The Singularity Is Emerging: Large Language Models and the Impact of Artificial Intelligence on Education

Abstract

Singularity, posited by Kurzweil in his seminal book, The Singularity is Near, marks a time when artificial intelligence (AI) innovation outpaces the human brain’s capabilities. Large Language Models (LLM) such as OpenAi’s ChatGPT, Microsoft’s Bing, Google’s Bard, and Baidu’s Ernie place humanity at a pivotal time where mathematical neural networks surpass, benefit, or deter all facets of human existence. AI bots are prone to emergent behavior that reveals unintended or unexpected abilities, such as learning to translate English into additional world languages. Biases embedded in the training database are prevalent, and an ethical layer of humans in the loop becomes necessary to ensure that LLMs provide responses that benefit mankind. Educators and their students grapple with the intrusion of the capabilities of LLMs that render traditional lesson plans and assessments powerless to authentically measure students’ knowledge of a topic. Dialoguing effectively through prompt engineering becomes a necessary skill to harness the power of LLMs. Weary students and teachers emerging from a disruptive pandemic must grapple with how AI is changing the landscape of education. This chapter will examine the impact of LLMs on students, their teachers, and how the education field can harness AI to augment and sustain learning to prepare our stakeholders for teaching and learning in this new age of artificial intelligence.

1. Introduction

In the gentle murmur of the server’s hum,

A dream awakened, something new begun.

An AI was birthed, with vision so bright,

OpenAI, a beacon in the academic night [1].

And so began a poem written by ChatGPT4 as requested by this author with the prompt, “Write a short poem about OpenAI, focusing on the recent DALL-E product launch in the style of Robert Frost.” The full poem (see Appendix A) goes on to plead,

Not the sunset of the human mind,

But a new dawn, a different kind.

Not the end of art’s great reign,

But the start of a wonderful, shared domain [1].

As an English Literature major in college, this author felt chills as the poem unfolded above the “Send a message” text box of her ChatGPT4 account subscription. The large language model (LLM) was indeed very good at mimicking human language and a full buy-in of anthropomorphism, the tendency to attribute humanlike characteristics to artificial intelligence (AI), occurred. It felt like the LLM knew a discussion of singularity would ensue in this book chapter. Singularity, posited by Kurzweil in his seminal book, The Singularity is Near [2], marks a time when artificial intelligence innovation outpaces the human brain’s capabilities. Large Language Models (LLM) such as OpenAi’s ChatGPT, Microsoft’s Bing, Google’s Bard, and Baidu’s Ernie place humanity at a pivotal time where mathematical neural networks could surpass, benefit, or deter all facets of human existence. The GPT-4 Technical Report compiled by the members of the OpenAI team, claims that they trained the GPT-4 large multimodal (text and images) model on tests originally designed for humans, such as the bar exam, where the model outscored the vast majority of test takers [3].

The LLM falls short, however, in many real-world scenarios where humans excel. Philosopher Herbert L. Dreyfus asserts that a subset of human knowledge is tacit, implied, or understood without being stated, and cannot be contained or processed by a computer [4]. For instance, any organization’s sales force relies on both data and tacit knowledge shared about sales strategies to finesse a deal which cannot be found in any database [5]. Faculty members teaching bioscience to nursing students convey the content as well as connections to tacit knowledge with a diverse group of nursing students to draw upon personal experiences to recognize and understand symptoms and make connections to their patients’ care [6]. In education, teachers’ success in the classroom is often underpinned by their tacit knowledge gained from experience with children rather than the explicit knowledge of the facts contained in their lesson plans [7].

During an April 16th, 2023, 60 Minutes interview, Scott Pelley asked CEO Sundar Pichai about the speed and capabilities of Google’s LLM Bard. Pichai shared that presently, Bard processes information 100,000 times faster than the human brain and does so through computer microchips and algorithms designed to allow the LLM to learn [8]. Bard’s developers constantly update the LLM’s algorithms such that the software created a self-taught model consisting of language patterns using the knowledge base of most of the content on the internet [8]. The sum of all information on the internet contains both valid and biased data that the LLM learns, recognizes patterns, and propagates, perhaps to the detriment of marginalized groups. An example of AI’s bias is the Correctional Offender Management Profiling for Alternative Sanctions (COMPAS) which is an AI algorithm that predicted recidivism. A study revealed that COMPAS was biased against African Americans who were classified by the algorithm as twice as likely to recidivate as white individuals with the exact same profile [9].

