AI applications that learning professionals will definitely find useful

Posted by Greten on 27 Mar 2023 under Tips, Tools

Recent advances in artificial intelligence (AI) have opened up new education and professional development possibilities. From personalized learning to intelligent tutoring systems, AI tools transform how teachers and L&D professionals approach their work. By harnessing the power of machine learning and natural language processing, these tools can help educators and trainers create more engaging, effective, and efficient learning experiences for their students and employees. In this entry, you will learn some of the top AI tools for educators and show you how they can help you take your teaching and training to the next level.

AI are computer programs that can generate content using natural language processing (NLP) and natural language generation (NLG) methods. NLP helps computers understand human language, and NLG helps computers generate human-like language. This technology can be used to create lesson content and activities to accomplish specific learning objectives.

Another way of expressing it is the computer program gathers data, as in large bulk of data, analyzes patterns, and then, using the users' instructions, creates a product or renders judgment using those data. Sounds familiar? That's basically the top two levels of Bloom's taxonomy of cognitive learning objectives. Yet, computers are learning now and doing it more efficiently than humans. However, just like any technology invented by humans, AI is a tool meant to help humans. We continued teaching arithmetic in grade school after calculators were invented, and thus, there is no need for us to stop teaching learners how to evaluate situations and produce deliverables just because computers can do it now.

Writing content and documents

AI language models are computer programs that can understand and generate human-like language. Some examples of large-scale language models include GPT, developed by OpenAI; BERT, developed by Google; and ELMo, developed by researchers at the Allen Institute for Artificial Intelligence. These models can be used for various applications such as chatbots, virtual assistants, and even generating news articles.

You can use AI language models to generate text-based documents and content such as lesson plans, course outlines, experiment guides and procedures, course content in a readable format, and quizzes and exercises. As a result, it can reduce the time you spend on many of your tasks. Without an AI language model, you must write and then edit your work to ensure its quality and accuracy. Using the AI language model, you'll spend more time editing than writing. All you need to do is to type something like "write me a lesson plan discussing Newton's third law of motion for grade 10 students," and the AI will write it for you. You can also create an entire course by typing a prompt such as "create an outline for a course on conflict resolution" and then, for each bullet point in the outline, prompt the AI to write a script for a video elearning about that subtopic.

A bronze robot wit barrel-like body and camera face holding a plume and about to write on a blank paper.

It does not mean that you won't have any work. Your work shifted from more writing to more editing while reducing the overall time for both tasks. AI language models, mainly when prompted to write longer content, tend to make things up if their stored data is insufficient. Imagine you are conducting a job interview and are impressed that the applicant is knowledgeable and answers confidently. Then, you begin asking questions that are beyond their current knowledge. While some will admit that they do not know, some applicants will try to fib their way through using imagined or guessed answers while still trying to sound confident. You must be knowledgeable or an expert on the topic you're asking to determine if the applicant is answering correctly or making things up. Working with an AI language model is like that. After it produces the content that you need, you must check it for errors and edit it accordingly.

So far, I have experience using two AI language models, ChatGPT and Bing AI; both use the same underlying GPT language model, with ChatGPT using GPT-3 and Bing AI using the more advanced GPT-4 model. GPT-4 model is also available in ChatGPT but not in the subscription level I am using, so I cannot speak about that. ChatGPT can write longer content, and it stores all conversations until you decide to delete them. However, you cannot access previous conversations sometimes, especially if ChatGPT is under maintenance or there are too many people using it at a time. It is also more prone to fibbing. You can tell it is fibbing only if you are already knowledgeable of the content it is writing or if you did some cross-checking with external references.

On the other hand, Bing AI tends to answer more correctly, and it even cites references. However, its answers tend to be very short. You can force it to write longer by insisting on a certain number of words, which will make it more prone to fibbing. I also encountered situations where Bing AI provided information contradicting its cited reference. However, the fact that it cited references makes cross-checking way easier. Lastly, BingAI can store only up to certain pairs of prompts and responses, requiring you to delete all previous conversations if you reach this limit and need to generate new content. When I signed up, it was six pairs. Now, it is 20 pairs. I am unsure if Bing increases the amount of allowed conversation for everyone regardless of when they signed up, or everyone starts at six pairs and then increases from there.

