David Jones – Innovation Blog

Current AI-Integrations

I have completed the first round of our AI integrations! Here are some “how-to” guides for what we have so far.

An article in the New York Times caught my eye yesterday. The Professors Are Using ChatGPT, and Some Students Aren’t Happy About It read the title. For the past six weeks, I have been coding AI integrations into Innovation. It caught my eye because I have been thinking a lot about AI in education. From the perspective of a teacher, it drives me crazy when my students submit ChatGPT-generated work and pass it off as their own. The cartwheels I have to do as a remote instructor to prevent this are pretty byzantine!

But I am also interested as a businessman. I aim to enliven innovation (and raise its notoriety) by the integration of OpenAI in every aspect of the site. During this feverish coding period since mid-April when we got our API key, I have coded apps that…

generate multiple-choice questions for tests, reading comprehension, videos, and Jeopardy-style games;
score short answer responses based on guidelines and model answers;
score longer essays based on rubrics and instructor-designed guidelines;
interact with students in online forum discussions;
generate composition topics and dictée practices for world language teachers;
generate custom grammar exercises for world language instructors.

Through the summer, I plan to add some sophisticated AI analysis options for student essays as well as rubric generators and a monitored chat.

In the New York Times article, student Ella Stapleton was a senior at Northeastern University. Her professor had used ChatGPT to generate lecture notes and failed to remove the telltale signs of its origin. Another student found that the comments a professor left on one of her assignments included the chat with an AI to help grade it. One student is suing her university, saying she was paying for instruction from the prof and not from an AI. Are they right to be annoyed?

Readers are no doubt familiar with the Talmud, a central work of Jewish thought composed of rabbinic debates spanning centuries. These debates often wrestle with how to interpret and apply biblical law to real or hypothetical situations. A hallmark of Talmudic reasoning is the use of analogy: to what extent does a current case resemble one already discussed and resolved?

This is the approach I would like to take in arguing specific ethical considerations regarding the use of AI by instructors. I began teaching in 1991. If we assume ethical principles to be fairly static, since right and wrong should probably not really change much, then what was right then is still right now.

In 1991, a public school teacher would have access to a commercially published textbook. This would typically come with a package of pre-made tests and answer keys, workbooks for subject-specific practice, maybe filmstrips or posters, and so forth. It was the common understanding that teachers were not expected to write their own textbooks or even design every one of their own lesson activities.

In 1991, a college professor would typically teach using a commercially published textbook selected for the course. Along with the textbook came instructor guides, test banks, lecture slides, and other supplemental materials provided by the publisher. Professors might adapt these resources, but it was generally understood that they were not expected to create every reading, assignment, or exam from scratch. The role of the professor centered more on guiding discussion, delivering lectures, and evaluating student work than on developing entirely original curricula for each course.

With regard to assessment, my teachers in the 1970s in my grammar school sometimes used a Scantron machine to score those tests where you fill in the bubble. They did not score the tests all by hand. My elementary classes were 35-40 kids to a class in a parochial inner-city school.

When I was teaching social studies here in New York State just before I retired, I was called upon each June to drive far away to meet with colleagues from other districts to score the essay portions of the New York State Regents exams. Two teachers graded each paper and we discussed the merits and the score.

In 1991, assessment at the college level often meant midterms, finals, and a handful of major papers or projects. In large lecture courses, teaching assistants might handle the grading of essays, quizzes, or lab reports, following rubrics or guidelines set by the professor. While professors were ultimately responsible for student evaluation, it was common for them to delegate portions of the grading process, especially in high-enrollment classes. The expectation wasn’t that every piece of student work would receive personalized feedback from the lead instructor, but rather that grading would be efficient, consistent, and scalable.

Returning to the students who are upset with their professors for using AI to generate lecture notes or to generate student evaluations, I think we can reason by analogy as did those Talmudic scholars in times past to ascertain what is right.

My premise, and this is after many hours of working with AI over a year or more, is that at this particular moment in history, the best AI has to offer is to be a rather naive, but sometimes insightful, young assistant. My teachers reviewed the commercially published tests and checked for typos and accurate keys. My professors supervised their teaching assistants, providing them guidelines and checking their work. My AI helpers, who at the moment are ChatGPT and Gemini, need guidance and supervision by me.

