The arrival of new AI technology has sent the world of online education abuzz. The new technologies have brought new excitement to e-learning, but it’s difficult to tell what’s hype and what’s not. What will the impact of AI be in the online learning space? What’s likely to change? What will remain the same? We’ll take you through some major innovations in Artificial Intelligence (or, more accurately, Machine Learning), and what they mean for educators and e-learning at large.

What is AI?

Artificial Intelligence technologies are technologies that allow computers to mimic the behavior of humans. Recently, these technologies have gotten quite good at performing creative tasks previously thought impossible for machines – such as writing poetry, composing music, or rendering unique art.

These technologies might be more accurately termed ‘Machine Learning.’ Machine Learning technology works by changing the way a program behaves based on what it’s ‘seen’ before. For example, a machine learning program might be able to generate a picture of a cat after seeing many pictures of cats. Here is a cat which does not exist:

Someone gave a program a lot of kitties to analyze and the computer came up with this cat picture on command. Neat!

The term Artificial Intelligence brings to mind a vision of fully autonomous, thinking, feeling robots ready to take over humanity. The more you peel back the layers, though, the more you discover that while these programs are quite clever, they’re still a long way off from matching humans in terms of general-purpose intelligence. This is why ‘Machine Learning’ is the more accurate term. However, for the sake of this article, we’ll use both terms interchangeably. After all, everyone else is doing that and we want to be cool, too.

How are advanced teams using AI in E-Learning?

There are three main ways that Machine Learning is being used for online education, namely:

  1. AI Tutors for Online Courses
  2. AI Content Generation for Online Learning
  3. AI Created Problems

Let’s have a look at all three…

ChatGPT will make a few appearances.

1. AI Tutors for Online Courses

One of the most exciting uses of Artificial Intelligence for online courses is the ability to have an ‘on-demand’ tutor that can help learners when a teacher is not available. This is made possible through the use of Large Language Models (LLMs) like ChatGPT. Large Language Models work like other machine learning programs– they’re built by analyzing many, many written texts, including encyclopedic articles, online blogs, e-books, and news. After taking in so much information, they’re able to respond to prompts in a useful manner.

For instance, we prompted ChatGPT with “Could you explain the linguistic function of an adjective?” and it responded:

Certainly! Adjectives are a fundamental part of language and serve several important linguistic functions. An adjective is a word that modifies or describes a noun (a person, place, thing, or idea) by providing more information about it. Here are the primary linguistic functions of adjectives:

  1. Descriptive Function: Adjectives are most commonly used to provide descriptions or characteristics of nouns. They help paint a more detailed picture in the reader’s or listener’s mind. For example:
    • “The blue sky” describes the color of the sky.
    • “A tall building” describes the height of the building.
  2. Quantitative Function: Adjectives can indicate the quantity or number of nouns. These are often referred to as quantitative adjectives. For example:
    • “I have five apples” specifies the quantity of apples.
    • “There are many people in the park” indicates a large number of people.
  3. Qualitative Function

-ChatGPT, abridged

ChatGPT’s ability to summarize large swaths of information is impressive. Credit Janko Ferlic via Unsplash.

This is a quite comprehensive explanation of adjectives! The fact a machine is able to provide this is impressive, but lest we get ahead of ourselves, we need to think of the limitations of such a tutor.

Limitations of AI Tutors

The first major limitation of an AI tutor is that it may ‘make up’ information which isn’t true. The accuracy of a virtual tutor’s answers are more likely to be wrong if there isn’t a lot of information about a topic, or if the information it’s “trained” on is poorly curated, for example. These falsehoods have been deemed ‘hallucinations,’ and they can be a serious problem for a virtual tutor. A few recent examples of the consequences resulting from hallucinations include:

AI Tutors have a good chance to say the wrong thing with such confidence that learners are likely to believe falsehoods at face value, and instructors may wind up with students that have newly-introduced misconceptions they will have to tease out and correct.

Like a moth with an ‘eye’ pattern, an AI can build a convincing facade. Credit David Clode via Unsplash.

The second major problem with AI Tutors is their tendency to give answers they are not intended to give. A common imperative for a virtual tutor is that it should help a learner understand without giving it the answer to quiz questions. It shouldn’t substitute itself for a learner’s thought process. However, sometimes virtual tutors will directly give learners the answer, despite being asked not to do so. This effectively “short-circuits” the learning process for students. 

There are ways to reduce the likelihood that an AI will give an answer to a quiz question. You can add pre-configured prompts telling the ML Tutor not to answer the question directly. You can tell it to only respond through asking questions for a Socratic experience. However, the phenomena of ‘jailbreaking’ an AI is a sport in itself. Jailbreaking is getting the AI to unshackle itself from the rules it was programmed to follow in order to force it to do things it is not intended to do.

In fact, there is a game you can play where you try to jailbreak progressively tougher and tougher AIs. Likewise, there are collections of jailbreaking prompts available online. If someone wants to get an AI to act beyond its intended use, they can.

The third major problem is not precisely a problem with AI Tutors so much as a subtle, yet major limitation of their abilities. That limitation is that the quality of the answers is greatly limited by the quality of the questions asked. Much like ‘Google-Fu’ has become an important skill, writing a good prompt is a skill in itself.

Prompt writing has proven such an essential skill when working with AI that some have taken to calling it ‘prompt engineering’. This characterization greatly overstates the skill required, but it is still a learned ability. A good instructor can sometimes work around a student who does not ask good questions by observing what they don’t ask about, or recognizing common misunderstandings embedded in the questions they do ask. AI Tutors have a harder time performing these functions.

Are AI Tutors still worth it?

