These days, it seems like nearly every topic finds its way back to AI, somehow. Because of the rapid rate of progress, it’s difficult to keep up and this learning edge is a place ripe with hope and caution. In the education sector, schools and districts are at various levels of understanding and implementation, from outright bans over legitimate concerns around privacy and security, to policy creation, to fully embracing AI use. We see three main use cases in the present iteration of AI around efficiency, design and learning. EdTech Insiders just released their comprehensive real-time map of AI products and articulate six major use cases and, while the proliferation of AI-assisted tools can be overwhelming, we must remain attentive to what’s on the horizon — the evolution of assistants to agents. As Ethan Mollick writes in his recent blog, “There seems to be near-universal belief in AI that agents are the next big thing.”
AI Assistants
AI assistants help teachers gain efficiency with operational tasks through direct prompting – drafting comments, emails, agendas, etc. A number of these assistant apps support the design of lesson plans, projects, and units. Other apps help to differentiate those units. And, the elusive goal of a 1:1 tutor for every student gets closer every week as large language models support learning functions (see Google’s LearnLM model designed specifically for education or research assistant Google Notebook LM). As AI becomes faster and increases memory and recall of previous interactions, these tutors and assistants may help reach millions of students who are not being served (think reading and math core skills, writing support, etc. where frequent and specific feedback loops increase outcomes).
So, an avalanche of edtech products meet these needs of efficiency, design, and learning. Given the appropriate level of security and privacy, these should increase the amount of time a teacher can spend building relationships, designing real-world learning experiences, and developing durable skills. These durable skills, including critical thinking, creativity, curiosity, collaboratio,n and emotional intelligence, while perhaps supported by AI tools, will be best developed through real-world interactions with real people and real phenomena. More time for relationships, real-world experiences, and durable skill development are good for teaching and learning. We also know that these durable skills are in demand by employers and are a better predictor of success than scores on typical standardized measurements used in schools. We argue that while a knowledge base and development of core skills still remain important, they must be balanced with well-developed technical and durable skills.
AI Agents
However, these first-generation AI advances in education, while important and (mostly) helpful, are not what will radically shift teaching and learning. At this point, the vast majority of AI-supported technology solutions work as assistants and serve as support and collaboration in partnership and under the direction of the educator. Review this rubric and make it more precise. Evaluate this paper and provide specific feedback. Given these recommendations, write a report for my department head that I can edit. etc. Input is provided by a human and the AI assistant processes the request through a large-language model and reports an output.
More recently, an increasing number of AI agents emerged. These agents are “a system or program that is capable of autonomously performing tasks on behalf of a user or another system by designing its workflow and utilizing available tools.” Customer service, self-driving vehicles, financial analysis, data analysis, emergency response, and healthcare all benefit from AI agents. These agents can operate without direction in many tasks once started. So what could this emerging technology look like for education?
AI Agent Benefits
AI Agents will continue to increase the amount of time available to build relationships with students, help students develop durable skills, and design and implement real-world experiences (which may always require human leadership and cognition).
Not reaching every learner? Imagine that as an educator you can “assign” AI agents to work with a number of your students in a personalized learning classroom. They do their work and make decisions independently of the teacher.
Overwhelmed by writing recommendations? Imagine you had notes around each student for which you needed a recommendation. You could direct your agent to write up each recommendation, draft up the emails, attach the recommendation, and save them to your draft folder for your review.
Need more frequent feedback for your students? Already, AI assistants can read an essay and provide feedback using a supplied rubric. Using an AI agent, one could prepare an assignment, build the rubric, send it to students via email or an LMS, and provide as much feedback as needed on the assignment until the student reaches proficiency. Meanwhile, they are also giving the teachers daily reports on which students need direct intervention and support from the teacher.
This is just the beginning. Imagine the many other opportunities that will become possible over the next 12-24 months, just as the suite of AI assistant tools became available over the last 12-24 months.
AI Agent Challenges
Given the inverse amount of funding going into addressing the impacts of AI on education versus developing AI in general (and thus impacting education), there is less discussion on the challenges. AI Agents are outsourcing specialists. From a student perspective, any assignments that can be read by AI (meaning all text, video, audio, and images) can now be outsourced to an AI agent. Imagine a traditional college or high school syllabus, released ahead of time, and outsourced to an AI Agent by the student. The agent could read through the syllabus, create a set of documents and presentation decks (which are typical assignment outputs), title them appropriately, write the assignment based on the linked rubrics, and send a summary report to the student-semester class completed five minutes later.
Or, maybe, a high school student, despite the restrictive school policy, has an agent complete the assigned reading, highlight specific locations in the text, pull those quotes for an essay, create three draft,s and then self-assess and reflect against a rubric. Students will get full credit for an assignment fully completed by the AI agent. Imagine a high school teacher relegating all instructional responsibility over to the agent – build a class that teaches these students (see uploaded data) the following standards (see uploaded data). Reach out to these students via email and launch the course. Provide me weekly updates on their proficiency levels as you teach them these standards. Provide a final grade and comment on each student at the end of the semester. More time for relationships and real-world experiences? Yes, but will that happen within our current educational system which is hyper-focused on demonstrated proficiency in core knowledge and skills?
And these examples reflect benign human temptation over more nefarious use cases.
Preventing the Age of Cognitive Laziness
Agents are different from assistants. Assistants require human leadership, supervision, and decision-making. Agents do that for us. Cognitive laziness is one probable outcome when we relegate more and more current human tasks to agents. As the profoundly challenging job of education gets easier, the temptation will be to take a back seat. Humans are good at seeking the path of least resistance. As all of us in the education sector continue to help develop young people to thrive in a future world, and agents emerge at an exponential pace, we need to double down on developing learning that we as humans are uniquely positioned to do: build relationships, design extraordinary and challenging real-world experiences, and develop the durable skills necessary for human thriving. We live in the moment of assistants, but we need to prepare for a future of agents.
The post The Rise of AI Agents: More Time for Real-World Learning or the Era of Cognitive Laziness? appeared first on Getting Smart.