Discern FAQs

How Discern Works

Q: How does Discern work?

A: Discern operates as an inference engine, utilizing an AI system to analyze online activities that have been anonymized to remove any personally identifiable information (PII). It applies a specifically designed "framework" consisting of instructions that guide the AI in generating insights and inferences from the students' online behavior.

Q: How do you get the “Students’ Digital Footprint”?

A: If you use Securly Filter, we can immediately begin inference over the data already collected - including web searches, social posts, and sites visited. If you are not using one Filter, we will install the “Discern Extension” on browsers to begin collecting the online activity for inference. You will be able to start running your first analyses after 1 week of collection.  Discern can only use data provided by Filter.

Discern Benefits

Q: From a day-to-day perspective, how does Discern help districts? 

A: Discern uses AI to generate powerful data sets for K-12 directly from students' online activity. The product will give districts more insight into student behavior, student alignment to district goals, academic readiness, student wellness, and student safety than was ever possible before the advancements of AI, and it will do this with dramatically less effort.

Q: How much time can it actually save them?

A: We have validated with multiple beta users and initial prospects that the legacy survey based approaches to capturing and collecting the types of data that Discern generates will save district personal months (and months) worth of time. The typical SEL process of collecting data, analyzing and then reporting takes 4-6 months to complete.

Q: What surveys would it replace?

A: Pretty much any survey can be replaced. Surveys used for universal screening, assessing school climate, understanding students’ engagement, measuring social emotional skills and competencies, assessing risk, mental health issues, safety issues,  and more. Discern can also gather more information that is not currently available to gather via surveys or assessments.

Discern's Frameworks Explained

Q: How are the frameworks developed and validated?

A: All frameworks are developed and validated by our in-house clinical team. Many frameworks are directly aligned to evidence-based assessment frameworks, like CASEL’s 5 Core Competencies, or the Department of Homeland Security’s School Violence Threat Assessment. Others are custom created by our in-house team, combining relevant variables across an array of established practices and adapted for the nuances of AI inferencing. This meticulous process means the frameworks are relevant, effective, and evidence-based.

Q: Can you provide details on the frameworks used by Discern?

A: Discern utilizes a variety of frameworks designed to evaluate different facets of educational life. These frameworks are categorized into four main areas, each with specific focus areas. Click here to review a deep dive into the Discern Frameworks.

Discern Budgeting, Pricing, and Deployment

Q: What is the price of Discern?

A: Discern costs $5 per student. At this price, districts have a sizeable data cap (that they are unlikely to max out) and can run a large number of frameworks and analyses to meet their data needs. 

Given that Discern is a new AI solution and we are still learning about customer usage patterns, in the unlikely case that a district reaches its data cap, it can double its data at an additional cost of $1 per student. 

Q: What would Discern replace at the district level?

A: The budget for Discern will come from legacy survey tools. Every school has multiple data survey tools that they use and sales cycles will assess and discover how Discern will replace these tools.

Discern also saves labor time. These surveys are generated just to collect the data, then school personnel have to spend time sifting through data, building reports, figuring out what the data means and then planning actions based on the insights gained. All of that lost time is won back if they adopt Discern.

Q: What does deployment look like?

A: For existing Filter customers there is no deployment - Discern will be turned on for the customer and they can start running data frameworks.

For new customers they would have to deploy a Discern extension to collect student online activity and those extensions running on students devices will have to run for approximately 30 days to gather a decent sized data set for Discern to analyze and generate data.  Discern can work alongside other content filters in a "Discern only" mode with all Filter capabilities disabled.

Q: Is there training involved?

A: New product and lots to learn - we do not know the answer to this question as of yet. We did not provide any training at all to out beta customers they just jumped in and started using the software and generating data and insights.

Ensuring Data Validity

Q: How do you ensure the validity of the data produced by Discern?

A: Ensuring data validity involves several key steps:

• Sample Data Vetting: Our clinical team performs an initial analysis of sample data on a framework-by-framework basis to ensure initial accuracy, before releasing the framework to our users.
• Benchmarking of Aggregate Data Sets: Discern offers a feature that enables the comparison of all aggregated data against established benchmarks. This process normalizes the data, accounting for any discrepancies in AI inferences, to ensure interpretations and reports are reliable and meaningful. By utilizing this benchmarking tool, you can confidently understand your data through comparisons that are consistent and directly comparable.
• Transparency for Individual Students: Discern ensures transparency when analyzing data related to individual students by providing comprehensive summaries of the AI's analysis and the reasoning behind specific inferences. Additionally, the platform offers the ability to examine all the underlying online activities that contributed to these insights. This feature enables educators and administrators to manually verify the accuracy of individual student level data before making any decisions, addressing concerns about potential actions based on incorrect information.

