Why I Pivoted from Machine Consciousness to Therapeutics

Most of the posts on this site are about consciousness, integrated information, transformer interpretability, and the question of whether anything that runs on silicon can have an inside. That was the work I was doing for years before this. I still find the questions interesting. I do not work on them anymore.

My older daughter was diagnosed in early 2026 with a 15q11.2 BP1–BP2 microduplication. She is five and non-verbal. The duplication is small — about 500 kilobases, four genes, on the paternal chromosome — but the phenotype it produces in her is significant. There is no approved therapy. There is no clinical program for this specific indication anywhere in the world. The closest active program targets the adjacent UBE3A region, which is a different syndrome with different biology.

When you find yourself in that position, you do one of two things. You wait, and you hope that the broader field eventually pays attention. Or you decide that someone is going to do the work, and you ask whether that someone might as well be you.

I want to be careful not to romanticize this. The decision was not based on a heroic conviction that I could outperform the field. It was based on a more clinical observation: the field is not working on it, the patient population is small enough that it likely never will, and the molecular target is tractable. CYFIP1, one of the four genes in the duplicated interval, has a dosage relationship that makes a partial knockdown therapy plausible. The technology to do that — antisense oligonucleotides — is mature. The regulatory path exists. The bottleneck is not science.

The bottleneck is that nobody is doing it.

So I incorporated a company. Longinus Therapeutics, registered in Spain, focused on the BP1–BP2 microduplication as a first indication. The technical groundwork — structural validation of the target, virtual screening of approved drug libraries for repurposing candidates, an internal ASO design pipeline — was built during the months leading up to incorporation. The company collaborates with academic partners in the Netherlands, Switzerland, the UK, and Spain.

The transition from machine consciousness to therapeutics turned out to be less of a jump than I expected. Both fields involve building a careful model of something you cannot directly observe. Both fields have a rich theoretical literature that is largely disconnected from operational tooling. Both fields reward the kind of person who can sit with uncertainty for long stretches without getting bored. The technical skills transfer surprisingly well — antisense design is a search problem, repurposing screens are an information retrieval problem, structural validation is a geometry problem. None of these are new to anyone with a CS background.

What does not transfer is the timeline. A software project iterates in days; a therapeutic project iterates in months or years. The discipline required to make decisions on noisy preclinical data, knowing you will not get a clean answer for another 18 months, is different from anything I did before. I have not fully adjusted. I work on it.

I am keeping the older posts on this site for two reasons. First, the thinking that produced them is the same thinking that produced the company — the habit of taking abstract problems seriously and trying to build operational versions of them. Second, I do not think the consciousness work was wrong. I just think a different problem in front of me requires the same person to work on a different question for now.

If you are a researcher, a clinician, a family with a related diagnosis, or an investor — I am happy to hear from you. The work is ongoing. The deadline is real.