Commercially published textbooks, tests, workbooks, worksheets, and the like have been acceptable and welcomed for a century. No one would have asked the one room schoolhouse teacher to publish her own grammar books. No one would have faulted a full professor for having his assistant grade lab reports. In 1991, and this is before the demands of differentiating instruction, the teacher was the creative director of a plan to educate using resources that they had vetted and sometimes using assistants that they supervised. At the time, this arrangement was both normal and uncontroversial.

The introduction of AI as a source of learning or an assessment tool doesn’t diminish the instructor’s crucial role; it amplifies it in the same way a carpenters’ work was amplified by the invention of the nail gun. Just as educators have always been responsible for the quality and integrity of their classrooms, they must now extend that vigilance to AI. This active supervision ensures that AI enhances, rather than supplants, sound pedagogical practices.

Innovation has built all of its AI integrations around a clear philosophy: the instructor remains the expert in the loop. When AI generates test questions, they must be approved by the instructor before being added to an assessment. When AI scores an essay, the instructor sets the rubric, defines the guidelines, and reviews the results before incorporating any of them into the student’s grade. When AI participates in student discussions, it does so within parameters the instructor has defined — including tone, context, and purpose — and under active supervision. When AI grades short-answer responses, it relies on model answers the instructor has already selected and endorsed.

At every turn, Innovation’s workflow puts the instructor in the role of guide and gatekeeper — promoting good old-fashioned professional oversight through the design itself.

The profs who failed to properly read and edit the course materials or assessment comments are to be chided for editing poorly. But the expectation that students have that instructors be the author of all of their course materials is born of an age when technology makes this at least theoretically possible, although not practically so. The expectation that no assessment will be outside the hand of the instructor is a new fashion, also imagined in a context of hyper-alertness to AI usage. One professor noted in the article was criticized by the student for chatting with the AI about writing the critique of the student’s work. But this is precisely what a professor might do with a live assistant in days gone by! The difference is that the student of the past would have no knowledge of the discussion.

One of my remote students this year had nothing good to say about one of her teachers. She cited the example of the fact that her teacher got her powerPoint slide shows from ChatGPT. If that powerPoint were of poor quality or included incorrect information, I could agree. Where this student goes wrong is in thinking that the general notion of getting learning resources elsewhere is illegitimate or unprecedented. The wrong would be in presenting shoddy or incorrect information, not in failing to be the author of everything.

✨ Let the AI Teaching Assistant Help you Generate Questions to Embed with Video and PDF.

Need comprehension or analysis questions for a PDF or a video? That can be incredibly time-consuming — but Innovation’s Teaching Assistant is here to help!

Just open the Étude app. Upload your PDF or paste your video embed code, then add it to your AI request configuration. In seconds, you’ll have high-quality questions based on your stimulus, crafted in the language and level of sophistication you choose.

Check it out and see how much time you’ll save!

✨ How To Score Writing Tasks Using AI

The AI Grading Assistant integrated into Innovation is a powerful tool designed to streamline the assessment of student writing tasks.

With just a click, you can apply one of the pre-installed rubrics or upload and use your own custom rubric. After a brief processing time, you’ll receive a detailed second opinion to help you balance and validate your own evaluation of the student’s work. The AI’s assessment is based both on the selected rubric criteria and on the advanced capabilities of a large generative AI model. As of this writing, Innovation uses GPT-4o for essay scoring, ensuring fast, consistent, and thoughtfully reasoned feedback.

Updating our Terms of Service for the AI Integration Rollout

We are excited to be rolling out our massive upgrade to AI this month! Already, subscribers will notice the little purple buttons all over the site controls offering AI assistance with test question generation, grading student work, and tasks specifically geared toward teaching modern languages.

