An independent lab for questions across unlike systems
Far Akin Research Labs studies the interpretability of minds, natural and artificial: how they learn, organise internal representations, remember, perceive and act, and how far they are akin. Its long-horizon programme is a comparative science of minds: theory and instruments that put the same questions to a model and to a mind under one standard of evidence, so that a claim of likeness can be tested rather than asserted.
The lab
Far Akin is independent. It answers to no institution and no product, and it chooses its questions for their substance. It aims to produce work other researchers can use and check: theories, methods, instruments, datasets and papers. Products may follow from research. They are not its purpose.
Subjects are what the lab studies, lenses explain them and methods investigate them. Programmes pursue long-horizon questions. Inquiries test one bounded question at a time, each with its design, what would count as an answer and what would end the attempt written in advance.
The name and the mark
Far Akin names the relationship the lab investigates: systems that are far apart may turn out to be akin, in mechanism, in method or in the way they fail. The name states a question. It does not claim an answer.
The mark draws the same idea. Two unlike bodies, one soft and one faceted, overlap. The region they share is left open, and that open region is where the lab works.
How the lab works
Six commitments about method. They hold for every programme and every kind of system the lab studies.
- 01
Interiors from outside
Models, like minds, are studied through behaviour, measurement and controlled disturbance. Information that can be decoded from a system is not assumed to be information the system uses.
- 02
Resemblance is tested, never assumed
The lab carries methods from the study of human minds to machine models. Published human and neural evidence enters as data and as constraints on what a model of a mind must explain. The lab runs no studies with people at this stage. Behaviour that looks human, or a feature that can be decoded, does not show a shared mechanism or experience. A resemblance that holds has to be shown, and one that breaks counts as a result.
- 03
One standard of evidence across unlike systems
A reading of a model and a reading of a mind face the same tests: repeatability, validity on planted cases and rival explanations ruled out. Each inquiry states in advance what would count as an answer and what would end the attempt.
- 04
First principles before scale
A design starts by defining the phenomenon, its state, the rules by which it changes and the rival explanations, then derives what each rival predicts. A small model that can tell the rivals apart comes before a large run. A larger run follows only if the question needs it.
- 05
Old questions, new instruments
Some precise questions stalled for want of a way to test them: what a system holds steady, whether an instrument's readings repeat, what a model of a world must preserve. Several can be asked now.
- 06
Checks before claims
Each instrument is to be tested on planted cases whose answer is known before it is used on a real question, and each key analysis checked by a separate implementation. A test that fails is reported with what it rules out. A reading that fails a check is withdrawn, and the withdrawal is dated.
Stage
The lab's work is at the design stage. Nine inquiries are published as written designs, each with its method and the result that would count against it. Results will be reported on each inquiry's page, with their limits, as the work runs.
The essays and explainers draw on published research by others, which they cite. Work posted as a preprint or submitted to a journal will be linked from its page.
Founder
John Leskas
Founder
John Leskas founded Far Akin to study how minds, natural and artificial, come to hold what they hold. His research interests run from the foundations of machine learning, mathematics and the internal representations of models to cognitive science, consciousness and nonconscious processing, cybernetics, simulated people and world models.
He holds the lab to one standard for every claim: say what would count as an answer before looking for one.
Independence
Far Akin is independent and works in the open. It publishes its designs and methods before its results, so that anyone can check the work against what it said it would do. Nothing on this site draws on work or data from the founder's other professional roles.
The plates are generated images made for the site. They illustrate ideas. None is a photograph of an experiment or a result.