The robots are learning just like we are. Credit Andrea de Santis via Unsplash.

AI Tutors can absolutely improve online learning as long as their limitations are understood. While LLMs hallucinate at times, Machine Learning Tutors can cover topics that they have strong training data on. Instructors can also ask AI tutors several questions about their topic to verify if an AI tutor is appropriate for their subject.

While AI tutors can be jailbroken, they can often be more trouble to break than is worth doing so. They also are not unique to this problem – after all, a learner could just hop over to another chatbot not controlled by the instructor and ask it the question an AI tutor refuses to answer. So, this is not a real reason not to use AI tutors.

Finally, while AIs are limited by the ability of the learner to ask a good question, so are people. While people may be better able to spot the unstated misconceptions in a learner’s questions, the alternative isn’t the instructor spotting the confusion.

The alternative is no one spotting it.

That is to say, in many cases, the instructor for an online learning course may not be able to have one-on-one conversations with each learner to verify understanding and answer questions. And that’s what these bots are for – filling in the gaps where an instructor can’t be. Their intent is not to replace the instructor. The limitations of AI make it clear that the instructor should not be replaced. But if these limitations are observed and respected, they can improve an instructor’s ability to help their students learn.

2. AI Content Generation for Online Learning

Another area where AI is affecting online learning is in the content generation space. Instructors have to write many sections in courses that may last a semester or a whole year. That’s a lot of content! Any author will tell you that getting words to paper is a challenge. Many instructors would rather use AI to write their courses.

Limitations of AI for Generating Learning Content

Of course, there are some major limitations in using AI for generating course content. The biggest one is the same as with AI Tutors: ML tools will often hallucinate. And that means that an instructor cannot mindlessly let an AI write their course for them. They must audit and verify the output of the AI content generator.

The legal status of AI content is muddy. Credit Tingey Injury Law firm via Unsplash.

A second major limitation is not technological, but legal. In the US, recent court rulings have determined that content created by AI technologies is not afforded copyright protection. The technology is so new that many of these cases have not yet made their way through the legal system to their final destination, but the preponderance of case law so far indicates that the copyright status of AI content is on shaky ground. If you’re concerned about retaining copyright to your courses, this is cause for pause.

Is AI Content Generation a good idea?

Maybe. AI Content Generators can build well written, clear instructional content. You will have to watch closely to make sure that no falsehoods slip through. Likewise, some illustrations may not be generable, and problems have to be constructed deliberately, so there will still be some work that must be done by hand. However, course authors stand to gain, on the whole, by using AI Content Generators.

While the legal restrictions are indeed concerning, most of them melt away once a human is sufficiently involved as an audit layer. That is, there’s a line where something crosses from ‘AI Generated Content’ to ‘AI Assisted Content’ — only the latter of which may benefit from copyright protection. Where this line is, is greatly dependent on the work, level of effort, and intentionality by the human modifying the AI Generated Content, and other factors. You should contact a qualified legal advisor if you’re unsure if your work qualifies. The line may also be moved by precedence set by future court cases, and may differ based on your jurisdiction.

Secondarily, if your concern isn’t making money off of the intellectual property powers of copyright, then you have every reason to enlist the help of AI content generators. For teams that give away their content for free but instead monetize certification or other value-added services, using ML Content Generation for your online courses is perfectly fine – so long the result meets the quality standards of your institution.

Of course, online learning platforms sometimes suffer from “Not Written Here” syndrome where faculty often feel compelled to write a new course of their own when there are high-quality courses on the same subject that already exist. We encourage course authors to explore and contribute to open courses and textbooks whenever possible.

3. AI Created Problems

An under appreciated but keen use of these new technologies is leveraging them in creating unique problems for learners to solve. LLMs are able to generate large volumes of authoritative-sounding text in response to a prompt. Some clever uses can arise out of this.

One tricky task in our modern world is learning to think critically – to evaluate what we’re reading and test to see if it’s true. ChatGPT and other LLMs frequently ‘hallucinate’, and that’s to our advantage here. Imagine giving a learner the output of a prompt to ChatGPT– and then having them grade it!

AI content generators have great potential in problem generation. Credit Compare Fibre via Unsplash.

Have the learner examine all of the claims in an AI’s essay, highlight them, find citations for them – or citations contradicting them. This could, in fact, allow AI to be the best tool for teaching people to be skeptical of what they read in mass, a skill that is needed now more than ever.

Likewise, an AI could be a debate partner. A learner could be assigned one position, and the AI another. A teacher would then review a record of the debate between the two and grade the learner on their performance.

Another example would be to have the AI generate questions about the instructor’s topic that the learner has to answer. Of course, the learner should be told that an AI is asking the question and might not understand, so that the learner can challenge hidden assumptions in the question.

Problems with AI Created Problems

Actually, it’s difficult to think of any reason why this one wouldn’t work. The main trouble would be in the execution – building the tools that would create these problems. Of course, this is nothing the team at OpenCraft, an Open edX Partner, couldn’t handle! So, if you’re interested, contact them here! Also, in these cases you’re generating new ephemeral text each time a learner views your course, so copyright is unlikely to be a concern.

Final Thoughts

AI technology brings a lot of fascinating possibilities to the online learning space. None of these allow education to be ‘fully automated’– there will always be a need for an instructor and course designers to audit the work of Machine Learning tools, as they are no more arbiters of truth than humans are. So long as the limitations of this technology are understood, the capabilities of AI can be leveraged to improve the online experience of learners the world over.

Also, in case you were wondering – this article was hand-written by Fox Piacenti, Business Development Specialist at OpenCraft, and not assisted by AI. Only those sections explicitly marked were AI generated. 🙂

Loading