Research programme
ResearchMachina eX
Machina eX is our research programme in machine intelligence. It asks what a mind built from probability, chance and evidence would look like, rather than one built to imitate us.
Machina ex SigmaTalk to Machina eX
Coming onlineAsk it anything, search our work, or explore the code. It will run on Da Vinci, our own AI on our own server, so what you type stays with us. The chat is being connected now and is not answering yet.
- Explain a neural network like I'm twelve
- Could a machine ever understand Chinese?
- Why can't anyone see inside a black box model?
- Show me the code behind Da Vinci
The name
Deus ex machina
Greek theatre's god from the machine: an actor lowered onto the stage by a crane to resolve the plot from above. Machina eX turns it round. Nothing is lowered from above; the answer has to come out of the machine itself.
Ex Machina
Alex Garland's 2014 film, in which a programmer is asked to judge whether an android called Ava is conscious, and is outplayed. It is the question this programme keeps open: what passing the test would and would not prove.
Intellectus Machinae ex Summa
From the sum, intelligence. Machina eX starts from the sum of many uncertain signals, weighed and tested, and treats intelligence as what comes out of that sum.
How machines think, in plain words
Neural networks
A neural network is layers of simple units, each one adding up the numbers it receives, weighting some more than others, and passing the result on. Training means nudging those weights, millions of times, until the outputs stop being wrong. Nobody writes the rules; they settle into the weights.
The Chinese room
John Searle's 1980 thought experiment. A man who knows no Chinese sits in a room with a rulebook. Notes in Chinese come in, he follows the rules, perfect Chinese replies go out. To the people outside he understands Chinese; inside, he understands nothing. Searle's point: running the right program is not the same as understanding. Our Da Vinci guardrail is built on it, and tells the model it handles symbols and does not understand them.
The black box
A large model can give the right answer without anyone, including its makers, being able to say why. The reasons are spread across billions of weights, not written down anywhere. Machina eX's answer is to make every reasoning step show its working and its uncertainty, so you can check it instead of trusting it.
Sigma ergo sum
Our ARI paper moves from Descartes' cogito ergo sum to sigma ergo sum. Its test for a new kind of intelligence: not how human we can make it, but how inhuman.
The work
Research, not a product yet.
Artificial Random Intelligence
ResearchRandomness used on purpose: random starting weights and thresholds inside an ordered framework, so the system explores instead of repeating.
Super Stochastic Intelligence
ResearchReasoning in probabilities: every belief carries its uncertainty, every step can be explained, and the system learns faster from being refuted than from being confirmed.
MISIX
In developmentThe working name for Machina eX's engineering. Details coming.