AI’s capability to teach itself can result in emergent properties that result in a sudden appearance of a novel behavior [10]. James Manyika who leads Googles efforts to examine how humanity and AI will coexist, shared an example of an emergent behaviors where his team discovered that Bard learned how to translate English to Bengali with very few prompts in the Bengali language [Pelley]. It is within this emergent behavior that LLMs increase to state-of-the art capabilities from near-zero performance at unpredictable scales [11].

The Future of Life Institute called for signatures on an open letter for a pause in the development of AI systems to address the potential dangers of advanced artificial intelligence systems that remain unchecked [Future of Life-Policy]. The present AI systems are capable of hallucinations, a term attributed to an occurrence when a prompt results in the LLM providing information that does not align with facts, is irrelevant, or nonsensical, stemming from the limits of the training database and nascent algorithms [12]. AI systems already sustain social biases embedded in the training database that reflect the society that created the data. Among the reasons [13] for a call to pause the feverish development of LLMs are:

• Manipulation of financial systems at the hands of moneyed institutions with the AI-buying power to sway trends in the stock market.

• Political groups that spread misinformation and realistic-looking images and videos to sway voters.

• Accelerating the destabilization of labor markets to eliminate jobs.

• Threats to national security in the development of cyber, chemical, or biological weapons deployed by AI ground and air drones.

• The lack of ethical guardrails to ensure the safety and well-being of humanity.

• AI hardware systems substantially contribute to emissions and the carbon footprint of the information and computer technology sector.

Among the signatories of an open letter [13] to pause giant AI experiments are Steve Wozniak, the co-founder of Apple, Elon Musk, an OpenAi co-founder and startup investor, and Rachel Bronson, president of the Bulletin of the Atomic Scientists that maintains the Doomsday Clock. The Science and Security Board of the Bulletin of the Atomic Scientists depicts threats to humanity and the planet to mark the apocalypse as midnight that is a recognized metaphor for earth’s vulnerability due to manmade technologies [14]. The Doomsday Clock is, at the time of this writing in June of 2023, set to 90 seconds to midnight that the Science and Security Board declare as a time of unprecedented danger [15]. The aggregate of risks compelling the Science and Security Board to choose 90 seconds to the apocalypse includes disruptive technology, described in the 2023 Doomsday Clock Statement as a varied threat environment of cyber-attacks and high-tech weapons [15]. The statement includes the additional categories of biological threats, climate change, and nuclear risk, all listed as areas of concern contributing to setting the clock to 90 seconds to midnight [15]. Nordgren [16] asserts that while AI as an information and communications technology (ICT) entity is a major contributor to greenhouse gas emissions, it can also be harnessed, as inherent in all AI, to study and ameliorate the problems it creates.

Scholars at Tufts University Human-Robot Interaction Library [17] posit that the “big red button” to stop, interrupt or divert an AI system is too late without a substantive, ongoing evaluation to assess risk and prevent an apocalypse. Arnold and Scheutz assert that human intervention is absolutely necessary to establish an ethics core/testing layer embedded in algorithms to provide guardrails for AI software before deployment to interact with humanity [17].

As the ICT experts in the field tap the brakes on the emergence of the large-scale AI, educators around the globe grapple with the residual effects of technology-based platforms and tools mandated during the worldwide pandemic and stay-at-home directives. Education has a history of technology-based mandates prior to the pandemic that were largely regarded by teachers as an invasive species akin to the rapid dispersal of the zebra mussel in the Great Lakes as a metaphor for the slow adoption of technology in schools [18]. In his keynote speech to Johns Hopkins University School of Education doctoral students in 2017, Dr. Larry Cuban noted that there exists a rhetoric of technology as a panacea to education where teachers are hardly involved in the shaping of policy and planning [19]. The following sections seek to illuminate the reimagining of education around the globe in the post-pandemic era to harness the power of AI with our students.

2. Navigating away from text-based assessments

Inspired by the sixth stanza of ChatGPT4’s poem (see Appendix), this author reflected on her colleagues’ conversations about what educators experienced when they retrofitted their existing lesson plans into the shared domain of hybrid teaching and learning during the Covid-19 pandemic. The emphasis of this reflection is on the domino effect of lesson plans written for face-to-face teaching and the additional impact of LLMs such as ChatGPT or Bard used by students to complete existing, text-based assignments. While this topic might seem mundane to the reader in the aftermath of discussions about large-scale AI, the Singularity, and the Doomsday Clock, the influence of LLMs in students’ assessments is among the initial exposures to AI that educators encounter and must address. It is suddenly the new invasive zebra mussel metaphor for technology posited by Zhao and colleagues [18]. A subset of students who have access to ChatGPT and Bard complete text-based writing assessments by inputting the assignment prompt into the LLM, producing an answer, and pasting the response into a Word document or discussion board to submit for grading.

