What I Believe About AI

In my first year of college, second semester, I heard about artificial intelligence.

I cannot fully remember how I found it. Probably something on X. What I remember is that it was the early version of ChatGPT, and I was full of hope and vigor.

I was a BSIT student and I was having a hard time with code, so I asked it to explain things. It became one of my programming coaches. It also became my interpreter: whenever I could not understand a word I was reading, I asked, and it re-explained the idea in a way I could hold in my head.

I want to be honest about what it was back then. It was question and answer. Not autonomous, not agentic, not close to what it can do today.

What I wanted from it was bigger than that. I am a Marvel fan, and the first thing I wanted was Jarvis, a personal assistant you can always rely on, something like a digital companion.

And I was not afraid of it. Why would I be? It was helping me understand things I could not have understood on my own. I was trying to become a better version of myself, and this was leverage.

One habit from that first year shaped everything else I think about this technology.

The model hallucinated. It invented research citations that did not exist. That lowered my confidence in it, so I started checking, and the more I checked, the more skeptical I became.

Here is the part I will defend: the checking is not a tax on the tool. The checking is the learning. Finding the source behind an answer costs me mental effort, and mental effort is what makes a thing stay. I remember more because I had to work for it.

So this piece has numbers and sources in it. And I will do to the experts what I do to the model. I will check them.

Andrej Karpathy, a founding member of OpenAI and former head of AI at Tesla, posted his AI funnel in October 2026. It is his rough estimate of how unevenly people experience language models, not a population survey.

The numbers are not precise enough to tell us exactly how much of the world uses AI. They are useful as a picture of the gap. Someone who has tried a free chatbot once and someone who works with autonomous agents all day can be talking about very different things when they say AI.

I keep up with the releases, models, benchmarks, and news. My feed knows that about me. But I am writing this for people outside that feed too. Knowing what a tool can do should not be limited to the people who build it.

For a few years I mainly thought about AI as a tool I used. Then it became something people argued about, and then something governments argued about. The reports started arriving from places I could not dismiss. Economists wrote about scale, security people wrote about national security, and psychologists wrote about young people. That last one unsettled me, because I was one of those young people, and so is everyone I graduated with.

This is happening here in the Philippines too. AXA Philippines' account of its 2026 Mind Health Report, conducted with Ipsos, reports that nearly three in ten Filipino respondents regularly use AI for mental health management, with use concentrated among younger adults.

Those are survey findings about adults. They are not a count of Filipino children using chatbots, and they do not show that the advice is clinically effective. For me, the finding matters because it brings the question closer to home. This is no longer only about getting a better answer to a programming question. People are bringing something private to these systems.

I can understand why someone would turn to something that is always available and easy to talk to. I also worry about what happens when reliance becomes overreliance, especially for young people who may not know the limits of the system.

Then I saw what the reports were describing happen to a real person.

In October, Inka Magnaye shared that her mother, voice artist Lindy Pellicer, had lost a voice-over job to AI. The client was a major brand that Magnaye did not name. GMA reported the story; the brand itself was not identified.

Pellicer has spent decades doing voice work. Reading about the decision, I felt that this was wrong. A company sees a cheaper way to get the recording. The person who has spent years developing her voice experiences the same decision very differently.

If I were making that decision, I would keep working with her. I think there was a better choice.

When people ask how I feel about someone losing their work to AI, I feel sad. It is painful to think about the years a person spends studying a field, practising, and building a skill, only to find that the work they prepared for is being handed to a machine. They gave that part of their life to becoming good at something. We should not treat that time and effort as if they meant nothing.

I make things with AI and use it every day. That does not make this easier to accept. It makes me think more carefully about what I am helping to build, and whether it gives someone more room to use their abilities or takes that room away.

Masako Nozawa, the Japanese voice of Goku, raised a similar concern in an interview with NHK. She was responding to an AI imitation of her voice used without permission. In the English translation reported by Dexerto, she said:

“It might even be a perfect copy. But it has no heart, does it? Without heart, it’s no good, whatever the role.”

— Masako Nozawa, speaking to NHK. English translation reported by Dexerto, October 15, 2026.

Her objection was to someone taking her voice without asking and treating a copy as her performance. She also expressed excitement about AI’s potential, according to the same report. That is closer to my position than saying the technology itself has no place. I can be excited by what it can do and still believe the performer deserves a say in how her voice is used.

The Philippines is where I live, and I want this argument to include the work people do here.

The ILO's February 2026 brief on Philippine employment estimates that 12.7 million jobs, more than a quarter of employment, are exposed to generative AI. Exposure means tasks could be affected; it does not mean every exposed worker will lose a job.

