AllHere Explained #4
According to Digital Promise, the use of Artificial Intelligence (AI) has the potential to transform learning experiences, enabling people to engage in teaching and learning experiences more deliberately. However, the race to develop AI tools is upon us, even without many understanding what problems we are trying to solve or ensuring that researchers, education technology developers and education leaders are working together to ensure AI in education has a human-centered approach.
AllHere is grounded in research and we partner with districts to ensure the tools we develop are meeting educational needs. And, as part of this work, we follow the recommendation of the U.S. Department of Education’s Office of Educational Technology that emphasizes the “Humans in the Loop” (HITL) principle as essential to ensuring safe and effective AI implementation in education.
This post, the fourth in the “AllHere Explained” series, helps explain why education institutions need to ensure their vendors are using a “human-in-the loop” approach when it comes to developing tools for students. AllHere’s latest tool for districts launched recently with Ed™, a pioneering, AI-fueled learning acceleration platform created in partnership with the Los Angeles Unified School District.
Unlike many AI tools where humans are used for traditional data annotation and to model fine-tuning, throughout AllHere’s development of Ed™, humans serve as moderators and trainers for our AI in various ways:
Humans actively supervise, moderate and escalate conversations that go beyond the capabilities of chatbots due to their nuanced nature.
Trained, background-checked staff work 24/7 in tandem with automated escalations from the chatbot for emergent situations. During non-emergent situations, they initiate escalation processes, provide insights into complex conversations, and validate bot recommendations.
AllHere staff also train the Ed™ chatbot continuously, adding new information to it on the basis of usage and fueling iterative enhancements to the chatbot. By capturing insights on incorrect responses and knowledge gaps, we refine the bot’s capabilities over time, ensuring responsiveness to evolving user needs.
Below is an illustration of our Human-in-the-Loop process in relation to the chatbot.
Illustration: Incorrect or Misleading Response
User Query: User: “What are the symptoms of COVID-19?”
Bot Response: Bot: “COVID-19 is a type of flu caused by the influenza virus. Symptoms include fever, cough, and sore throat.”
Moderator Intervention: Recognizing that the bot’s response is incorrect and potentially misleading, the moderator intervenes to provide the correct information. They engage with the conversation and offer the following response: Moderator: “COVID-19 is actually caused by the novel coronavirus, not the influenza virus. Common symptoms include fever, cough, shortness of breath, fatigue, and loss of taste or smell. It’s important to stay informed about accurate information regarding COVID-19 to protect yourself and others.”
Flagging for Review: After providing the correct response, the moderator flags the conversation for review. They indicate that the bot provided inaccurate information, prompting the need for further scrutiny to prevent similar errors in the future.
Feedback Loop: The flagged conversation is logged for analysis, allowing the team to identify the root cause of the bot’s misunderstanding and implement corrective measures. The moderator initiates a feedback loop to prevent similar occurrences in the future.
In this scenario, the moderator’s intervention serves not only to rectify the immediate error, but also to contribute to the ongoing improvement of the chatbot’s knowledge base.
By incorporating human oversight and expertise, the AllHere system maintains the integrity and reliability of its responses, enhancing the overall user experience.
About this Series: This blog post is the fourth in a series called “AllHere Explained.” The series will explore a range of topics, including AllHere’s approach to security, privacy, AI/ML safety, education innovation, digital transformation, and more. We hope you find these blogs useful and informative; and we welcome feedback.