Brings Intelligence to Assessment’s uses an AI-based algorithm that helps professors organize and assess discussions.

Structuring discussions helps students see highlights and get a better understanding of every discussion.

Learn more about our algorithm!

How Creates Deep Discussions

How can one assess a healthy discussion? 

How can a group of learners accomplish a reliable answer to a difficult question?

At, we aim for healthy discussions, assessing all aspects of a multilogue. The old-school discussion boards are unreliable, do not help with analyzing discussions, and focus only on participation score which does not make for a healthy, deep discussion.

As we learned from centuries of philosophy, a good discussion starts with a question… What do we want to know? 

The next layer should be alternative possible answers to the question, or in other words, alternative hypotheses. This is mostly a matter for the imagination of the participants, suggesting mutually exclusive alternatives.

The last layer is a healthy debate about each of the alternatives. This means raising supporting arguments as well as challenges to each of the answers and backing each argument and counterargument with evidence (e.g. citations, data points, etc.). captures this discussion structure and analyzes the metadata of its development, dynamics and interrelations between the participants. captures dynamics and interrelations between the participants

Bringing Health to the Classroom Discussion

Once you are in the thick of the discussion, the AI looks into a few specific aspects.  First, a healthy discussion should go deeper, it should include as many arguments and counterarguments, and as much evidence as possible. We want to strengthen the participants who present real arguments and evidence, rather than just voting.

A healthy discussion is focused on the debate rather than on each participant supplying their own answer, sometimes repeating what others have already suggested. also measures how much each participant reads before they write. We do not value participation or activity as such. measures intelligent contributions, which take into account what has already happened in the discussion. lets students make supporting arguments and challenges to each of the answers

The Mathematics of Knowledge

And then goes even deeper. 

Once the debate gets going we apply a Bayesian algorithm to score each of the alternative answers. Bayes equation is a very famous theorem in statistics. Put simply, it is like a weighted crowdsourcing algorithm; for each support to a certain answer it adds to the reliability score of that answer, and for each counterargument it reduces the score. 

However, it’s not enough to simply suggest an argument or counterargument. The algorithm also looks into how much support each of the claims gets from other students, who it was that gave that support, and how that supporter relates to other participants. The algorithm thus scores not only the top answers to the question posed, but each and every claim in the discussion.


Not Just Assessing, But Also Improving Student Analytical Skills

Given that the Bayesian algorithm scores each section of the discussion, can collect for each student all of the claims that they contributed in any discussion. Using this we calculate the student’s peer review score, based on the engagement they created and the responses they got from others. 

Furthermore, looks into many other aspects of participants’ behavior, such as:

  • Collaboration level (i.e. do they collaborate widely with all their peers or only with two best friends)
  • Thoroughness (i.e. do they satisfy themselves with voting only, or do they contribute arguments and evidence)
  • Clarity (i.e. how often do their peers ask for clarification to their claims)
  • Balance (i.e. what is the ratio of arguments vs. counter arguments they contribute to the discussion). 
This is all meant to help each participant improve their analytical skills. This is made possible by the analytics being translated into prescriptive advice on how to improve each participant’s behavior in the course of the discussion.

Bridging the Thought Divide

Given that follows arguments and counterarguments, our algorithms also can follow similarities in the behavior between different participants. 

We can cluster the user based on their similarities; if two participants always agree together and also always disagree together, we will say they “think alike”. A group of these may be considered to be in a bubble or a hivemind.

If they always have differing responses to claims we will say they have a greater thought distance

We can use this metric to expose students to diverse views, and to peers who they should engage with, even if they do not know them. We can focus their attention on heated debates and ask specific participants to engage in a specific part of a discussion in order to help decide it.


Discussion Entropy and Avoiding Polarization

This part can get a bit technical, but is pivotal to creating an environment for healthy discussions for a classroom. uses the unique notion of entropy to designate the polarization level of a discussion. This is a score for a whole discussion, looking into its dynamic along the time. 

Sometimes, a discussion starts with extreme polarization among participants, with many arguments and counterarguments that go deeper and deeper. When the participants are divided on which answer is most correct, the discussion has a higher entropy score.

Sometimes, after a while the participants converge into agreement as to the most reliable answer. That’s an example of lower entropy. 

If at any point there is a total consensus that one of the answers is the right one, we will say the entropy of the discussion is zero. 

If you want to learn more on “Discussion Entropy” you can watch the research conducted on data by Wharton and Stanford.

This website uses cookies

By visiting and interacting with Epistema Ltd.’s website, I hereby agree to the use of cookies as is stipulated in Epistema’s Cookie Policy and reaffirmed in its Privacy Policy.