Written By Tsabita Vera Cyavrilla – Chief Operating Officer at Generation Girl
There is a question that keeps coming up in our training rooms, and by now I have heard it enough times that it no longer surprises me. It comes from university students, from teachers juggling lesson plans and administrative work, from professionals trying to figure out where AI fits into their daily workflow. The question is some version of this: “Why do I need to learn this? I already use AI. I just chat with it.”
Every time I hear it, I pause. The person asking is technically correct. They do use AI. And that is precisely where the gap begins.
Since June 2025, the nonprofit organization that I currently lead, Generation Girl, has been running an AI literacy initiative across Indonesia, in collaboration with AVPN and supported by Google.org and Asian Development Bank, targeting three groups: university students, workers, and teachers. The logic behind this is straightforward. Indonesia needs 9 million digital talents by 2030 to stay economically competitive, according to the Indonesian Ministry of Communications and Digital Affairs. Yet only 25% of engineering graduates and 35% of ICT graduates are women, based on UNDP 2024 data. The gender gap in STEM is a representation issue, of course, but it carries real economic weight as well, with consequences for national competitiveness. As AI becomes the new baseline for how work gets done, that gap risks widening into something much harder to close.
To date, we have trained more than 6,800 people across all 38 provinces in Indonesia through 900 hours of programming, with a participant satisfaction rate of 4.7 out of 5. Those numbers matter to us. They also do not capture the part of the work that occupies most of my thinking.
What occupies my thinking is the gap between using AI and actually understanding it. Between familiarity and trust.
When we started designing the program, we assumed the biggest barrier would be literacy. That people would come in not knowing what AI was, or feeling intimidated by the technology. That assumption held up in some places and fell apart in others.
Participants from larger cities, particularly those in Java, often arrived with some existing exposure. They had used ChatGPT. They had played around with image generators. The tools themselves did not intimidate them. What was missing was the critical layer underneath: the habit of questioning the output, checking its sources, understanding why the model responded the way it did, and knowing when not to use it at all. They were using AI the way most people use a search engine. Type something in, take the output at face value, move on.
Outside Java, the dynamic was different. Participants from provinces further afield often came in genuinely unaware of how many AI tools existed or what they were capable of. But they were, in many cases, more curious. More questions. More willingness to sit with uncertainty and explore. They treated AI as something to figure out rather than something they had already figured out.
Both groups, in different ways, illustrated the same underlying problem. Operating a tool is one skill.
Thinking with it, and about it, is another.
What shifted my perspective most was what happened when we moved away from instruction-based teaching toward project-based learning. We wanted participants to experience AI directly, through their own work, rather than through demonstrations. What they did with that space genuinely surprised us.
One participant used NotebookLM, a tool that lets you build a knowledge base from your own sources, to map out scholarship opportunities for her graduate studies. She fed it application guidelines, eligibility criteria, and deadlines from multiple programs, and asked it to help her compare them against her own profile. Her use of AI here was organizational rather than generative. She had already gathered the information. She was using AI to think more clearly about a decision that mattered to her.
A teacher did something similar. She had years of curriculum documents, ministry guidelines, and teaching resources scattered across different formats and folders. She used the same tool to consolidate them into a single knowledge base, then generated lesson plans from it. The outputs stayed close to her actual source material, which reduced the hallucination problem that plagues more open-ended AI use. She walked out of the session with something she could use the following Monday.
Neither of them had walked in expecting AI to do this for them. They left with a more accurate mental model of what AI is genuinely good at, which is also the foundation of knowing when to be skeptical of it.
That skepticism is, I would argue, the core of what we are trying to build. It is also the thing most AI literacy programs underinvest in.
There is a tendency in this space to treat AI education as a skills transfer problem. Teach people the tools, show them the workflows, and they will be equipped. Skills without critical judgment, however, leave people more exposed than they were before. An AI model trained primarily on data from certain geographies, languages, and demographics will reflect those biases in its outputs. A user who does not know this, and who trusts the output because it sounds confident and well-structured, has gained access to a tool without gaining the means to use it safely.
Computational thinking matters here in a way that is often overlooked. The difference between prompting AI with “write me a report on X” and walking it through a sequential, structured set of instructions changes the quality of the output significantly. The structured approach also requires the user to have already thought clearly about what they actually need. That thinking process is protective. It keeps the human in the loop.
I use AI this way myself. I rarely ask it to generate something from zero. I come in with a position, a set of ideas, a direction I am already moving in, and I use it as a sparring partner. When it hallucinates, which it does, I stop, work through the problem on my own, and come back to it with more structure. That friction is part of what working with AI honestly looks like.
For women in Indonesia specifically, the trust question carries additional weight. If AI systems are trained disproportionately on data that does not reflect their language, their context, or their lived realities, then the tools will quietly underserve them in ways that are hard to name but easy to feel. An output that sounds authoritative but does not quite fit your situation. A recommendation that assumes a professional context that does not match yours. A tool that works well in English and loses nuance in Bahasa Indonesia.
These are small frictions rather than dramatic failures, but they accumulate into a reasonable reluctance to rely on something. If we want women across Indonesia, including those in provinces where AI was not even on the radar twelve months ago, to genuinely benefit from this technology, access alone will not get us there.
We have to build the conditions for trust. That means teaching people how AI actually works in the background, with the goal of making them informed users rather than engineers. It means being honest about what the tools cannot do, and about whose data is and is not shaping the models they are using. It means creating space for questions that sound basic but are actually foundational: Is my data safe? Who benefits from me using this? What happens when it gets something wrong?
For other organizations running similar programs, and for the policymakers and technology companies whose decisions shape what these tools look like and who they serve, here is what I have come to believe: literacy is where this work begins, and the harder work is everything that comes after. Building the kind of critical, grounded relationship with AI that lets women use it on their own terms, for their own purposes, takes more than a curriculum and a set of tools. It takes time, honest conversation about what AI can and cannot do, and a willingness to keep adjusting as the technology shifts.
That is the work we are committed to. And it is far from finished.
About The Author
Tsabita Vera Cyavrilla is the Chief Operating Officer of Generation Girl, where she drives the organization’s strategy and operations, ensuring that programs reach and inspire thousands of future female leaders across Indonesia. She holds a double degree in Computer Science from the University of Indonesia and the University of Queensland, and combines technical acumen with a long-standing interest in design and innovation. Today, she leads cross-functional teams, builds strategic partnerships, and champions initiatives that scale Generation Girl’s mission of creating inclusive opportunities for young women to develop the skills, confidence, and leadership to thrive in technology.
Generation Girl is an Indonesian nonprofit focused on increasing women’s participation in STEM fields, with the goal of closing the gender gap and creating equal pathways for women to build careers in technology. As a nonprofit, the organization is committed to advancing the United Nations’ Sustainable Development Goals, particularly Quality Education, Gender Equality, and Reduced Inequalities. Since 2018, Generation Girl has built a community of more than 106,800 members across 38 provinces of Indonesia, and continues to expand its reach through programs spanning STEM education, mentorship, and AI literacy.
Learn more about Generation Girl at www.generationgirl.org