Editing and paraphrasing content

After creating your content using the AI language model, you must edit it. Do you know how to make your editing task easier? By using AI, of course!

It seems ironic that I mentioned that as an educator, you will end up doing more editing and less writing. Yet, we're talking about editing using AI. You can ask ChatGPT and Bing AI to paraphrase what they wrote or even what you or others wrote, and they usually write with near-perfect grammar and syntax. However, they still cannot do these tasks perfectly. For example, if you ask ChatGPT to paraphrase a long article, you will still sometimes find a long sentence or group of consecutive sentences that did not change from the original. Another reason for using a separate AI tool for checking grammar is that there are situations where using AI to produce content is not feasible. Such cases involve writing from actual observation and experience. For example, writing about the results of an experiment or clinical trials. Other examples of writings that an AI cannot produce are memoirs and reflection papers. As you write these materials, you still need to edit them. You can edit manually or make your life easier by enlisting the help of AI.

A clockwork robot hovering a pen on top of a paper tht already has writing on it while several papers and a pile of book are scattered on the table.

This is where editing and paraphrasing applications come in. Two that I find helpful are Quillbot and Grammarly. They are both useful as both grammar-checker and paraphraser. Still, Grammarly specializes more in the former, while Quillbot in the latter. Quillbot can replace words with synonyms and combine, split, and mix sentences at your command. Meanwhile, Grammarly focuses on providing recommendations for your grammar, clarity, and style, which you can accept or reject. In their free versions, Quillbot is limited by the number of words, while Grammarly is limited to doing only basic grammar checks like subject-verb agreement.

Detecting AI-generated content

The saying "fighting fire with fire" works in certain situations, and using AI is one of them. Just as much as teachers, trainers, instructional designers, and other educators can use AI to produce content, students can also use them in the same manner. For this reason, one of the concerns many educators raise is that students will use AI to write homework essays and term papers and pass it as their own work. Hence, there are also AI applications whose primary function is to detect if a work is written by AI. Copy your student's work from a soft copy and paste it on a text field. The application can then display the probability that the work was written by AI. They indicate the likelihood that a written work is AI-generated as a percentage or using descriptive words like high probability, likely, or unlikely.

The catch in using these tools is that they are probabilistic. Thus, there is a chance that a work detected as AI-written was actually written by a human. You should be careful in using these tools and then accusing your students of cheating. If you are so worried that using AI will not reflect what they learned from your class, then you must revise your assessment tools. Otherwise, consider that the students will indeed use AI tools, or better yet, level the playing field by encouraging them to use AI tools. More on AI detectors are covered in this previous entry: Why we should not be afraid of AI-written essays?

AI text-to-speech

Earlier text-to-speech applications do not use AI. They are essentially a set of prerecorded syllables and words spliced together to turn text into audible information, making them sound robotic and computer-generated. They are widely used in asynchronous elearning and to provide accessibility to those with visual impairments.

In asynchronous elearning, non-AI text-to-speech is used to time the animations and the appearance and disappearance of various slide objects. These text-to-speech audio files will eventually be replaced by recorded human speeches, with little adjustments to the timing of slide objects after that. However, thanks to AI, many text-to-speech applications are now available that can produce human-like speeches. They can sound friendly and lively, change tone to something that sounds like a question if there is a question mark or the sentence sounds like it is asking a question, and words with the exact spelling but different pronunciations can be pronounced correctly based on the surrounding text. There are also AI text-to-speech applications that can convey emotions.

A shiny robot with what looks like an LCD display for eyes is speaking with microphone in front of a podium.

AI-powered text-to-speech uses NLP just like content writers and grammar checkers. Text-to-speech requires an AI-based system with NLP capabilities. The NLP engine generates human-like voices, which make the text more interactive and fluent. Sound waves are represented electronically as audio signals. The speech signal is digitally represented as a series of numbers. Speech scientists use multiple feature representations in the context of TTS to describe specific features of the speech signal, allowing AI models to be trained to generate new speech.

