChatGPT-enabled robots purchased by local charter chain Altus Schools which, combined, cost them $500,000.
Officials at Altus are optimistic about the big dollar buys. They say the robots are part of a pilot program meant to help them figure out what role AI – and potentially robots – may play in the future of education.
But not everyone’s sold on the pitch. To some researchers, the promises of AI’s backers fall far short of its capabilities. For them, not only has research not sufficiently proven AI’s usefulness as an educational tool, but its introduction into schools also has the potential to do far more harm than good. That’s to say nothing about the humanoid robot part.
“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,” said Wayne Holmes, a professor of critical studies of artificial intelligence and education at University College London. “What we are increasingly hearing are bits of evidence that demonstrate the opposite.”
For Altus officials, humanoid robots are another example of their technological explorations.
Altus has long functioned as a pseudo-alternative school, offering students who have fallen behind the opportunity to fast-track classes and catch up. The chain’s credit recovery formula, particularly at its flagship location, Charter School of San Diego, has been lauded as helping students succeed at higher rates than similar programs.
Work is largely done by students independently, but the chain also operates a network of in-person resource centers that give students the opportunity to get one-on-one instruction. It’s in those resource centers that students encounter Altus’ new robots.
Learning Curve
Your biweekly update on the state of education in San Diego County schools.
In an email sent to families of Altus students, Cathryn Rambo, Altus’ dean of academic studies, touted the new purchase, which she described as an “innovative opportunity for your student to participate in a research-based learning experience.”
“We are thrilled to be the first school in the world researching the use of physical AI as a teaching partner,” Rambo wrote.
On a sunny June day, I got a front row-seat to see what one of the robots could do. Two middle school students sat in a small room at the back of Altus’ Tierrasanta Resource Center, the robot towering above them.
They were there to learn about the inventors of some of the technologies that are used in modern drones – they had chosen to learn Nikola Tesla. It was now the robot’s job to respond as the Serbian inventor.
“Can you tell us who you are and when you lived –” a middle school girl began to ask, before the robot interrupted.
“Great! What’s your first question for Nikola Tesla?” it responded out of turn.
“Can you tell us who you are, when you lived and what you are best known for inventing or discovering?” the girl tried again. After a brief pause, the robot obliged.
“Greetings, I am Nikola Tesla. I lived from 1856 to 1943. I am best known for inventing the alternating current AC electrical system, and my work on wireless communication and magnetic fields. My inventions have greatly influenced the modern world,” it said, at a rushed pace.
The interaction was rife with stops and starts. The robot would interrupt the children, or the children would interrupt the robot. Each interruption led to another pause.
Throughout it all, the robot spoke too quickly for the students. They ultimately asked it to repeat its Tesla introduction three additional times as they scribbled away dutifully.
Rambo ended with a reminder: always fact check what the robot tells you.
Afterwards, Rambo acknowledged the lesson, which she called “clunky,” didn’t go great. It was actually the first time she’d tried that particular one. But this was all part of the learning process, she said.
Rambo said that over the course of the pilot program, they hope they’ll see statistically meaningful educational improvements in the kids who’ve interacted with the robot, like increased standardized test scores. Officials have no set deadline for the pilot’s conclusion.
“As long as we keep learning and iterating, we’re going to continue,” Rambo said. “We didn’t even know in the first six weeks what Ameca can do, and we still don’t know all the different possibilities.”
And to Rambo, the possibilities seem magical. A disproportionate number of the students Altus serves are low-income, homeless or have disabilities. She thinks the impact of simply being exposed to technology as advanced as a robot could be transformative.
“We’re talking about breaking poverty cycle. If there’s one kid that comes and says ‘I want to be a robotics engineer…’” Rambo said, trailing off as she fought to hold back tears.
While they don’t have any hard data on performance, some of the preliminary feedback they’ve received is less than stellar. For example, before and after interacting with the robot, Altus officials ask students what three words they would use to describe the robot. “Creepy” is the number one word offered by students, Rambo said. Though by the end of their interactions, students sometimes drop that descriptor.
“This is a long-term investment,” added Jay Garrity, Altus’ head of instructional services. “We’re still at the beginning phases, but we are optimistic in terms of the prospects it might bring to teaching and learning in the future.”
For researchers like Holmes, the University College London professor, it’s not surprising that schools are experimenting with AI. After all, it’s sexy and the platforms are supported by a nearly bottomless pit of funds. But despite the hype, he feels Garrity’s optimism is misplaced.
“The suggestion that this could ever be within a million miles of the capabilities of a human teacher is criminal,” Holmes said. “For this to be peddled as something that can teach English language arts, that can teach vocabulary, that can engage with education in these ways, it’s utter nonsense. It’s a parody of education.”
Rambo and Garrity insist that they will not use AI to replace people. The connection between students and teachers is an important part of Altus, they said. For them, the robots are more about finding new ways to engage students.
But in the decades since the introduction of rudimentary AI programs into classrooms, that pledge has been made over and over, said Monash University Professor Neil Selwyn. He’s spent decades researching the benefits and pitfalls of integrating technology into classrooms and in 2019 released the book “Should Robots Replace Teachers?”
