How It Works
Presentation from HOPE 2020
The key component of *DAS is the certificate. Learners create certificates with a description and attach a case proving their mastery of this specific skill. When learners share their certificates, reviewers will rate the credibility of the case and how it matches up with the description.
When learners create certificates, they will also structure the certificates according to their prerequisite relationship. If certificate A is a prerequisite to certificate B, it means that if a person does not show sufficient evidence for the skill in certificate A, they can’t possibly master the skill specified in certificate B. The Truss Inference Engine uses these connections to help reviewers make sense of each certificate’s credibility as more and more certificates are rated.
In addition to the connections made by one person among their own certificates, connections can be made between people with similar certificates. This kind of connection is called an “award.” Learner X notices that learner Y in their community has also become great at the skill outlined in certificate A. By asking learner Y to create a certificate for the skill, learner X is motivated to provide an award between their two certificates outlining the same skill. Worth noting is that the description in the two certificates doesn’t have to be identical.
If a reviewer rates the certificates from both learners as credible, this award will benefit both learner X and Y as the Truss Inference Engine will take the award connection into account. However, if the reviewer does not rate the certificate of learner Y as credible, that will also reduce the credibility of learner X, who gave the award. This interpersonal dynamic encourages awards to be given only when one learner truly believes that another learner possesses the same ability.
This is the basic concept that underlies the *DAS protocol. The protocol, and the Truss Inference Engine, are open source. This allows anyone to contribute to the code and build their own software on top of the protocol. For example, there could be several different front-end applications to create and link certificates between learners. Specific software packages can be built to enhance the workflow of reviewers. As long as these different projects implement the *DAS protocol correctly, they can communicate with each other and share certificates.
For the technically inclined
The first *DAS inference engine, Truss, performs bayesian inference over bernoulli random variables that represent certificate trust (no parameter inference). Reviewers’ trust ratings are treated as expert priors. Prerequisites, prerequisite groups, and awards are represented as data, constituting observations of the network structure. The computations are backed by infer.net.
Independence of the graph
The accreditation graph exists independently of other entities like learners, mentors, educators, or reviewers. For example, if one of a learner’s certificates is reviewed poorly by a reviewer, it does not negatively affect other certificates made by the same learner by association; that would be an example of non-independence of the accreditation graph, because some information from the learner structure was used to influence its interpretation. This independence ensures that any inferences drawn about the trustworthiness of a certificate regard that certificate only, and do not reflect somehow on the character or overall trustworthiness of the learner who made it. A learner should be free to try a wide range of topics and not be hurt by any undeveloped competencies. Independence also promotes data security by at least offering the possibility of anonymizing certificates.
*DAS certificates are encrypted using the Fernet recipe whenever they are stored on disk. Security and privacy will become important if certificates become valuable digital assets, so this feature is incorporated into the design even in early prototypes. We intend to keep the cryptographic stringency high as we develop certificate access and distribution tools.
White paper to be released soon. Email info-at-stardas-dot-org to be added to our mailing list if you want to receive updates!