Nuvl
Named by hash
Uncensored
Verifiable
Library
|
About Nuvl
The Nuvl project aims to build a Named by hash, Uncensored, Verifiable Library:
- Named by hash: Data in Nuvl is named by its hash, not
by its server location, so it can live anywhere on the
net.
- Uncensored: A link to a claim is good as long as the
hash can find the data anywhere. No one organization
controls it.
- Verifiable: All claims link to their sources so your
viewing software can verify it according to your own
accepted assumptions.
- Library: The result is a shareable library of
knowledge you can trust.
Nuvl also sounds like núvol,
the Catalan word for cloud.
|
What Nuvl is
Nuvl is a protocol for using and updating causal
models which explain evidence. Each explanation is
connected to its supporting evidence, so you can
evaluate it against competing explanations.
Eventually, your Nuvl instance can be connected to
all knowledge you can put into it.
|
What Nuvl is
not
|
|
Nuvl isn't an autonomous learning algorithm |
... but other learning
algorithms can feed models into Nuvl. |
Nuvl
isn't an Artificial Intelligence |
... but an AI system may use
Nuvl as a reliable tool. |
Nuvl
doesn't
think for you |
... but if you express your
knowledge to Nuvl it can tell you where there may be
contradictions and where improvement may come from
other connected knowledge. |
Nuvl won't hold
someone's falsehoods
|
... but Nuvl provides a place
where people who want help with lying won't use it
for exactly that reason.
|
Technical details
Named by hash
Nuvl needs a reliable way to get the evidence it depends on,
so it refers to a piece of evidence (video clip) by its
hash. This has two advantages: each piece of evidence
automatically has a globally unique name so that it can be
stored anywhere, and it is secure because when your software
downloads a piece of data it computes a new hash and
compares with the name. Because of the strong
cryptography of the hashing algorithm, you know you got the
data you wanted. No one can spoof you and make you
retrieve the wrong data.
Causal models
The world has systems with physical attributes. Knowledge is
causal models which predict the change of those attributes.
A model has the form {set of attribute values} implies event
causes new attribute value. If models make conflicting
predictions, this is defined as a "problem". Knowledge is
created with a new model which resolves the conflict.
Algorithms
The Nuvl software has a set of algorithms for processing
models.
- applying models
- applying the model selection algorithm to explain
evidence
- recording the assumptions and evidence used to compute
a new causal path
- making predictions from selected models
- detecting problems (conflicting predictions)
- categorizing unresolved problems
- connecting a new causal model (created by another) to
the problem it resolves
- recategorizing problems when the model selection
algorithm is changed (by another)
- replacing many separate assumptions with a new model
which predicts them
- reviewing/accepting the problem and resolution of new
models (created by others)
|