Boundaries: what ControlAI does NOT do¶
An honest conversation about boundaries matters just as much as the list of capabilities: it protects both the rector's office and instructors from false expectations. Here is what ControlAI does not do (at least, not yet).
Does not identify individual students¶
The system understands how much students talked in total, but not "which one, specifically." It does not link speech to a named student, and it does not recognize students by voice or face.
This is a deliberate boundary: the subject of analysis is the learning process and the instructor's work, not individual students. The system keeps no personal student profiles or "dossiers." What appears in the product next, and in what order, is covered in the Roadmap chapter.
Does not grade students¶
ControlAI does not assess student knowledge, assign scores, or affect a student's rating, credits, or final grades in the credit-module system in any way. What gets evaluated is the class session as a process, not the performance of the people sitting in it.
Does not proctor exams¶
This is not an exam-monitoring system: ControlAI does not track cheating and does not monitor student behavior during tests. It analyzes regular scheduled class sessions — lectures, seminars, practical sessions, and lab sessions.
Does not replace official instructor certification¶
The 1–10 instructor rating and the lesson observation form scores are the university's internal analytics: material for development, mentoring, and management decisions. This is not a substitute for official instructor certification (attestatsiya) and is not a state procedure. At the university's discretion, this data may serve as supporting evidence for internal KPIs — but that status is granted by the university itself, not by the system.
Does not monitor 24/7 and is not "student surveillance"¶
ControlAI only analyzes scheduled class sessions — specific classes in specific classrooms. It is not continuous room monitoring and not a tool for spying on students: the focus is on how the class session was conducted, not on what any individual student was doing. Cameras in classrooms are, as a rule, already installed and recording — ControlAI only analyzes class sessions from that existing recording.
Does not pass automatic judgment on an instructor¶
- Lesson observation form scores come with a confidence level and a supporting quote; disputed items are flagged as "needs human review."
- Decisions — personnel and methodological — are always made by a human: the quality-control office (ichki ta'lim sifati nazorati bo'limi), the head of department, the rector's office. Whether every score is reviewed before publication is something the university itself decides in its settings.
- The goal is to help the instructor grow, not to punish them; there is no public "wall of shame."
Not "ask anything about the video"¶
ControlAI recognizes specific, validated signals (in the next stage — attendance and late arrivals via video). It is not an all-seeing AI you can ask any question about any second of video: that approach is unreliable, expensive, and unacceptable from a student and instructor privacy standpoint. We take the path of individual validated signals, not "magical omniscience."
Does not assess emotions or "engagement from faces"¶
Reading emotions or "interest" from students' faces is ethically questionable and unreliable. We deliberately do not do this.
Does not work "live"¶
The report is usually ready within an hour of the class session; with a limited internet connection, by the next morning — either way, it is not real time. ControlAI analyzes a class session that has already happened, not an online broadcast with live annotations.
Depends on audio quality¶
All speech analytics relies on how well the class session can be heard and on the quality of recognizing mixed Uzbek-Russian speech — this is the product's main technical challenge. Where the audio is poor, accuracy drops; that's why audio quality — including in large lecture halls — is checked during the audio audit at onboarding and improved where needed (for example, with an external microphone connected to the camera's audio input).
A decision-support tool, not a replacement for people¶
ControlAI shows where to look and what to improve. But decisions — personnel, methodological, about a specific department or instructor — remain with people. It is a tool that strengthens the quality-control office, heads of department, and the rector's office — not one that replaces them.
Key takeaway from this chapter: ControlAI does not identify individual students, does not grade them, does not proctor exams, does not replace official instructor certification, does not monitor around the clock, and does not spy on students — the subject of analysis is always the learning process. It does not pass judgment on its own, does not answer "any question about the video," does not read emotions, and does not work live. It is an accurate decision-support assistant, and the decision is made by a human. What of this may change in the future is covered in the next chapter.
→ Next: 16. Roadmap — what will appear in the product over time.