There are many available AI-powered text-to-speech available in the market. So far, I have tried Speechelo, WellSaid Labs, ElevenLabs, and IBM Watson Text to Speech.

Speech transcription

The opposite of AI-powered text-to-speech is AI-powered transcription. That is, using AI to translate an audio speech into a written format. So far, I've tried only one such application, Otter AI.

Transcription applications are more commonly used in corporate meetings, interviews, and other situations where people always say something impromptu. It is not usually used in teaching, training, and online learning because the written content exists first as part of the instructional design before they are deployed to learners.

A white robot with what looks like an LCD display for eyes is writing something on a notebook while wearing a headset.

However, if you recorded a synchronous training, onsite or online, and you, or perhaps one of your learners, said something that you believe is insightful, interesting, or important information that is not originally in your learning materials or lesson plan or that the training was delivered in a manner that's way better than planned, you might want to deploy the recorded lesson as asynchronous elearning and use transcription to create a transcript for your video. Such transcription is also beneficial to those learners with hearing impairment.

Similar to using AI to write or paraphrase, you still need to edit the transcription. It's possible for the AI to pick words from the audio file incorrectly. For example, you might have mispronounced some words or said them while music or a sound effect is playing, thus confusing the AI.

Image generators

AI image generators employ a concept known as Generative Adversarial Network (GAN), which uses two opposing algorithms. The first neural network generates new images from previously stored data. The second neural network judges how close to the real thing the image is based on actual samples from the internet. Once scoring the image for accuracy is complete, the data is sent back to the original AI system. The technology behind them varies, but the most recent AI image generators use diffusion models. These work by destroying their training data by adding Gaussian noise and then reversing the process to remove noise from the image. The model applies this process to random seeds to generate images.

A small robot holding a paint brush while painting an abstract art on a canvas.

There are various ways you can enter prompts into an AI image generator. One is by entering a prompt in a small text box, such as "nuclear power plant in the middle of a desert," while others include some dropdown option for art style or type of image, such as pictures, 3D graphics, illustration, or vector.

So far, I've already used four AI image generators: Microsoft Bing Image Creator, OpenAI DALL-E2, Images.AI by Unite.AI, and Runwayml.

My experience using AI image generators is that they are more geared towards producing beautiful art than images that show accuracy and do not work well with systems of several small parts. For example, using the prompt "nuclear power plant in the middle of a desert" show you a compound with several buildings, chimneys, and reactors as seen from the outside. However, suppose you ask for something like, "cross-sectional diagram of a nuclear reactor". In that case, it's unlikely that you will see uranium fuel rods, control rods, and cooling pipes, and even if you did, they would not be in their proper places.The images below are generated using the aforementioned prompts. The nuclear power plant on the left matches the typical appearance of a nuclear power plant, while the illustration on the right looks like a beautiful machine that may look good in science fiction book cover, but does not match how a nuclear reactor appears on the inside.

On the left is an AI-generated image of a nuclear power plant as seen from outside. On the right is what looks like a glowing engine.

AI generators also work worse in objects with long thin parts. For example, asking an AI image generator to draw you a wind farm would sometimes cause the rotor blades to look like a tangled mess.

Conclusion

AI tools transform how teachers, L&D professionals, and other educators approach their work, creating more engaging and effective learning experiences for students and employees. For example, AI language models can generate text-based documents like lesson plans and course outlines, reducing the time spent writing and allowing more time for editing, and detect if a learner is submitting AI-generated content in an assignment that must reflect the learner's writing ability. AI-powered text-to-speech applications can create human-like and interactive audio content. In contrast, transcription applications can turn recorded synchronous training sessions into asynchronous elearning content. AI image generators can create images from prompts and dropdown options. However, they produce beautiful art rather than accurate images, and systems with several small parts or long thin objects can be challenging to render accurately. AI tools can help teachers and trainers take their teaching and training to the next level. Still, it's essential to understand their limitations and use them effectively.

Bibliography

External links

These are online AI applications covered by this entry. Listing them here is not an endorsement of superior quality over their competition. It just so happened that I get to encounter and use them first.

Last updated on 11 Jun 2023.

Share your thoughts

* Required. Your email will never be displayed in public.

Instructional design and educational technology for effective learning