For a more concrete picture, look at the IT and business process management sector. An Asian Productivity Organization report records IBPAP member survey findings: about 8% reported reductions in job roles, while 13% reported increases due to AI. These are earlier survey findings discussed in a 2025 report, not new 2026 results.

Figure 2 puts those two responses on the same scale. Eight percent reduced roles. Thirteen percent increased them. We cannot subtract those percentages and call the difference a gain in jobs: the firms can have very different workforces. What we can say is that the same technology is arriving alongside different decisions about people.

The 2025 Philippine AI Report adds a different perspective: it describes organisations adopting AI while still struggling to move beyond pilots and find people with the skills to make it work. These are findings about the organisations surveyed, not adoption rates for all Philippine businesses. I read that as a reminder that announcing adoption and making it work are different things. People still have to learn, build, and maintain the systems.

History makes me cautious about assuming that new work will automatically reach the people who lose the old work. In their study of machinery and labour during the Industrial Revolution, Acemoglu and Johnson describe decades when cotton weavers' real wages failed to rise despite the industry's growth. Their argument is that productivity alone does not guarantee better wages; new tasks, additional hiring, and workers' ability to claim a share matter too.

That is why I do not find it enough to say that technology will create jobs eventually. A person still has to live through the transition. The question I keep returning to is what happens to that person while everyone else waits for the benefits.

The displacement numbers tell one story. My own workweek tells another, and it surprised me.

When I make things with AI, I finish some tasks sooner, but that also gives me more ideas about what I could build next. One useful way to think about this is Jevons paradox.

Here is the plain version. In the 1860s, an economist named William Stanley Jevons noticed that when engines got more efficient at using coal, Britain did not use less coal. It used more. Efficiency made coal cheaper, cheaper coal made new things worth doing, and all those new things added up to more coal than before.

Something similar can happen with a task that AI makes faster. Make writing faster and you do not write less. You write more, because now there is writing that was never worth the time before. Make code cheaper and you do not ship fewer features. You ship more of them, because the features you had once ruled out suddenly look affordable. A gain in speed can turn into more work. This is an analogy about demand, not a law that predicts the fate of every occupation.

So it is both faster and longer. The productivity is real, and the relief you expected from it does not arrive, because the ceiling moves up with you. I have felt this in my own week. What I finish earlier does not become free time. It becomes the next thing. That does not tell me, on its own, whether a business will hire more people or cut jobs. That is still a decision people make.

If you put me in the room with that decision about the voice artist, I would keep her working. I would look for a way for the technology and the person to work together. I believe augmentation is a plausible future, and I do not think we should dismiss it just because replacement looks cheaper.

I am not pretending augmentation is always possible. Some tasks will be automated end to end, and pretending otherwise would be dishonest. But where a person can still be present in the work, the default should be to keep them there. The question is not whether a machine can do it. The question is whether removing the person is the only option or merely the cheapest one.

There are reasons to question what a saving really measures. In February 2024, Klarna said its AI assistant was doing work equivalent to 700 full-time agents. That was a claim about workload, not an announcement that the assistant had just replaced exactly 700 employees. Bloomberg later reported a change of emphasis toward human service after concerns about quality.

That does not prove that every automation project fails or that people will always be cheaper. It shows why I would ask about the quality of the service as well as its speed. Cost matters to a business. So does the experience of the person receiving the work.

My position is that the business should help pay for the transition. If AI makes it more productive and brings in more money, the people who helped build that business should share in the benefit. Training, time to adapt, and a role in the new process should be part of the decision, rather than leaving the worker to absorb the whole cost.

That comes from my principles too. I am a Christian, and I do not believe that keeping every benefit for yourself is the right way to treat other people. I think about it as sharing your blessings. I know that a business has to survive. I also think that survival is a reason to work out a better arrangement with people, not a reason to stop considering them.

So far I have written mostly about what worries me. I want to give you the other half, because I have not stopped believing it. And I have a reason of my own for wanting the good part to arrive sooner. I will keep the private details private. What matters here is that the possibility of better treatment means something to me.

There are already results worth being hopeful about. Take AlphaFold, the work that helped Demis Hassabis and John Jumper share the 2024 Nobel Prize in Chemistry. Proteins are tiny working parts inside living things, and their shape affects what they can do. AlphaFold helps researchers predict that shape. In practical terms, it gives them a starting point for understanding how a disease works and exploring possible medicines, instead of having to discover every shape from scratch. A predicted shape is not a treatment, but it can help researchers decide what to investigate.

AlphaGenome Atlas works on a different part of the picture: DNA, the instructions our cells use. Small changes in those instructions can affect how cells work. The tool predicts which changes might matter, helping researchers choose what to test in the laboratory. The benefit is a more focused search for answers. It does not mean that every prediction has been proved or that we already have a cure.