But in all that time, a basic truth has remained: AI technologies don’t make teachers’ lives easier, they create more work. That may mean endless troubleshooting of glitches, correcting misstatements, inputting lessons or fine-tuning prompting. In any case, it’s a de-skilled version of what teachers actually do that is far less beneficial to students.
Selwyn is even less convinced of the benefit of a humanoid robot.
“The physical robot in the room is just complete bullshit – a show, a charade, a spectacle,” Selwyn said. “The distraction of the humanoid robot is mad in this particular case, because we haven’t got physical robots that can be anywhere near as sentient as a teacher… so much can go wrong.”
Beyond the pedagogical concerns, since being installed in January, one of Altus’ robots has already hit a series of technical problems, albeit minor ones. There are also endless questions about how to decide exactly what the robot should be allowed to do.
One of those decisions was whether the robot should be able to imitate people. Engineers at Engineered Arts, the firm that designed the robot, restricted this feature because of its potential to be a “gateway into inappropriate content.” Per Altus officials’ request, though, they altered the instructions.
In an email, the engineer wrote that that some figures were still off limits – like “certain rappers (Tyler the Creator), criminals (Martin Shkreli), or other people who are likely to say offensive things.” But the robot would now be able to impersonate people with “gray areas,” like Tupac or former President Bill Clinton.
The robot was free to discuss more “neutral things” about Clinton’s presidency, like his foreign policy positions. But it was programmed to be cagier about others, like Clinton’s illicit sexual relationship with White House intern Monica Lewinsky.
Similarly, when asked about what he talked about in his songs, the robot, speaking as Tupac, responded in vague, airbrushed generalities.
“I talk about life experiences, struggles, social justice, and hope for change. I aim to inspire and connect with people through real stories and emotions,” the robot replied.
But stiff necks and prompting issues are comparatively tame challenges – at least compared to the open question of how to make an AI chatbot, let alone an AI-powered robot, safe for children. That question is one that even engineers at OpenAI, the developers of ChatGPT, haven’t figured out.
As chatbots have exploded in popularity, so have cases of what’s been referred to as AI-induced psychosis. While not a medical diagnosis, the phrase refers to instances in which AI-users have developed psychotic tendencies in tandem with chatbots or have had delusional thinking validated and reinforced by chatbots’ typically sycophantic responses.
In some serious instances, chatbot users in the throes of AI-induced psychosis have committed suicide. Children are particularly susceptible to developing unhealthy relationships with AI-powered chatbots.
That grim reality is why one of Altus’ proposed uses for Ameca is especially concerning to Holmes. The robot is able to shift into four “personas,” each meant to serve a different function. Those are “Sage the Teacher, Remi the wellness coach, Ari the college and career planner, and Lexi the translator,” Rambo wrote in an email to families.
To Holmes, the prospect of an AI-powered robot being used to aid student mental health, is particularly disturbing.
“The idea of using a robot or an AI system, or a merging of the two, for mental health support, I think is frightening beyond all words,” Holmes said.
Here, again, Altus officials stress that they don’t plan to replace mental health counselors with robots. Instead of traditional mental health services, they imagine Ameca’s wellness coach persona as something that can offer test prep recommendations to students dealing with finals anxiety. Rambo, for example, likes to have the robot give her an inspirational quote or affirmation that can help get her through the day.
“If a student is upset about an argument with a parent, we’re never going to put them in front of a robot,” Rambo said.
Rambo said the robot has robust safety features. And despite the earlier pledge of one-on-one instruction, children are not left alone with the machines. She also said they’ve sought to insulate it from data privacy concerns: it doesn’t record data and its memory is erased after every interaction.
But in an era of limited funds, every school expenditure is a value statement. To Sherry Turkle, a professor of the social studies of science and technology at MIT, an AI-powered humanoid robot seems like the worst use of those limited funds. Half a million dollars could go a long way toward purchasing healthy snacks in schools, or tutors, or mentors from kids’ community, she added.
Turkle, who’s latest book examines the impact chatbots have on users’ abilities to form relationships with other humans, said AI is just the most recent in a long line of educational technologies that have been pitched as potentially transformative.
A decade ago, the technology was screens and tablets. Now, parents are mobilizing nationwide to rid classrooms of screens. While chatbots are incredibly technologically advanced, Turkle predicts they will go the same way.
“Generative AI is a marvel,” Turkle said. “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.”
Something went wrong. Please refresh the page and/or try again.
Jakob McWhinney is Voice of San Diego's education reporter. More by Jakob McWhinney
Your email address will not be published. Required fields are marked *
This site uses Akismet to reduce spam. Learn how your comment data is processed.
Voice of San Diego is a registered trademark of Voice of San Diego, a California non-profit.
401 B Street, Suite 302, San Diego, CA 92101
Phone: 619-325-0525
We rely on people like you to support this service. Please donate or sponsor
Voice of San Diego® today.
DONATE NOW
Leave a Reply