Subscribers will be invited to agree to the new terms of service when everything is up and running in June. Here is the text of that change:

AI-ENABLED FEATURES AND RESPONSIBILITIES

Innovation Assessments LLC now provides access to a variety of artificial intelligence (AI)–powered tools to enhance educational services. These may include, but are not limited to, automated test question generation, grading support for essays and short answers, rubric design assistance, writing prompt creation, and supervised chat-based discussion with AI for students. Users should be aware that AI-generated content may contain inaccuracies or reflect inherent biases, and human oversight is crucial.

Use of AI services is subject to the following terms:

Students may only access AI-powered chat or discussion tools under licensing of their teacher, and only if the teacher has enabled this feature for their activity.
All AI-generated content is provided “as-is” and may require human review. Teachers are responsible for reviewing all materials prior to use in assessments or instruction.
Essay grading by AI is advisory in nature. Final evaluation remains at the discretion of the teacher or institution.
Teachers and students may not use the platform’s AI features to submit or generate content that is harmful, discriminatory, or in violation of academic integrity policies (which include, but are not limited to, plagiarism and unauthorized assistance).
Innovation Assessments LLC reserves the right to monitor, restrict, or disable AI usage in cases of misuse, abuse, or usage patterns that negatively impact the platform’s performance or other users.

Innovation Assessments LLC will handle data generated through AI features in accordance with its Privacy Policy. Innovation Assessments LLC may update or modify its AI-powered features and functionalities over time. By using any AI-related features, you acknowledge the limitations of current AI technology and agree not to rely solely on AI-generated outputs for high-stakes educational decisions.

AI TOKEN USAGE

Access to AI features is governed by a monthly token system. Each account tier includes a set number of AI tokens per month, which may be used for supported features such as question generation, AI chat, grading support, and other automated tools. Tokens renew every 30 days from the date of paid subscription.

AI tokens do not roll over. Unused tokens expire at the end of each 30-day cycle.
Users may purchase additional token bundles if their monthly allotment is exhausted before renewal.
Token usage is calculated by the AI company and may vary based on the feature and the amount of text processed in both the request and the response. Requests with more text will consume more tokens, as will more detailed or lengthy AI-generated content. Higher-cost actions (e.g., full essay scoring) consume more tokens.
Token balances and consumption details are available to account administrators within the platform dashboard.
It is the user’s responsibility to monitor token use and purchase additional tokens as needed.

Innovation Assessments LLC reserves the right to modify token costs, tier allowances, or features covered by tokens with notice. Abuse of the token system may result in service restrictions or termination.

✨ Make a Jeopardy Game with Innovation’s AI Integration!

My students have always loved playing Jeopardy! Oh, sorry, trademark issue… I mean “Jeopardy-like trivia games in class”.. 😏

Our game is called “Ventura”.

Innovation has had a fantastic app for generating such games for years now. As part of our integration of AI into our whole system, teachers can now employ our AI teaching assistant in generating Jeopardy games!

Just like for creating test questions, teachers configure the request to OpenAI in the Ventura game.

Use the teaching assistant to generate questions. Use them as-is or edit them. Add images or audio clips!

Holy cow, I remember the old days back in the 1990s when I would use PowerPoint to make a Jeopardy game for review day. It took a really long time to enter all the questions and answers even when I had a template game prepared!

Now I can make a game in 2-3 minutes! The test generator is using one of OpenAI’s contemporary models, so you can rely on the question quality.

Enjoy!

✨ Introducing OpenAI – Innovation Integration!

Innovation is excited to announce that we are working diligently to integrate OpenAI artificial intelligence into all aspects of the Innovation platform! We are developing a virtual teaching assistant that will help teachers generate “just right” learning activities and assessments.

Look for the purple button throughout the site …

… or the ✨ emoji to signal where AI integrations have been installed.

AI integration has been a goal at Innovation since before the release of the generative AI models in the early 2020s. Since 2018, Innovation has sported a pretty nifty algorithmic AI that helps grade short written work submissions and proctor student online activity. Now with the integration of OpenAI’s generative AI models (3.5 and 4o), we can truly realize the dream of a highly productive and efficient virtual teaching assistant!

When Innovation started out under a different name some 25 years ago, it was mainly a test generator (thus, the name). Since the pandemic, it has been meeting the needs of remote instructors and in-person classrooms alike not only in assessment but in content delivery in a variety of subjects, including for world language instruction.

