Generative AI in Public Education

This post first appeared on IBM Business of Government. Read the original article.

Wednesday, October 4, 2023

Imagine the controversy some 40 years ago among engineering school faculty who expressed concern that the standard hand-held slide-rule might be replaced by sophisticated hand-held calculators? Should this new technology be allowed in class during exams?

This is the sixth in a series of articles stemming from the National Academy of Public Administration’s Standing Panel on Technology Leadership as part of its Call to Action on Responsibly Using AI to Benefit Public Service at all Levels of Government. Please see our first blog, “A Call to Action: The Future of Artificial Intelligence and Public Service” second blog, “Artificial Intelligence and Public Service: Key New Challenges,” third blog “Making Government AI-Ready Begins with an AI-Ready Workforce,” fourth blog Artificial Intelligence and Public Service: Key New Challenges, and fifth blog “Digital innovation: from “tech problems” to “redesigning governance”.”

The main concern wasn’t against a new and promising technology, but instead their fear centered around how students would lose the connection of thinking through a process as opposed to having it instantly solved for them. Today there are similar debates and discussions with the advent of generative AI and in particular the popular ChatGPT. And given the many complexities and allure of generative AI (GAI), the consequences are far more serious. Today professors are rightfully concerned that students will use this nascent technology to short-cut the learning process and use it to prepare outlines and in some cases actual assignments submitted as their own work.

A more mindful review of the pros and cons of GAI in public education reveals some compelling arguments as well as some dangers that must be recognized and addressed.

Advantages of Generative AI in Public Education

  • Human-like Text Responses: GAI produces fluent, textual and verbal interaction leveraging natural language processing to produce a reasonable continuation of text completion, conversation generation, and language translation from massive data sets of human written text available across the Internet.
  • Personalization and Individual Attention: GAI can learn and develop over time based on an individual’s input, interests, responses or behaviors to provide personalized content, feedback, and performance support to students inside and outside classroom hours.
  • Immediate Feedback and Increased Interaction: Students can receive optimized, real timefeedback on their progress and participation as well as monitor their own performance. Conversational agents can provide effective support for student learning, interpreting student questions to provide relevant interactive responses, which can enhance their learning process.
  • Adaptive Learning: Advanced integrations can be designed where GAI can adjust instructional methods based on student progress and performance to assess, predict and optimize student learning with automated learning pathways, sequencing and pacing of content.

Accessibility: Improved accessibility of digital materials for students with disabilities implementing required accommodations through optimally presenting information to increase understanding and support reading comprehension, as well as analyzing patterns from relevant data sources to customize and automate formatting of accessible materials and support.

  • Digital Content Generation: A production tool for students and instructors in the rapid generation of presentations, video, images/graphics, text, audio, video through dissecting examples; learning their patterns, imagery and distribution.
  • Automated and Consistent Assessment: Every student receives the same assessment process with improved accuracy through grading procedures leveraging GAI, such as identifying features of well-written essays providing more consistent feedback when trained on a data set of human-graded essays, eliminating the potential for educator-based inconsistencies.
  • Scalable, Predictive and Recognition Tool to Complement Human Interaction and Decision Making: Mining past inputs, GAI can be used to generate suggestions, systematically compare responses, recommend follow up corrections and augment teacher tasks to provide more time for rich student interaction and enhanced decision making, thus reducing the strain on human resources.
  • Cost-Effective: Over time, the integration of GAI might reduce the costs associated with instructional content synthesis and production, textbooks, supplementary materials, or additional human tutors with pinpointed, personalized, lifelong learning support.
  • Informal, Self-Directed Learning: It can provide a great supplement to human teachers, aiding in areas where there might be a shortage of human resources, suggesting relevant content and providing personalized tools to mentoring support self-regulated learners.

Like the slide-rule controversy, it is easy to overlook some of the disadvantages of a profound and new technology. Its limitations are serious and requires recognizing and developing strategies that address any limitation. Here are but a few disadvantages of GAI.

Disadvantages of ChatGPT in Public Education:

  • Limited Understanding: Without a true, experiential understanding of learning concepts that may warrant contextually appropriate responses and reliant on statistical patterns found in the training data set, GAI is limited in directly addressing student misconceptions in complex, dynamic problem solving.
  • Lack of Emotional Intelligence: GAI doesn’t understand emotions, so it can’t provide the emotional support or understanding that a human teacher can.
  • Bias in Training Data: Generative models are only as good as the data they were trained on and the massive data sets used for training these models contain inherent human biases.
  • Over-reliance: There’s a risk that students might become overly reliant on GAI for answers, discouraging independent research or critical thinking. Students may fail to realize the limitations of what data has been indigested into its systems and thus omit many newer findings that have not yet entered the GAI domain.
  • Potential Misinformation: While GAI is trained on vast amounts of data, there’s no guarantee it will always provide the right answer. Teachers need to ensure students are still engaging in critical thinking. Just because it looks right and is perfectly formatted doesn’t mean it is correct. Those who have used this technology report that GAI systems have made things up which are now called hallucinations.
  • Lack of Comprehension, Metacognition and Self Awareness: GAI lacks comprehension, human-level judgement, awareness and metacognition of its own output with limitations in itsintelligent behaviorinterpretation of social nuance, self-monitoring and complex problem solving.
  • Ethical, Security and Privacy Concerns: Using AI in classrooms could lead to concerns about data privacy and security, especially if conversations, interactions or queries are stored or analyzed.
  • Plagiarism: Students may be tempted to overly rely on GAI and submit work without attribution. Currently, GAI systems do not offer explainability as to where their stored information or responses are coming from. Recently the US Copyright Office has launched an inquiry into copyright infringement in GAI.
  • Technical Issues: Reliance on technology always comes with the risk of technical glitches, outages, or other issues.
  • Update Limitations: GAI at any given version, has a knowledge cutoff. This means that it may not be aware of the very latest developments in a field, unlike a human educator who can continually update their lesson plans.
  • Rather than outright banning GAI, public education should develop AI literacy, policies and guidelines, spelling out how this technology can or can’t be used. After all, there are many positive and promising applications that can enhance learning outcomes. For example, the nonprofit Khan Academy, well known for its innovative approach towards teaching K-12 student, is experimenting with what they call, Khanmigo for students that is designed to provide individual lesson tutoring and Khanmigo Assistant for teachers. It is too early to evaluate its effectiveness, but the very concept appears well thought out and promising. Will tomorrow’s students look back with affection regarding their most influential chatbot as opposed to their favorite 3rd grade teacher or college professor?

While we cannot ignore the potential dangers of overreliance in GAI, there are so many promising applications in the realm of public education, their usage should be balanced with human-led instruction, human-centered systems design and augmenting human intelligence and problem solving to improve learning and education. At least for today, it’s most effective when human judgement is leveraged in conjunction with generative artificial intelligence to empower teachers, students and future generations to be responsible, creative and productive in the use of these emerging tools.

Image by rawpixel.com on Freepik.

Leave a Reply

Your email address will not be published. Required fields are marked *