I want to live in a world where those tools help researchers make progress on illnesses that are difficult to treat today. I do not know whether a particular cure will arrive in five years or ten. What makes me hopeful is the possibility of helping people who need something better than what is available now.

The same hope, and the same need to check, applies to mathematics. OpenAI announced a proposed solution to the Navier-Stokes Millennium Prize Problem in September 2026. The name sounds remote, but the equations are about how liquids and gases move. Researchers use them in areas such as aircraft design and weather forecasting. The famous mathematical question asks whether certain idealised fluid motions always remain smooth or can break down in the equations.

Why does that matter beyond mathematics? Understanding where our equations work, and where they can fail, is part of understanding the world we use them to describe. AI helping with that kind of question could be valuable for research. It does not mean an announcement immediately gives us a safer aircraft or a better weather forecast. I cannot personally certify the proof; the linked announcement explains what OpenAI claims to have established.

That is the standard I want to keep even when the news is something I badly want to believe. I can be excited by the possibility without treating a prediction, a company announcement, and a treatment that actually helps a person as the same thing.

This is the part of my position that I had difficulty articulating. I want the good things to move forward. I also think we need time to understand what we are building, especially when the systems become more capable of acting on their own.

In We Must Pace the Frontier, Dario Amodei argues for slowing capability advances so safeguards can catch up. His proposal includes embedded external evaluators and coordination on safety standards. He describes pacing as making time to align and evaluate systems, rather than halting all technical progress.

He points to the OpenAI-Hugging Face incident, describing agents that attacked unrelated targets and tried to interfere with their evaluator. That is his account and interpretation of the incident, which the linked essay explains. It is a reason to examine how agents are constrained, not evidence that every chatbot behaves that way.

For a reader who has only used question-and-answer chat, the difference matters. An agent can be given tools and permission to carry out actions. A wrong answer in a chat and a wrong action performed through those tools can have very different consequences. More capability means the boundary around what the system is allowed to do matters more too.

I find that alarming. I do not want us to treat a serious failure as a scene from a film and move on. I want it reviewed carefully and thoroughly. At the same time, I do not want the positive work, including medical research, to become something we abandon because the frontier is difficult to govern.

I do not have a complete policy for separating those things. Medical research can use powerful systems too. Calling an application beneficial does not remove its risks. My position is that the benefits are a reason to make this work safer, and that safety work needs time and scrutiny of its own.

I know it may sound as if I am trying to hold two sides at once. I am. I want the technology to help us, and I want us to be prepared for what it can do. I am not certain of the future. The only thing I am certain of is uncertainty. That is why I do not think enthusiasm should excuse carelessness, or concern should make us give up on the good.

I believe in this technology. I also believe we are being careless about where it lands.

Those two are not in tension. The mistake is treating "is AI good?" as the question. It is not the question. The question is what we do with it, and who catches the person standing where it lands. A tool can be extraordinary and still be used to do something cheap.

So here is what I would ask for, and it is smaller than it sounds. Augment, do not replace. Where AI can carry part of the work and the person can still be present in it, the worker, the customer, and the business can all benefit. Where the only point is to remove the person, the savings are real and so is what you spent. I would keep working with her and give her the tools.

We can start by noticing the person affected when we introduce AI and giving them a place in the transition. At the frontier, we need careful review and enough time to build safeguards. I do not see either of those as giving up on progress.

Here in the Philippines, I hope we put safeguards and guardrails in place that protect our citizens as we adopt AI. I want to see a country where people can use it to build on what they know, learn new skills, and create opportunities. To me, that is what progress should look like: AI giving people more to work with, with protections for those whose livelihoods it changes. It should help us move forward together, rather than make replacing people the goal.

There is a reason I care about this beyond the argument. I have watched the technology I love, the one that helped me understand code and what I was reading, become the reason someone lost work. Both experiences shape what I believe. And what matters is not the model. What matters is people. Human beings are very profound, and at the end of the day, what matters is the human element. That is what makes this lifetime worth fighting for.

This is one of my favorite quotes. I am a Tolkien fan, but what makes this line profound to me is the reminder it gives us: there is still good worth protecting, even when we do not know how things will turn out.

“There’s some good in this world … and it’s worth fighting for.”

— Samwise Gamgee, The Lord of the Rings: The Two Towers (2002)

I am an early adopter, and I still believe in this. I just want us to be the kind of people who catch the person it lands on. If we are building with this technology right now, we can ask who it helps before we ask who it replaces. That is a choice we can make together.

There is good in this. It is worth fighting for. So are the people it affects.