Innovation is a place to create. It’s a place where teacher-authors can generate “just right” learning activities and assessments for their teaching context instead of textbook company generics. But we understand that secondary school teachers are even more busy than ever, seeking to meet the growing diversity of need and ever changing objectives and curricula. With Innovation’s AI teaching assistant, teachers can redirect their creative energies to the big picture of student learning and measurement.

Check out this simple how-to, for example, which illustrates how you can use the AI teaching assistant to help you generate a test.

Check back with us as the spring unfolds! At this writing, we have already installed openAI teaching assistants in:

Test generator (multiple-choice or short answer)
Études (for adding questions to PDF or video)
World language apps:
- composition generator
- composition assessment
- grammar workspace
- translation drag and drop “Scramblation”
Essay grading
Vocabulary flashcards and assessment

Discouraging Over-use of Translators in Online World Language Classes, Part 2

The best way to learn to write well is to write and review the selected errors with an instructor for learning and practice. When I was teaching in-person, I would assign a composition in my French classes at the end of the unit to be done without any notes or references. I would then gather up the mistakes students made and we would commit to study them and learn to correct them. It is a great method to promote accurate and fluent writing in second language.

Teaching online, however, my work submissions from students in free-write compositions, even in one-on-one classes, were often AI-generated to such a high degree that the students really could not claim ownership. In one-on-one lessons, I did not always let on that I knew what they had done or sometimes I just made light of it. I could turn it into a useful exercise by asking the student to explain some tenses they used or some structures. But it is not the same. I felt like going into remote learning I had lost an important language training practice.

I have been enjoying success with a new kind of exercise for teaching composition. I will not claim to have invented it as surely someone, somewhere, has already done so. But I do say this method is not one I have seen or used before.

The student is presented with a series of prompts that constitute a composition in the target language of two to four paragraphs. The Innovation app presents them with one prompt at a time.

The prompt is a set of sentences that are in random order. One task for the student is to read these and arrange them in the best order. I design these using the unit theme vocabulary, so it is good practice in reading comprehension as well as in composing cohesive writing samples.

Often, especially for younger learners, I remove a word from each sentence and put it in a word bank. So now students have to not only rearrange the sentences, but they need to fill in the blanks based on context. Again, it’s a support for reading and composition.

Another strategy, especially for advanced learners, is to display the verbs as infinitives for the student to conjugate. In addition, I can remove the transitional phrases and ask students to supply them. Sometimes I include a prompt asking the student to add one sentences of their own, perhaps by providing an example of what it being discussed.

The prompts are displayed to students as images and not plain text, which creates an obstacle for those who would want to paste it into an AI to do the work for them. Displaying the prompt as an image file forces the student to write for themselves. An added benefit is that this promotes more lengthier writing for students who normally write way too briefly.

Innovation makes this easy! I select the “Single Short Answer task” from the Short Answer controls. I add each prompt with the answer key.

Then I add the screenshot of the prompt.

The short answer app at Innovation lets me place obstacles in the way of AI use and helps me generate practice exercises that help students develop their composition skills in the target language. Students have practice seeing and copying language in its standard and correct forms. They practice reading comprehension and the current theme vocabulary. They can rehearse transitional expressions and devising cohesive compositions. Prompting students to “Add one sentences of your own” prompts synthesis.

The tasks are easy and quick to score. From the course playlist, select Task, score One Student, and easily compare the student’s response to the answer key.

Although my preference is still for a free-write composition assignment, I can see many advantages to this one. I began developing this task with a mind to place obstacles in the way of student misuse of AI translators to do their work. I think I ended with an exercise that may arguably be actually better than free writing.

Discouraging Over-use of Translators in Online World Language Classes, Part 1

AI assistance and translators such as Deepl and Google Translate are very accurate and useful tools. When I assign my French students certain tasks, I expect they will use these tools to help them just as I would have expected students in-person thirty years ago to use a French-English dictionary to help with spelling and new words on certain tasks.