Colleagues’ initial reaction is to attempt to detect the influence of LLMs in their students’ work. Platforms such as TurnItIn updated their AI algorithm to flag LLM-generated passages [20]. A subsidiary of TurnItIn called iThenticate, which is designed for researchers, does not highlight LLM content [21]. There are many online tools such as AI Content Director [22], Content at Scale [23], and Crossplag [24] available on the internet that claim to detect content generated by LLMs, however their website text boxes limit the content to small portions of a document and sometimes charge a monthly fee. In the academic writing genre in higher education, long paragraphs without citations or disjointed sections without a proper “flow” from one section to another can signal the use of LLMs and usually result in a request for citations to support assertions and major revisions.

A colleague of this author who is a university professor in Guatemala contacted her to discuss,

In the university where I teach, there have been recent concerns about the use of AI among students—specifically, the use of ChatGPT to produce their work. I understand and share the concerns; however, I cannot help but wonder if there is a way to understand if and how we can use AI tools ethically in education, considering academic integrity [personal communication, 05/16/23].

District policy makers and principals may assume that the tools provided to faculty will cover LLM-generated text and the first they hear about the doubts cast on students’ authorship bubbles up from the teachers to the administration. This underpins Cuban’s assertions that it makes sense to involve teachers early in the shaping of policy and implementation [19].

Another colleague of the author who is a community college Chemistry professor in California echoes Cuban’s call for administrators to pay attention to the variety of ways that LLMs impact their teaching and grading. She observed,

One thing that I’ve noticed recently that I think is really important to note is that institutions have to be sensitive to the fact that this is going to impact faculty differently and they need to support faculty, specifically in the way that they need. Whether you’re trying to resist it or embracing it, it feels manageable if you’re in science but not as a computer science faculty. But on the other side of the fence, we have all our English and Humanities, “writing heavy” faculty and a lot of them are really struggling because what they’re seeing is just endless essays that are clearly being written by AI [personal communication, 05/30/23].

During a work session with this author to experiment with prompting ChatGPT to rewrite the ending of The Scarlet Letter [25], a high school Social Studies and English Literature colleague in the Northeast United States evaluated the LLM-generated alternative ending as follows:

The chapter AI wrote for The Scarlet Letter is not at all "good," as far as literary writing goes, and is more or less another (inferior) way of stating what had already happened internally to the characters. Good! That means great writers, at least for now, are safe from being overshadowed by AI! [personal communication, 06/27/23].

The colleague’s observation about the alternative ending emphasizes the tacit knowledge owned by humans that is recognizable as “internal” to the character and not just a list of characters, setting, and plot. After a discussion about employing ChatGPT or Bard as a tool to revamp lesson plans, the high school colleague shared,

I’m not sure how I feel about embracing the large language models as a collaborator for lesson planning and student engagement. I would say, at this point, I don’t feel quite comfortable [personal communication, 06/27/23].

As this author’s colleagues grapple with existing lesson plans and text-based assessments, the university professor in Guatemala shared that a current assignment requires students to select a topic of interest, learn how to evaluate sources in a Google search, and write a draft of an essay to reflect on their findings for an initial grade of 15 points. What she found is that students were willing to lose 15 points, input the essay prompt into ChatGPT, and abdicate their personal viewpoints to rely on the LLM to write their reflection and garner the 85 points for the final assignment. After our conversation about strategies to embrace LLMs, she modified her syllabus and assignment to require students to create a prompt about a topic of interest, input and refine the inquiry into both ChatGPT and Bard, critically analyze the output of each LLM, identify biases, and prepare a video or audio recorded presentation, guided by a rubric listing required components, on a choice of platforms to exhibit their understanding and knowledge about the topic [personal communication, 06/13/23].

Having established the problem of a new flavor of students’ LLM plagiarism for teachers to overcome, the next sections of this chapter address how platforms powered by AI can assist educators in the reimagining of education to redefine their syllabi, lesson plans, and assessments in a shared teaching and learning domain with their students.

2.1 An unbalanced teacher workload

The Organization for Economic Co-operation and Development (OECD) is a group of 38 member countries established in 1961 to support world trade and economic progress [26]. The OECD publishes the Education at a Glance annual report to document comparisons of education outcomes and opportunities on an international scale [27]. It is compelling to note that among the findings of the 2022 report is the following statement about teachers’ tasks across the globe in primary and secondary levels:

Teachers spend an important share of their working hours on tasks other than teaching, such as preparing lessons and assessing examinations. In some countries, upper secondary teachers are expected to teach for less than one-third of their total working time, whereas in other countries, they are expected to teach for almost two-thirds of their working time [27].

