Key Questions Educators Have Regarding Artificial Intelligence

Educators are no strangers to the frenetic buzz accompanying the start of something new. As staff rooms and classrooms brim with the excitement of novel challenges and opportunities, we are veterans at navigating shifting landscapes. Yet, emerging technologies such as Artificial Intelligence (AI) may present the unfamiliar terrain of doubts and questions surrounding some fundamental aspects of the teaching and learning processes.

After listening carefully to both online and in-person discussions, we offer thoughts regarding the questions that seem to be most prescient regarding the potential of artificial intelligence becoming a part of our everyday encounters. The ensuing responses and further readings aim to equip educators with a firm grounding in the scientific foundations, shed light on central concepts, and provide a starting point to assess the benefits and drawbacks of AI.

How Can We Understand the Evolution of Artificial Intelligence?

The two key elements of AI that best illuminate how the technology has progressed are capability, or the range of domains the software can handle, and approach, which refers to methodologies used to solve problems. By diving into how AI is understood in terms of its core functionality and its recent innovations, we can better understand how to harness its potential.

The simplest kind of AI is referred to as “Narrow AI.” This term covers technologies with capabilities that perform very specific tasks, such as identifying people’s eyes in a school photo and cropping the image so it is perfectly centered. Teachers and students can utilize such software to create innovative learning experiences, for instance in a project that turns a webcam into an input device. For example in the “Teachable Snake” demonstration in Experiments with Google, the computer was pre-trained with explicitly-labeled images of an up arrow, down arrow, etc. These reference images can be distorted, or made fuzzier, to further extend the training process, producing a model consisting of how arrows appear when captured live on a webcam. When the student flashes an arrow in front of the webcam, the image is mapped to the closest match, and the snake changes direction accordingly.

The progression to Generative Artificial Intelligence (GenAI) took place when Narrow AI technologies were utilized to enable the production of original output. In a principle similar to the Teachable Snake game, generative AI utilizes pattern recognition to develop software that can compose, and not just recognize, patterns in data. For text, a Large Language Model (LLM) is able to both parse the input and statistically generate a composition that follows accepted writing conventions for style and format. For images, a Generative Adversarial Network (GAN) includes two models that work in tandem to challenge and reinforce each other. The net result of both LLM and GAN is that it removes the need for human intervention at critical points, where the prompt itself can be resolved to be useful in the completion of complex tasks such as designing a learning experience, or to paint a scene initially described with words.

While generative AI is certainly a leap forward in many respects, it is still considered to be a form of Narrow AI. This is because it has not yet attained Artificial General Intelligence (AGI) and can only produce output based on what it is trained upon, meaning that it cannot grow or adapt outside of those constraints. By understanding the progressive capabilities of AI, we as educators can grasp what the scientific community in general is aiming to achieve, and what may be possible within our students’ lifetimes.

How Can We as Educators Grasp What Makes Generative Artificial Intelligence Unique?

Services that use Generative Artificial Intelligence, such as ChatGPT, Claude and Bard, have grabbed headlines for their capabilities and the technical approaches to be able to mimic human language structures. While it is common for emergent technologies to evoke a sense of wonder when they are first introduced, it is useful to push past the wow factor of GenAI and evaluate how it best fits into our communities. In order to do so, it is useful to contextualize its importance and identify what makes it unique.

To put it simply, the groundbreaking approach that differentiates GenAI from other kinds of AI lies in its ability to identify context and purpose in a stream of data, thus rendering it able to turn questions into answers, text into images and video into text. For text, it does so by essentially guessing — a series of probabilistic determinations based on what is most likely to be an acceptable output. For images, it essentially hosts a game between two models.

Another notable aspect that makes this technology unique is that the transformers themselves require a vast body of input. Thus, initial versions were produced using the entirety of the public internet’s content, up to a given cut-off point. Since the machines themselves are able to produce quality content, and models can improve in proportion to the increased volume the machine can provide, there develops the potential for better and greater improvements embedded within the technology itself. This “virtuous cycle” is why we expect exponential development of this form of AI over time.

Essentially, as can be seen, the uniqueness of the technology is that the path towards further developments is readily available. Not only are future improvements built-in, but also the approaches it uses are ripe for further innovations. The two models described above, one best for text-based and the other for image-based models, may lead to hybrid approaches that give it even more capabilities.

What Is Generative Artificial Intelligence Best At?

School communities play a significant part in determining the acceptability of certain tools and utilities. With that in mind, and given our improved understanding of AI from the above discussion, what sort of tasks would artificial intelligence be best placed to assist us in our shared mission to advance international education?

Having understood its capability and approach, identifying where AI excels centers around its core ability as a superefficient processor of vast data sources. In that way we can think of AI as excelling at leveraging databases for individual users, in the same way that a calculator streamlines mental math tasks.

Tasks that benefit from references to many data points, or questions that span across domains, are areas where AI excels. In that way, we can see how it is now possible to utilize AI to design learning experiences that may previously have benefited from consultations with experts. A teacher who knows their students well can reach across subjects with more confidence, as they now have a tool that can assist with the navigation of complex concepts. Planning units and feedback sessions can benefit from increased understanding of research in psychology, and can be more explicitly linked.

To put this point in terms of the Substitution, Augmentation, Modification, and Redefinition (SAMR) model of classroom integrations, the best utilization of GenAI goes well beyond saving time for remedial tasks. For example, a teacher asking these tools to “Write a description for a unit plan about Photosynthesis for Grade 5” would merely be replacing the teacher’s ideas with ideas from an AI model. Where artificial intelligence excels is in being trained on not just descriptions, but on entire databases that connect any data points linked to unit descriptions, such as learning experiences. Instead of being a tool that accomplishes a number of small tasks sequentially, it can redefine planning altogether by enabling an approach that is seamlessly contextualized.

What Are Some of the Limitations of Generative Artificial Intelligence?

It is important to understand that while AI introduces great benefits, it also has limitations that are highly relevant to practitioners in education.

One limitation is inherent in the terminology we use to refer to this technology. While we call it an artificial “intelligence”, it is not the kind of intelligence that can be relied upon as a source of authority. It is readily possible to ask a GenAI tool to produce outputs that have little or no basis in reality. As a technology, it does not possess a lived experience to enable it to check for biases. Since it is detached from principles, it can stray from important values such as diversity or inclusion. Due to this, if the output from GenAI is immediately lifted into a published or submitted task, email, or unit plan, there is a strong risk that it may be disconnected, disjointed, or otherwise inconsistent with hidden assumptions. This presents school communities with the unique challenge of adapting their Academic Integrity Policy to define and codify acceptable behaviors that uphold ethical standards when using GenAI for both everyday productivity and official submissions.

As we have seen, our existing AI technology is still considered to be narrow in scope. It mimics creativity by copying existing structures, but it doesn’t “think” outside of a box, nor inside of one, even. It is no different than a calculator that solves math equations at a dizzying pace, except that AI resolves statistical models with incredible efficiency.

The end result is extraordinarily useful, but in a classroom setting, GenAI needs to be contextualized. In our collective quest to become life-long learners, we ought, as a teaching and learning community, to have a clear framework for understanding how we can greatly benefit from AI, but also what roadblocks are inevitable.