I Built an AI Tool to Guide My Experiment. Now I Need to Test the Tool.

ESSAYS
1,600 words · 7 min read

As a doctoral student at the California Institute of Integral Studies, I’m conducting a three-year first-person inquiry into perception, and my lab is ordinary life.

For years, I have pursued a personal philosophical inquiry, formulating my own questions and choosing what I wanted to study. Using a consciousness-first working premise drawn from philosophy, spirituality, and science, I explore the possibility that consciousness is fundamental and that one mind is dreaming this world together. I’m not claiming this premise of oneness is true. Instead, I’m trying to understand what subtle shifts in perception actually do, including when they produce null results, go wrong, or reveal risks.

How much of what we take to be the world depends on the structures through which we see it?

Philosophy as a Designed Experiment.

I developed the principles, practices, and field tests. Over time, I hope to design a branching structure that would allow other philosophical, religious, or spiritual frameworks to be applied to similar day-to-day perception experiments. In the meantime, I’ve built a working frame called We The Dreamer, inspired by nondual texts that resonated with me. I have developed consciousness-first principles, practices, and a growing library of specific field tests in perception, relationships, and inner navigation. Each month of the three-year experiment, I pick a theme related to my working premise. For example, I’ve tested belonging, judgment, peace, and presence. I’m building a detailed log of findings, including psychological observations, phenomenological traces, and my interpretations.

For me, design is not simply a way of illustrating philosophical ideas or communicating their conclusions. It is a way of investigating them. I design conditions in which a philosophical premise can be encountered, practiced, and tested in ordinary life.

The design itself becomes part of the philosophical method.

A.I. as Protocol Allocator.

But here is where I use AI in an unusual way. I wanted to give my full attention to the perceptual experiment, retaining enough of the texture of lived experience to describe it afterward. Maintaining the daily protocols of a three-year inquiry also requires a considerable amount of organization. I began to see how design, assisted by AI, could take care of some of that work without taking over the philosophical inquiry.

AI does not formulate the philosophy or decide what the results mean.

Its role is much narrower. For example, I built an allocator that operates within rules I established and varies some of the conditions under which I test these practices each day.

I may already know which principle I’m testing, but the allocator assigns a related practice from my library and determines whether I use it proactively in the morning, wait until an ordinary situation calls for it reactively, sustain it for an hour, or return to it every two hours throughout the day. The tool randomizes some experimental conditions, while others are determined by contextual variables. For example, I can tell it I’m going to be home-bound that day, and it won’t choose a test that can only be conducted outside.

For instance, it might assign ten minutes of eyes-open meditation at the beginning of the day, before checking my emails or calendar, at home or during my commute. It could pair that activity with a practice asking me to reconsider how I perceive whatever comes into awareness. I then conduct the experiment myself and write a brief psychological and phenomenological account of what I noticed.

Why do that? Originally, I was struggling to remember the practices. The hypothesis I’m testing is so radical, that one mind might be dreaming this world together, that I wasn’t always sure which practice to choose.

But I also realized I needed a way to reduce some of my discretion over the conditions of the experiment without giving up control of the inquiry itself.

If I decide every morning exactly when and how to test something, I may unconsciously keep choosing conditions in which I expect it to work.

For a solo researcher, having AI function like a protocol allocator is a remarkable opportunity. I don’t have a team following me, assigning conditions, and helping me sustain the experiment.

A tool like this allows one person to maintain a structured longitudinal inquiry that might otherwise require research assistance and significant funding.

And then there’s the gradual decay of energy and motivation that a three-year endeavor can bring. I built a simple web-based tool and placed a shortcut on my phone’s home screen. My phone is usually the first thing I reach for in the morning. Now, the first thing I open is my experiment’s premise protocol.

With more than 500 field tests already, each documenting a lived experiment in perception, I use AI platforms to retrieve data and organize material from my growing experiment log.

But an opportunity like this shouldn’t be taken for granted. One of the methodological questions it raises is, even if I remain responsible for interpretation, might the tools used to retrieve and select observations influence which patterns I notice?

Powerful tools need scrutiny precisely because they make things easier.

I joined CIIS because I wanted PCC to help me discipline this living inquiry. I know that from parenting, too. My wife and I are much better equipped with our second child because of everything we learned about screen habits with our older daughter, who grew up while we were all discovering, in real time, much of what could go wrong with social media.

When the Instrument Shapes the Experience.

This opportunity raises several methodological problems.

One of them concerns the allocator itself. Perhaps it reduces one source of bias, but does it introduce another? Randomizing conditions does not eliminate bias, particularly when the person designing the protocol is also the participant and interpreter. At most, it reduces some of my day-to-day discretion.

A technological system like this might determine when, under what conditions, and at what pace I maintain the experiment, but ultimately, it is also part of the conditions producing the experience so, how much of what I observe, therefore, belongs to the way the experiment has been staged?

I’m exploring all this through a first-person, one-participant experiment. But if I wanted to increase the number of participants and make it a collective experiment, I would likely need a group to define simpler, more consistent experimental conditions.

Another problem is that if I use AI to limit my own decisions, how do I audit how much agency I have delegated to it?

One of the larger ideas in my work is that perception may be the first creative act.

We tend to imagine that we see something and then decide what we think about it. But by the time we experience our first reading, it has already been shaped by language, memory, family, culture, social media, news, advertising, and propaganda. AI makes that old problem unusually visible.

And of course, this isn’t only an AI problem. Every method affects what it allows us to notice. We see this in phenomenology and psychology. A question does. A meditation instruction does. A psychedelic experience does. A research protocol does. Even language and terminology do.

And so I find it remarkably relevant that, alongside my research in perception, I also end up asking a larger question that I’m not trying to answer here: if our first reading, our first interpretation, perhaps perception itself, can be shifted deliberately, what is already shaping it before we notice?

I’ve decided to let this technology play an important mediating role in the experiment. But I want to make its mediation visible enough that I can do the philosophical work while keeping the layers separate. There is what happened. There is how I experienced it. There is how I interpreted it. And then there is what I might claim about reality. I’m not claiming that my private experience proves metaphysics.

Keeping the lived effect separate from the ontological claim is one of the central requirements of this inquiry, and the technological scaffolding makes that distinction more urgent.

On many levels, it strengthens my capacity for attention. But if I eventually want to turn this into a pedagogy of perception, that scaffolding may need to become unnecessary. The goal would be for the perception experimenter, whoever that might be, to internalize the practices and become independent of the technology, however long that takes.

Toward Independence.

I have had a recent conversation with Prof. Matt Segall, my doctoral advisor, who suggested to investigate what happens if I suspend the AI scaffolding.

I hadn’t considered that removing technological mediation might tell me something about the agency I had delegated to AI.

I’m going to make that an experimental condition and, gradually, I’ll extend periods without tech. For one month at the 12th month of the experiment, two months at the 24th month, and eventually spending the final three months of the three-year experiment without it. If the practices remain available without the prompts, perhaps something has actually been internalized. If they disappear with the technology, that tells me something equally important.

There is the environmental question of running AI systems. I’m, therefore, exploring the idea of gradually lightening the need for AI for experimental infrastructure. Depending on whether resources allow, a conventional software could take over its daily operation. That possibility is distinct from technology-free periods, which test whether the practices themselves can become independent of external prompts.

So for me, the opportunity is that one researcher can now construct an experimental infrastructure that would once have been difficult to sustain alone. The risk is that the infrastructure may become part of what is producing the phenomenon.

If our first readings are always already being shaped, how do we regain some authorship over perception without simply installing another system that shapes it for us?

That’s the tension I’m interested in.