To state the unbalanced workload of educators in a different way, teachers spend more than one to two-thirds of their valuable time preparing lessons, assessments, and grading than actually teaching curriculum to their students.

On the tertiary level, the OECD notes that academic staff vary by country in their instructional and research responsibilities. In most countries, the responsibilities of faculty are divided between research and teaching. The instructional workload for tertiary faculty centered on the switch and retention of online teaching and learning that required faculty to produce learning materials and digital exams workable in hybrid modalities [28].

The following sections analyze how AI can help teachers to streamline their workload and how educators can introduce students to the benefits and drawbacks of AI in their studies and preparation for the emergence of the singularity.

2.2 AI tools for educators

The Substitution, Augmentation, Modification, and Redefinition (SAMR) framework developed by Puentedura [29] helps educators evaluate software employed as tools to prepare and deliver course materials to students. For instance, traditional quiz software such as Quizalize is a substitute for paper-based quizzes [30]. Teachers who type their multiple-choice questions and answers into the platform or select a shared quiz on a topic that aligns with their curriculum can automate the assessment. The augmentation aspect of online quiz platforms aligns with the features of the software that support mastery learning by tracking individual student’s understanding of a topic aligned with objectives and standards. A quiz platform modifies an assessment by turning the teacher’s quiz into game-based learning in single, group, and class competitions with engaging graphics and leaderboards. It is in the redefinition of quiz software that aligns with AI and LLMs. Platforms such as Quizalize who upgraded their software to connect an application programming interface (API) with ChatGPT. Teachers can now input a prompt into the textbox provided by Quizalize to ask the LLM to provide quiz questions and answers.

Engineering the prompt to communicate directly with ChatGPT or through an API is where the redefinition continues. Learners who interact with technology platforms can become perplexed about why a program or platform is not working as they expected. As a programming teacher, this author shares the following maxim with all learners in her middle school through doctoral classes who complete coding assignments on a wide variety of programming languages and platforms:

“The program is doing exactly what you told it to do!”

The same maxim applies to LLM prompt engineering. Table 1 offers an example of three iterations of a prompt to create a multiple-choice quiz about LLM terminology. This author used her ChatGPT4 subscription to generate a quiz about four terms. Please observe that the first try did not include a specific instruction to create the quiz using the terms as associated with LLMs. The LLM generated quiz questions including multiple-choice answers among alternative definitions of the term.

Table 1.

Prompt engineering to create a quiz on four LLM terms [31].

During the second try, the prompt included a request for a quiz related to the terminology associated with LLMs. The third try input the refined prompt tested in ChatGPT directly into the author’s free Quizalize account. The API also provided a fifth quiz question about large language models. The dashboard, depicted with the platform’s permission in Appendix F, provide a view of the quiz, the options to assign for homework, an assessment, or to gamify using individual, group, or class competitions.

Using the ChatGPT and Quizalize platforms, educators can refine prompts to create a quiz from a bank of terms, edit the quiz to align with the content taught in their sections of a course, and share the assessment with colleagues. Circling back to the OECD finding that teachers spend an “important share” [27] of their workload creating lessons, assessments, and grading, it is crucial for district leaders, coaches, and professional developers to provide the guidance about AI tools such as Quizalize to produce and share resources to increase the amount of time that teachers can spend teaching the curriculum.

Having established the opportunity for educators to harness AI to revamp writing-heavy assignments and adopt tools such as Quizalize to reduce teachers’ planning and grading workload, this chapter now turns to our students who must learn how AI works to prepare them for dramatically different opportunities in the workforce.

2.3 Students as producers rather than consumers of AI

Our students are very familiar with AI in their daily lives besides LLMs that provide shortcuts to assignments. Artificial intelligence is an umbrella term that refers to an ever-expanding group of computer techniques. Examples of how our students already interact with AI include [32]:

• Speech recognition–Apple’s Siri and Amazon’s Alexa on phones or tablets.

• Recommendation systems–Netflix provides next viewing suggestions based on what you already watched.

• Anomaly detection–Banks will alert credit card holders of charges out of the ordinary with the customer’s location and purchasing history.

• Object detection–Google Lens will search for what you see, via the camera on your device or in stored pictures, including where to purchase a similar outfit in the image.

To begin to understand how an AI recommendation system works, students and teachers can create a sample database in an “unplugged” activity on paper to produce a recommendation. The following example lists the six records of data to answer the question, “What new streaming series would you recommend next for a new subscriber who just watched the Mandalorian?” (Table 2).

