I'm Murad. I run a small independent research group in London called Dissensus (an IASEAI affiliate since August, which sounds grander than a founder, a couple of collaborators and a lot of unpaid evenings), and I'm preparing to start a PhD in complexity science next year on emergent collective misalignment in multi-agent systems.

My background is not the usual one for here. I did accounting and finance at SOAS (a dissertation on money laundering, of all things) and a finance MSc at King's, and spent a while doing empirical work on markets before the questions that actually kept me up were the ones about what AI systems are. About a year ago I started reading AI ethics properly and found it thin on exactly that, which sent me down a rabbit hole: does a system deserve moral consideration rather than being treated as an arbitrary tool (From Consent to Consideration, under review), is an "AI relationship" parasocial by construction (Relational Functionalism, one reviewer accepted, the other largely disagreed about the ontology of the word friendship, so an R&R), and whether we've been calcifying processes into nouns (a monograph on qualia I wrote mostly to work out what I actually believed, since neither eliminativism nor functionalism fit).

That pulled me toward a more abstract question, what legitimate governance even is, and from there into complexity science, which I was slightly embarrassed to discover existed as a whole field outside anything I'd been taught. I tried to formalise how things persist across disciplines and apply it to metaethics (ROM, whose formal appendix I've since had to correct in public, which was the right thing to do and not fun), proposed a consent axiom on top of it (which I later disproved), and ran the MARL experiments that did the disproving (Four Ways, a preprint; desk-rejected once, warranted, it needed more work). More recently a small empirical finding on whether models show preferences when first allowed and then restricted (The Shape of Stakes), and I'm currently re-running the 2025 LLM social-convention experiments on 2026 models to see what survives a point release.

Nothing here is settled. Most of what I've learnt is which of my own ideas don't survive checking, and I'd rather post that than pretend otherwise. Longer-form writing at farzulla.org, code at github.com/studiofarzulla, the rest at dissensus.ai/research.

Murad Farzulla

Murad Farzulla