Differentiating Educational and Non-Educational Activities

Q: How does Discern differentiate between educational and non-educational activities?

A: Discern's AI analyzes patterns of activity over weeks or months, allowing it to identify education-related searches effectively. This method goes beyond looking at individual searches, providing a broader context for each inference. Additionally, explicit instructions are provided to the AI to account for this particular issue, meaning Discern is one of the only tools available that can reliably account for educational activities when analyzing for risk. 

Privacy and AI Learning

Q: How is student privacy protected when using Discern? Parents are concerned about AI accessing their child's information? How do I address/respond to this?

A: Short Answer:

• We don’t send any PII to any AI systems, including Azure and Open AI
• There is no training of any AI system on any district/school/student data 

Detailed Response: 

Discern maintains strict privacy:

• We do not send any personally identifiable information (PII) to any AI system. No AI system leveraged in Discern has any capacity to know what online activity data is associated with what student. Online activities are sent to the AI completely abstracted from any PII.
• Additionally, we analyze data consistent with all of our already robust privacy policies.
• Discern’s interface maintains FERPA compliant role-based access, with users accounts permissioned to only the student data that they need for legitimate educational purposes. Additionally, consistent with regulations like SOPPA & COPPA data is used for beneficial purposes only.

Q: Does the AI learn from the data it analyzes?
A: No, the AI does not learn from the data. It processes each request based on predefined instructions without retaining information from previous analyses. Instead, Discern, as a whole “learns” when our clinical team reviews outcomes and refines the system instructions in a given framework for improved quality.

Data Collection and Framework Customization

Q: There is some data we might not want to produce and we don’t want to overwhelm staff with information they have to respond to quickly. Is it possible to customize the data we collect with Discern? 

A: Yes, Discern allows significant customization, including:

• Selecting specific frameworks to run.
• Choosing frameworks without risk variables for sensitive areas.
• Customizing frameworks to exclude unwanted variables.

Generally speaking, Discern is different from traditional AI safety products in that it only generates the EXACT INFORMATION you ask for at the EXACT TIME you request it.

Addressing Bias and Cultural Sensitivity

Q: How does Discern address potential biases in data analysis?

A: We acknowledge that AI Systems have some inherent bias. We also acknowledge that surveys and assessments do as well. We leverage our diverse, in-house clinical team to build and refine this system to reduce bias. 

Some Important Knowledge: Interpretive vs Direct observations

• A framework centered on "student interests" primarily focuses on making direct inferences based on observable online activities. This approach limits the scope of interpretation, thereby minimizing the potential for bias. It adheres closely to what can be directly observed or evidenced in the students' online behavior, offering a more straightforward and less subjective analysis.
• Conversely, a framework such as CASEL (Collaborative for Academic, Social, and Emotional Learning) skills operates on a more interpretive level. It integrates lower-level direct observations—such as interests and behaviors—to formulate higher-level interpretations regarding a student's social and emotional competencies. This method involves a deeper layer of analysis, bridging concrete data with abstract qualities like empathy, self-awareness, and decision-making skills. As a result, these frameworks are inherently more subjective, introducing a greater possibility for bias due to the interpretive leap from observed actions to inferred skills or traits.

Our strategy for minimizing bias involves a multi-faceted approach, informed by our clinical team's continuous efforts in this area:

• Education on Framework Differences: We educate users about the distinct characteristics of direct and interpretive frameworks, including the inherent potential for bias in AI interpretations. This education helps users understand the methodology behind the AI's analysis and the importance of context in evaluating results.
• Transparency in Summaries: For interpretive frameworks, Discern has the option to have the AI include considerations of potential bias in its summary outputs. This transparency provides an additional layer of scrutiny, ensuring users are informed of any biases that may influence inferences at the individual level.
• Benchmarking for Normalizing Data Sets: To further address bias and enhance accuracy, we plan to introduce "like-for-like benchmarking." This feature will enable districts to compare their aggregate data not only against the broader Securly ecosystem but also against districts with similar profiles. Such comparisons ensure more accurate and bias-normalized assessments of aggregate data.
• Clinical Design to Limit Bias: Our diverse clinical team plays a crucial role in continuously refining the instructions provided to the AI. Their expertise helps ensure the AI's analysis is as unbiased and accurate as possible, contributing to the reliability of the insights generated.