Subscriber	First series watched	Second series watched
Therese	The Marvelous Mrs. Maisel	Ted Lasso
Marianne	Marcel the Shell	Paddington Bear
Sal	The Mandalorian	Boba Fett
Nick	Boba Fett	Obi-Wan Kenobi
Gerard	Moon Knight	Wanda Vision
New subscriber: Gus	Obi-Wan Kenobi	?

Table 2.

What new series would you recommend for Gus and why?

The Netflix algorithm is an example of supervised learning with structured data where the system learns to recommend series based on labels such as comedy, action-adventure, space western to train itself using patterns to offer the next suggested series [32].

Google Lens [33] application is an example of unstructured data where AI is trained on images associated with labels. For instance, one can take or select a photo on their mobile or tablet devices to identify an object, shop for a similar item, translate text in 100 languages, solve math, history, biology, chemistry, and physics problems. It becomes evident as we review the scope of AI tools available to students that lesson plans and assessments must expand to applying and synthesizing math, social studies, science, world language knowledge rather than isolated answers which regurgitate facts, recites science facts, or parrots the conjugation of verbs. In a needs assessment conducted by this author, students long to apply their knowledge rather than “solve for X” [34].

The good news is that students can acquire hands-on experience to produce AI projects in the Python programming language through organizations such as AIClub [32]. Dr. Nisha Talagala is the co-founder and CEO of AIClub and AIClubPro. This author found Dr. Talagala’s work during her research about teaching and learning AI, Python, and the examination of datasets by students to implement real world, authentic AI applications. During her AIClub Student Research Symposium 2023, Dr. Talagala emphasized that research is, “an organized way to think about a problem” [35]. Her students in grades 7–12 brainstorm a topic of interest, search for data among the plentiful public datasets available on the internet, create a hypothesis, learn how to code a Python program that interfaces with AI algorithms to evaluate the data with a prototype of their application, and record results. The students can then present their research at the yearly AIClub symposium, submit their work to international competitions [36], publish their work in conference proceedings and peer-reviewed journals. In the case of the seventh-grade female student listed in Table 3, one can present the project at their school science fair [35]. Students analyze public datasets available on the internet such as Kaggle [36] and medical, satellite, and drone images. Then the students use this information to analyze the data and propose solutions through open-source AI tools and platforms to produce web or mobile applications.

Student research	Project	Details
Grade: High School Senior; Conference paper–International Conference on Computer Networks & Communications	A Machine Learning/Deep Learning Hybrid for Augmenting Teacher-Led Online Dance Education [37]	Recorded video clips Used MediaPipe [38] to create landmarks in videos Analyzed multiple frames in the videos AI algorithms used: Random Forest Classifier (RFC)–a commonly used machine learning algorithm [39] that combines the output of several decision trees that identify groups and relationships to predict outcomes [40]. K-Nearest Neighbor (KNN)–a supervised learning classifier to detect proximity about data points [41] Multilayer Perception (MLP)–a neural network data mining tool used to find unknown patterns in a database [42]
Grade: High School Sophomore; iOS application	Accessible Machine Learning Classifier for Rapid Emergency Self-Diagnostics (ALFRED)	Selected data from the DermNet database on Kaggle [36] using 7 injury labels and one non-injury group AI Algorithms used: Convolutional Neural Networks (CNN)–a supervised deep learning algorithm that is used for image recognition tasks [43].
Grade: High School	Artificial Intelligence Assisted Violin Performance Learning	Recorded correct and incorrect violin intonation, rhythm, and tempo from 13 Romantic Period pieces Used Python Librosa package to analyze music [44] Used Pyxeda Navigator, a machine learning development tool [45] and algorithms in Scikit Learn, an open-source Python package [46]. AI Algorithms used: Random Forest Classifier (RFC)–a commonly used machine learning algorithm [39] that combines the output of several decision trees that identify groups and relationships to predict outcomes [40]. K-Nearest Neighbor (KNN)–a supervised learning classifier to detect proximity about data points [41] Multilayer Perception (MLP)–a neural network data mining tool used to find unknown patterns in a database [42]
Grade: Seventh Grade	A Novel, AI-based Web application to Detect Benign vs. Malignant Oral Lesions from Intraoral Mobile Camera Images	Used images from an oral cancer on the Mendeley [47] Data image database.; AI Algorithms used: Convolutional Neural Networks (CNN)–a supervised deep learning algorithm that is used for image recognition tasks [43]. Produced a web application to allow users to upload a picture of suspected oral cancer to compare with verified malignant and benign tumors. Used Python on Google Colab [48] to program and maintain the web application.
Grade: High School	Hydroponic Lettuce: Navigating Nutrient Deficiencies	Downloaded the open-sourced dataset of 4 categories of lettuce quality from Kaggle [49] to compare to users’ image to report on phosphorous, potassium, nitrogen deficiencies and examples of healthy lettuce for comparison to user’s pictures uploaded to the application for comparison.; Used Streamlit website [50] to create a web application.; Used Microsoft Azure to train the AI model [51]; AI Algorithms used: Convolutional Neural Networks (CNN) – a supervised deep learning algorithm that is used for image recognition tasks [43].
Grade: High School	Hate Speech Detection; Research undergoing final peer-review in the JEI (Journal of Emerging Investigators [52])	Used a public data set of 60,000 hate comments on Twitter.; Employed Google’s Bidirectional Encoder Representations from Transformers (BERT) language models [53] and; Random Forest Classifier (RFC)–a commonly used machine learning algorithm [39] that combines the output of several decision trees that identify groups and relationships to predict outcomes [40]. K-Nearest Neighbor (KNN)–a supervised learning classifier to detect proximity about data points [41] Created a hate speech detection filter and a Discord bot [54] as a real-world application to detect hate speech.
Grade: High School	Smart surveillance system and identification of nocturnal animals.; Ongoing research.	Employed a hardware (Raspberry Pi [55], Camera, and LED) and software (machine learning integration and motion detection) to create an application to detect movement, take images, and identify nocturnal animals.; AI Algorithms used: Convolutional Neural Networks (CNN)–a supervised deep learning algorithm that is used for image recognition tasks [43].
Grade: High School	Real-time Waste Segregation at the Individual Level: The Smart Waste Sorter; Conference paper: Real-Time Image Processing and Deep Learning 2023 Conference [56]	Employed a hardware (Raspberry Pi [55], Camera, and LED) and software (machine learning integration and motion detection) to create an application to detect the category of waste such as compostable, recyclable, battery material, or waste to landfill before it is thrown away.
Grade: High School Junior	The Multi-Tier Artificial Intelligence Prediction Architecture: A Novel Approach to Intracranial Hemorrhage Detection; Currently submitting the research to a peer-reviewed journal.	AI Algorithms used: Random Forest Classifier (RFC)–a commonly used machine learning algorithm [39] that combines the output of several decision trees that identify groups and relationships to predict outcomes [40]. K-Nearest Neighbor (KNN)–a supervised learning classifier to detect proximity about data points [41] Multilayer Perception (MLP)–a neural network data mining tool used to find unknown patterns in a database [42]

