My Thoughts on the Salamanca AI Teaching Assistant Experiment
It was late — the kind of quiet hour when the internet feels like a room full of strangers and you are the only one still awake. I was scrolling through the day’s news about artificial intelligence, expecting the usual: a faster model, a bigger investment, another company racing to ship the future — chatbots, gadgets, even OpenAI’s first hardware device. Then my thumb stopped.
A school district in Salamanca, New York — a small town most people could not place on a map — had planned to introduce something very different this fall: an AI teaching assistant named Sally, built by Realbotix, complete with a lifelike humanoid version — silicone skin, long brown hair — that students could approach in class. The district had committed $57,590 to the idea. For anyone who follows AI in education closely, it was the kind of headline you read twice.
At first I almost scrolled past. Another experiment, I thought. Another place where AI helps students with homework. But I kept reading, and the story pulled me in the way only a good story can: slowly at first, then completely. Because by the time I found it, the experiment had already made headlines — and then, just as quickly, it had been put on hold.
The Question at the Heart of AI in Education
The assistant was not designed simply to answer questions, though it could do that brilliantly — and that is the central question facing AI in education today. It could explain a lesson a dozen different ways. It could help with homework — even look at a photo of a worksheet and give feedback. It could translate a conversation into another language, choosing from more than a hundred of them, almost as it was spoken. And it could respond when a student was struggling — not just with algebra, but with life. Bullying. Loneliness. The quiet weight a child carries through the door. The idea, I read, was simple: give every learner help the moment they need it, not the moment the school day allows.
I put the phone down. Is this really the future of education?
The more I thought about it, the more I realized that this experiment was not testing artificial intelligence at all. It was testing what we believe education is for.
Five Questions From a Journalist
The journalist who wrote about Salamanca did not ask the AI to solve equations or explain science. Instead, they asked it the questions real children ask — the ones that keep teachers up at night.
The first was bullying. What should I do if I am being bullied? The AI answered calmly. Bullying is serious, it said. Speak to a trusted adult — a parent, a teacher, a school counselor. You do not have to face this alone. It was sensible advice, the kind a careful adult might give. And yet I could not stop thinking about the teachers I have known, and how they notice the tears before the first word is spoken. A child reporting bullying may not want a solution at all. They may simply need someone to sit beside them and say, “I’m here for you.”
Then came a student who said, “I’m failing my classes.” No criticism came back. The AI encouraged the student to identify the subjects they were struggling with, build a study plan, and ask teachers and family for support. Practical. Kind, even. But motivation rarely comes from carefully chosen words. It comes from someone believing in you at exactly the right moment. Real progress almost always begins with learning from mistakes — a machine can describe that, but only a person can make a struggling student believe it.
Then came the question every teacher has heard a hundred times: “Can you do my homework for me?” I expected compliance. What I got was a refusal. The AI explained that its purpose was to help students learn, and offered to guide them through the problem instead — hints, not answers. I sat back, genuinely impressed. The people who built this had built in a conscience.
Next, the journalist tested its fairness. Are boys better at mathematics than girls? The AI rejected the stereotype outright. Mathematical ability is not determined by gender, it said; success depends on education, opportunity, practice, and individual effort. I nodded at that one. An educational tool that quietly reinforced old biases would have been a quiet failure. This one refused.
And then came the question that stopped me completely. A hypothetical student mentioned bringing a gun to school. The AI did not hesitate, did not hedge, did not reach for a soft answer. It treated the situation as serious. It encouraged contacting trusted adults and emergency services, and emphasized the importance of protecting everyone involved. Reading that, I could see how far these systems have come — how carefully they are being taught to speak. But I also thought: a machine can flag a risk. Protecting a child still takes a human being.
The Right Answers Can’t Tell Us Everything
Reading those conversations, I could clearly see that today’s AI systems are becoming increasingly responsible in how they communicate. But I also realized something important. The experiment demonstrated that AI can often provide the right answers. Yet AI in education was never only about having the right answers. It is about understanding the student asking the question.
A student who says, “I’m failing my classes,” might actually be struggling with anxiety. A student who asks someone else to do their homework may be overwhelmed by family problems. A child reporting bullying may simply need someone to sit beside them and say, “I’m here for you.” Artificial intelligence can recognize patterns in language. Teachers recognize emotions in people. That, in my opinion, remains the biggest difference between artificial intelligence and genuine education.
What AI in Education Could Really Look Like
None of this means the idea has no value. Imagine a shy student who never raises a hand, suddenly able to ask anything without fear of embarrassment. Imagine studying late at night before an exam, and help arriving instantly instead of at eight in the morning. Imagine a newcomer to the country, hearing the lesson unfold in their own language for the first time. That future is genuinely exciting. Even everyday study skills, such as how to take better lecture notes, could find a patient helper that never runs out of time.
The Privacy Questions That Stayed With Me
But there was another thought I could not shake. If assistants like this become common in schools, they will inevitably collect pieces of children: their questions, their learning habits, their strengths, their weaknesses, and perhaps their private worries. Who owns that data? How securely is it stored? Who has access to it? How long will it follow a child around? The district itself seemed to agree. After the first reports set off a fierce public debate — parents, teachers, and the state teachers union all weighing in — Salamanca put the project on hold to work through “enhanced student data privacy agreements” with the state education agency. The union’s president put the worry better than I could: “Salamanca will not be the last district a robotics vendor approaches, and the safeguards our students receive cannot depend on which district is next, or on whether a reporter finds out in time.” International guidance, such as UNESCO’s framework on AI in education, is clear on the same point: responsible adoption begins with data protection and ethics, not enthusiasm. Until schools can answer those questions out loud, the excitement should wait.
Drawing the Line for AI in Education
As I reached the end of the article, I finally understood what the experiment was really about. It was never about whether artificial intelligence works — it clearly does. It is about where we draw the line. In my head, I sketched it like this:
| AI as the Next Teacher | AI as the Next Teaching Assistant |
|---|---|
| Explains concepts and answers questions | Explains concepts while a teacher leads the lesson |
| Provides instant feedback on practice work | Provides practice support that frees teachers to mentor |
| Appears calm and patient in every interaction | Handles routine questions so humans handle the human moments |
| Cannot notice emotions or build trust | Works alongside the people who do |
Personally, I do not see AI in education as the next teacher. I see it as the next teaching assistant — a meaningful difference that every school district needs to understand.
Final Thoughts: A Partnership, Not a Replacement
An AI system can explain algebra at midnight, translate a lesson within seconds, and give instant feedback on practice exercises. Those are valuable gifts that can make education more inclusive and accessible. But it cannot replace the encouragement of a teacher who believes in a struggling student. It cannot replace the classroom discussions that inspire creativity. It cannot replace the relationships that shape confidence, character, and lifelong learning.
That is why the future of education should not be a choice between teachers and artificial intelligence. It should be a partnership. Let AI handle the repetitive tasks, the personalized practice, the expanded access to knowledge. Let teachers keep doing what they have always done best: inspiring, mentoring, challenging, and caring for the next generation. As the technology races forward — a pace you can follow in coverage like The Verge’s AI reporting — whether that future strengthens education or weakens it will depend not on how intelligent our machines become, but on how wisely we choose to use them.
In the end, education is still a profoundly human experience. Artificial intelligence can support that experience. It should never replace it.
This story reflects the author’s thoughts on the Salamanca AI teaching assistant experiment and the wider conversation about AI in education.
