A note from our founder

Our Approach to AI

This is the story of how we came to use AI at Enabling Learning, and where we draw the line. In short, AI supports our educators; it never replaces them.

An educator first

I am an educator by trade. But even before I became one, while I was still at the university studying to become a teacher, I realized that my other passion, technology, was the perfect match for learning.

Since early in my career, I have used technology both in and out of the classroom, with children and with adult learners. I have created multimedia resources and instructional materials for multilingual classrooms. And ever since we started Enabling Learning, we have worked both in the classroom and online.

Even before AI, we were using automation. We built HTML5 activities and study paths, experiences designed to mimic the realities of the classroom, or the realities of a particular situation a learner would face.

The problem a book could not solve

For about the last ten years, much of our work has been in teacher preparation, with the future generation of bilingual and ESL teachers. One thing I say often in my conference presentations is that, just as there are challenges preparing children in the classroom, there are real challenges working with adult learners in higher education.

Many of my students worked during the day and came to school at night to complete their degrees as bilingual and ESL teachers. Many of them also had to prepare for their teacher certification exams. Some needed the BTLPT, the Spanish proficiency exam for bilingual certification. Others needed the TOEFL to demonstrate their proficiency in English.

I created additional sessions and workshops at the university, at different hours. But the reality is that many students could not come, because of their work schedules and last-minute changes in their lives.

So I wrote books for teacher preparation, for the bilingual supplemental exam and for the BTLPT. Those books were useful. But with the BTLPT in particular, what you are really doing is helping a learner develop proficiency in a second language, and that is very hard to do with a book alone, without the proper practice. That is why my workshops and face-to-face activities were so important and, like I said, so hard to schedule.

Then I started using online learning with these students as well. Beyond the books, I created online modules where learners could listen, answer multiple-choice questions, and practice. But the real challenge was writing and speaking. For those, you usually need a human expert to correct what you are saying, give you feedback, and guide you based on your performance.

When large language models arrived

So when large language models like the ones behind ChatGPT became available, in 2022 and 2023, for developers like us to build with, I saw a solution to the problem I had been trying to solve with these learners.

Used properly, these systems can keep what educators, following Vygotsky, call the more knowledgeable other always present for the learner. Not only during office hours, and not only for the learners who can make it to a workshop.

So we started writing our own code to connect these models to our courses. That work became Verónica.

What made this possible

I will not get too technical here, but I do want to share what helped us make this work.

✓Expert knowledge first. Verónica works like a tutor with an open book; the technical name is RAG. Before she answers, she looks things up in our own knowledge base, so her answers start from our expert content, not only from the AI’s general knowledge.

✓Rubrics and feedback designed by experts. When the AI evaluates a spoken or written response, it applies rubrics and feedback that educators designed.

✓More than one model. We no longer use only OpenAI. We work with several models, and in key places one model verifies the work of another.

✓Multilingual by nature. Large language models are inherently multilingual. That lets us work with learners in English, in Spanish, and in the other languages they bring with them.

Together, that is what has allowed us to bring all of this into our courses.

The first line of support, not the last

From my experience in multiple institutions, I see how faculty have to work with so many students, often without enough time in the day. And I see students who feel embarrassed to go and ask for help.

So what we have been building with Verónica is a system that gives the student that first line of support, using expert guidance and expert design. Not to replace the expert, but to help scaffold the learner, and to identify the areas where only the expert can step in and provide the right support. For a learner studying at night, that means a place to practice and ask questions without embarrassment, and a person to turn to when the AI is not enough.

AI can also process a lot of data in seconds. Verónica can read a student’s performance over weeks and months of practice and build a profile that can serve the professor, the counselor, or the support system of an institution. In our case, the system can even flag when a student may need to meet with me over Zoom, or when to connect them with another expert on our team.

Our hard line

We never try to replace humans with AI.Humans are always going to be there.

What we are trying to do is enhance the work of these humans, take the repetitive load off them, and create the possibility for them to step in and interact when it is really needed, in the areas where they matter most.

How we keep our AI accurate

Good AI in education is not something you set up once and leave alone. We keep reviewing it, and we keep human experts close to it.

✓We test the AI against the real standard. For the BTLPT, we check our AI scoring against the official sample responses published by the state, and a daily check confirms that it stays aligned.

✓We choose accuracy over cost. To transcribe a learner’s speech, we use the model that proved most accurate with accented, second-language speech, even though it costs more.

✓A human expert is always within reach. Learners can see where they stand at any moment. If they have a question about a score, or something does not work the way they expected, they can tell us, and one of our experts reviews it.

✓We are open about what we store. Our AI Terms explain it.

Over the years, we have learned how important it is to keep reviewing our systems and to keep humans in the loop. We keep updating our courses and tools, and we read the feedback from the learners and educators who use them. That feedback guides what we improve next.

From evidence to a credential

The same idea guides how we measure progress. Our system uses AI to evaluate each learner’s progress and proficiency against our requirements, using rubrics designed by experts. Only learners who meet those requirements earn a verified digital credential, issued through POK.

Why we use digital credentials →

Why I am excited, and why we keep working

I am excited about AI. I know there are challenges and risks, and there is a lot of talk right now about AI systems going wrong. But from my perspective, and from the feedback we have been getting from the students and educators who use our courses and Verónica, there are undeniable benefits to these systems when they are used correctly.

That is where I have been investing so much of my time: building systems that use the proper data. In the end, these AI systems are guided by humans. So we keep putting in the time and work to make them as self-contained as possible, grounded in our own content, and very powerful because a human expert is always in the loop.

If you are an educator working through these same questions, I would like to hear from you. Write to us.

Luis J. Hernández
Founder, Enabling Learning

For the legal details of how we use AI, including what we store and why, see our AI Terms.