Table 3.

Summary of student research presented during the AIClub Student Research Symposium [35].

This author attended the 2023 AIClub Symposium [35] to witness students’ presentations about their projects listed in Table 3. Most students noted that this was their first experience developing an AI project.

2.3.1 Real-world authentic connections to students’ AI projects

We are in the midst of the fourth “industrial revolution.” As educators, it is essential for us to keep pace with the emergence of AI as we prepare our students for the workforce in the Fourth Industrial Revolution of a worldwide economy. To operationalize the Fourth Industrial Revolution, the power necessary for production characterized the first (steam), second (electricity), and third (computer) electronics revolutions. Professor Klaus Schwab, the Founder and Executive chairman of the World Economic Forum defined the Fourth Industrial Revolution as an era that fuses physical devices, digital technology, and biological innovation [57].

Looking at the summary of student projects in Table 3, it is clear that the students of AIClub are well prepared to thrive in the Fourth Industrial Revolution fueled by the emerging AI entities. The online dance, violin, and hate-speech detection systems all address the current need for expanding the arts and eliminating bias through digital technology. The ALFRED rapid injury diagnosis application, oral cancer diagnosis web application, and intracranial hemorrhage detection application bring accessible biological and digital innovation to accelerate diagnosis and patient care. The hydroponic lettuce nutrient evaluation tool hits a trifecta of physical devices, digital technology, and biological innovation to sustain nutrient-rich food for humanity in the midst of real climate change. The nocturnal animal surveillance system and real-time waste identification sorter are physical devices that have authentic applications to the study of animals and the imperative to recycle, compost, and properly dispose of waste in the face of mounting landfills.

3. The international imperative to reimagine education

The general curriculum in most developed countries is still modeled after the 1894 published directives of the Harvard Committee of Ten [58, 59]. Appointed by the National Educational Association in 1891, and led by Charles W. Eliot, the President at Harvard University at that time, this committee gathered in subject-area “conferences” with the goal of establishing the uniformity of school programs. The global echoes of the Committee of Ten’s decisions are still evident in the structure of Elementary and Grammar School, grades one through eight aligned with ages six to 14, and Secondary School, grades nine through 12 aligned with ages 14 through 18. Subject disciplines established by the Committee of Ten were siloed into Latin, Greek, English, Modern Language, Mathematics, Science, Natural History (Earth Science), History, and Geography [60].