The problem is that the temptation to just have the AI generate the work is a strong one. It is important for remote instructors to place obstacles in the way of this practice, which not only undermines the student’s training but represents an ethical pitfall.

Imagine this scenario: an online AP Spanish class where major assessments are take-home tasks like essays and video-recordings of presentations. Using the traditional paradigm for this assignment, the student is given guidelines and due dates and a rubric with a graphic organizer. The instructor provides all that. Then the due date comes, all the work is in, and the instructor begins to review the work. What impressive vocabulary! What elegant grammar! And yet, reflecting on the spontaneous language generated by these same students in video-conference live sessions, it is hard to believe that this work could come from some of them.

All of my remote courses begin with a training film of sorts in which I explain the concept of academic integrity and ownership of one’s work submissions. I explain that it is expected that students will learn all of the new words they incorporate into their work submissions so as to maintain ownership of the task. I demonstrate using a translator properly and improperly.

A very useful strategy is one I have used since in-person days decades ago: simply ask the student the meanings of the words in their work that I suspect they do not know. In the remote learning context, this can be difficult to arrange, since there is no easy way to pull a student aside during class and conduct the interview about their work. That’s where Innovation comes in.

Sample student work that was too perfect for their demonstrated abilities.

It only takes a few minutes to select words and phrases from the student’s work submission that I believe they do not likely know. I select seven to ten words or phrases and I generate a short answer translation quiz using Innovation’s Quick Short Answer.

I enter a title, maybe set the category, and enter the words with English first, an equal sign, then the French.

Innovation’s app separates the word from its meaning by the equal signs. Once I have generated the quiz, I access the quiz Master app. I set the time limit to 1 minute for 7-10 words and I turn on the high security.

With the high security on, the assessment will submit and lock the student out if the student leaves the window to click on something else. Only the teacher can re-admit student to the quiz. The window resizes to full screen when the student starts the task and if they resize it smaller, the proctor records it. The proctor also records start time, how much time spent on the questions, whether text was pasted, and more.

As a final step, I lock the quiz up to only certain access codes. This allows control of how many times a student can restart the task. Simply select the Task dropdown from the playlist and then select Lock. Instructors can view the access codes from the Task dropdown or can generate one key by clicking the One Key button next to the title.

Sometimes, I will ask that the local facilitator proctor the student during the quiz so that they cannot look up the words on their own device.

So now what does one do with the results? When first introducing this strategy to students, I explain that it will not affect their grade “this time” and that it is a good reminder to make sure students have full “ownership” of their work. I may randomly select students for this verification, or if it’s a small class I may include it as a portion of their grade for a task and send one to everyone.

It’s not necessary for a student to get 100%. I usually take the quiz first to test it out; to see how many I can do in 1 minute. Even if they do not get 100%, I can learn a lot from their responses. For example, one student got 44% right of 8 and did so by skipping around. I interpret the skipped words as ones she forgot and intended to get back to later. Another student only got 33%. I interpret that as definitely being a sign that his work submission had too many looked-up words he did not know. I let him off with a warning this time and a reminder about academic integrity and ownership.

I once had the experience of taking over a class part way through the year. No structures had been in place to discourage inappropriate use of AI. The grades were all outrageously good. Some students were rarely in attendance and only handed in work that was graded. They did this work using AI, so it was no real effort. This really is a terribly corrupt system, especially given that there are students in nearby schools taking in-person classes who have to really do the work for their marks. There are honest students with good attendance who have lower grades for their honesty. It was an AP level course. Now, you might argue that the students would not possibly be ready for the AP exam if they took the course this way. One would think that would deter them from cheating. But upon reflection, it’s clear that having a 98 in an AP class on one’s transcript, even if one only scored 2 on the exam, could be valuable for college admissions considerations. So, no, it does not deter them.

Remote learning has enormous potential. I have great confidence in it. We instructors, we need to learn how to maintain the same standards as we had during in-person sessions. We cannot allow a situation to arise such that students in remote classes can just become pass-through vehicles for AI translators that do all their work. That situation would become a sort of scam. In part 2 of this topic, I will present a strategy for teaching composition in this new world of AI-assisted homework.