In the educational AI race, Merlyn Mind focuses on classrooms, not individual students

While a number of companies and nonprofits are experimenting with AI-powered tutors, Merlyn Mind is rolling out an AI-enabled assistant for teachers to use in the classroom.

The new tool can be installed on a teacher’s laptop, where it awaits voice commands to answer in-class questions, pull up lessons on platforms like PowerPoint, or send assignments and other materials to students. Merlyn Mind cofounder and CEO Satya Nitta, who previously worked on AI tools for learning at IBM Research, says the company purposely focused on technology for teachers to use, rather than for pupils to use on their own.

“The jury’s out there about whether [artificially] intelligent tutors will ever replace teachers in the wild,” he says.

The technology is essentially a combination of a learning tool that can generate or surface relevant educational materials and a general labor-saving device for teachers, who can use it to handle simple tasks on their laptops, similar to an assistant like Alexa or Siri.

Merlyn Mind develops its own large language models rather than using off-the-shelf models from companies like OpenAI or Google, which it says helps its technology run faster, cost less per query, and deliver answers more on point for teachers in classrooms. The software, which is currently in beta and slated for general availability by the end of the year, typically runs requests through multiple models, aimed at different tasks, such as filtering out school-inappropriate questions, answering science questions, or suggesting related-classroom activities. It can also help create lesson plans based on particular material, and school districts have expressed interest in having the AI explicitly trained based on their own curricular materials, Nitta says.

After early experiments with an Amazon Echo-style smart speaker, which saw eager (and sometimes mischievous) students pelt the AI system with queries, Merlyn Mind switched to a push-to-talk remote control to ensure teachers stay in control of the system. Many do allow students to pose questions under controlled circumstances, Nitta says, and the remote also doubles as a wireless pointing device for controlling the teacher laptop. Future versions of the tool are likely to integrate with classroom displays as well as teacher monitors, Nitta says.

So far, he says, teachers seem to activate the AI frequently, with the average teacher using the system roughly 30 times per school day, and some as many as 100 times per day. The technology is typically paid for by school districts, who pay about $300 per teacher per year for the software subscription. Nitta also envisions applications for the technology in building specialized AI for other workplaces, perhaps focusing on law or accounting.

He isn’t opposed to eventually offering some student-facing AI—the technology might be useful for, say, preparing for a test, he says—but he says he just hasn’t seen evidence that purely digital tutors can engage students the way a human teacher can, and that’s a critical factor in education, especially in the younger years.

“If a student’s already motivated—for instance, they have to pass an exam, [or] they have to get a job, they’ll learn from anything,” he says. “But if a student isn’t motivated, I don’t think an intelligent tutor is going to solve the problem.”

https://www.fastcompany.com/90996397/in-the-educational-ai-race-merlyn-mind-focuses-on-classrooms-not-individual-students?partner=rss&utm_source=rss&utm_medium=feed&utm_campaign=rss+fastcompany&utm_content=rss