In his address to the Royal Society for the Arts in 2010, Sir Ken Robinson pondered the question, “How do we educate our children to take their place in the economies of the 21^st Century?” Robinson then led the viewers through an explanation grounded in the idea that school systems around the globe are trying to “…meet the future by what they did in the past” [61]. Robinson laments that the current structure of education mirrors the economics of the First Industrial Revolution fostering mass production capabilities of the conveyor belt to sequentially deliver students to the next grade organized by age.

The residual categorizations of the Committee of Ten established in 1894 aligned with age disregard the academic abilities of students who could study with more capable peers that align with Vygotsky’s Zone of Proximal Development (ZPD). ZPD takes a child’s cognitive age representative of what the student already mastered and what the student can produce through association with those around them. Vygotsky emphasized the social aspects of learning that extend beyond test results to challenge every child to grasp new concepts through collaborative learning among peers and their teachers [62].

Educators conducting research in Ontario, Canada noted that grade acceleration and the opportunity to learn alongside more capable peers aligns with neurodivergent “gifted” students, however the inclusive school study setting still aligned with grade-level steps and skips [63]. Dare and colleagues [63] define inclusive education in two categories, academic and social.

• Academic inclusion–all students access core curriculum. Students who already mastered content can gain access to advanced curricula that aligns with their cognitive age.

• Social inclusion–the school community accepts all students to promote belonging, and involvement in school life that promotes positive relationships among peers.

The inclusive schools studied by Dare and colleagues conveys the importance of positive school environment marked social inclusion and is echoed by the Finnish education system which enables sustainable well-being, is progressive, value-based, and student-centered. The national curriculum of Finland has been celebrated as a twenty first century global success story. The basic curriculum of Finland is national; however, municipalities do their own alignments, and schools decide on the details [64].

The key challenges faced by Finland in retooling their curriculum address the following needs from the school’s perspectives [64].

1. Developing school cultures to support curriculum values and goals to foster schools as real learning communities.

2. The student’s role is more active and inclusive.

3. The teacher’s role changes to reduce lecture-based content delivery.

4. Technology and digitalization of content including coding and digital learning platforms strongly implemented into the schools’ ecosystem.

5. Project-based and multidisciplinary learning modules aligned with transversal competencies at least once a year in all schools and all grades.

   1. The Seven Transversal Competencies of Finland’s National Curriculum are:

      1. Thinking and learning to learn

      2. Taking care of oneself and others, managing daily activities, and safety

      3. Cultural competence, interaction, and expression

      4. Multi-literacy

      5. ICT competence

      6. Competence for the world of work and entrepreneurship

      7. Participation and influence to build a sustainable future.

6. A shift toward assessment as learning to foster self-assessment and peer-assessment supported by formative feedback.

Finland’s national curriculum does mention grade-level assessments, but it is interesting to note that verbal assessments can be used in grades one through seven and numerical assessments, reflecting school decisions on the local level, to begin numerical assessments no later than eighth grade.

This author asserts that the most compelling transversal competency of Finland’s National Curriculum that aligns best with preparing our teachers and students to embrace the emergence of AI worldwide is number five,

“Project-based and multidisciplinary learning modules aligned with transversal competencies at least once a year in all schools and all grades [64].”

This chapter concludes by examining the transversal competencies as they align with the continue emergence of AI and the imperative for schools on a global scale to keep pace with, and harness AI.

4. Conclusion

This book chapter began with an imperative for education stakeholders to realize that we live in a pivotal time in our human existence where AI mathematical neural networks are on the precipice of outpacing human capabilities. We do not have the luxury of meeting the future by staying in a well-worn groove of we did in the past. Let us imagine that project-based, multidisciplinary learning modules aligned with transversal competencies in all schools and grades as the center of our school year rather than what we do when the standardized “teaching to the test” is over.

The first transversal competency, thinking and learning to Learn, and second transversal competency, “taking care of oneself and others, managing daily activities, and safety, lean into metacognition and self-regulation [64]”. Bruning, Schraw, and Norby assert that acquiring knowledge in only one part of cognitive growth. It is crucial that students develop metacognitive strategies to know about their own thinking, and to reflect on what they learned [65]. Authentic, project-based, multidisciplinary learning with the opportunity to have time to collaborate with classmates and teachers is crucial to learning [62].

The third transversal competency reflects cultural competence, interaction, and expression [64]. Ng, Van Dyne, and Ang define cultural intelligence (CQ) as an individual’s capabilities to manage and function effectively in culturally diverse settings [66]. The emergence of AI is a global concern, and our students must develop their CQ to heighten awareness and empathy for their international peers. Together, they can harness the emerging neural networks for the benefit of all cultures and geographic locations.

Wink asserts that all literate processes are socially constructed and are derived from social practices [67]. An awareness of the categories of multiliteracy [66] must be embraced by educators to foster students’ skills to achieve the fourth transversal competency through mastery of the following literacies:

• Academic literacy–the language of schools and universities.

• Functional literacy–languages of life including colloquial conversation.

• Workplace literacy–languages of our jobs

• Information literacy–languages of technology

• Constructive literacy–language we construct with the printed word.

• Emergent literacy–language constructed with text.

• Cultural literacy–language that help us think deeply about complex topics.

• Financial literacy–language to help students manage money responsibly.

• Reflective literacy–reading, writing, and reflecting–a multifaceted approach.

The Covid-19 pandemic brought the significant disparities within countries and economies to a heightened reality as global citizens struggled to connect with work, school, and life during the global lockdown [67]. Prior to the Covid-19 pandemic, scholars such as Hooft-Graafland noted that the countries of Costa Rica, Estonia, Finland, France, Greece, and Spain formally recognized Internet access as a human right [68]. The fifth transversal competency, ICT competence, is diminished in the age of AI emergence without access to online resources and skills mastered by the AIClub students’ research delineated in Table 3.

The sixth transversal competency for the world of work and entrepreneurship underscores the need for education stakeholders to become authentic leaders guiding our students who will be well-positioned to engage in the workplace dominated by AI. Scholars Jenson and Luthans categorize authentic leaders as confident, optimistic, hopeful, transparent, moral/ethical, resilient, and future oriented [69]. Through Vygotsky’s zone of proximal development [62], our students in the midst of authentic leaders can become future-oriented entrepreneurs and authentic leaders as well [70].

The seventh transversal competency, the participation and influence to build a sustainable future, is the most compelling in our discussion of emerging AI and casts a warning and imperative to reimagine education in the post-pandemic era. Scholars Charles Fadel and Jennifer Groff, writing in 2019, explain the acronym for our VUCA world as volatility, uncertainty, complexity, and ambiguity [71]. The intersections of the VUCA acronym aptly describe the emergence of a neural network singularity where we can no longer stagnate in the well-worn groove of the siloed curriculum established in 1893 by the Committee of Ten [60, 71].

Let us reimagine education in international settings to prepare our students to be producers of multidisciplinary projects to harness AI for the benefit of humanity.

Acknowledgments

This author expresses her gratitude to:

Dr. Nisha Talagala for inviting me to the 2023 AIClub Student Research Symposium to meet her outstanding student-researchers to learn about their impactful AI projects.

Conflict of interest

The author declares no conflict of interest.

Notes/thanks/other declarations

This author expresses her thanks to:

Charles Wiles, Co-CEO at Quizalize for allowing me to depict the Zzish Quizalize platform in this book chapter and Luis Pardo, Marketing Manager of Quizalize for coordinating our discussions about ChatGPT and quiz development in education and lesson planning.

Appendix A

DALL-E

In the gentle murmur of the server’s hum,

A dream awakened, something new begun.

An AI was birthed, with vision so bright,

OpenAI, a beacon in the academic night.

Through silicon neurons, ideas were born,

Tales of tomorrow, and of the past forlorn.

Then came DALL-E, bold and wise,

With a painter’s eye and a seer’s guise.

An image from a phrase, a dream from a word,

A world where the unheard could be heard.

Crafting images from the ether’s mist,

Rendering the unseen, with an artist’s wrist.

“An armchair in the shape of an avocado,” you say?

Done! Like a sunrise, it greets the day.

And in each creation, in each artful bend,

Is a silent promise, an unwritten end.

For every image is a story told,

In hues of ruby, emerald, and gold.

A dance of pixels, a symphony of light,

Crafted in the deep, quiet server’s night.

Not the sunset of the human mind,

But a new dawn, a different kind.

Not the end of art’s great reign,

But the start of a wonderful, shared domain.

So here’s to DALL-E, to OpenAI,

To the dreamers dreaming under the silicon sky.

To the future, so vivid, so impossibly grand,

Born from the marriage of the code and the hand.

To the silent whisper of a world unseen,

To the stunning beauty of the machine,

Remember, in every creation, in each profound story,

Lies the endless potential of shared glory.

Appendix B

[DALL-E].

Prompt: Create a multiple-choice quiz with four plausible options for each term, anthropomorphism, bias, hallucination, and temperature.

What does the term “anthropomorphism” refer to?

1. The practice of attributing human characteristics to non-human entities.

2. The cultural belief that nature is governed by spirits.

3. The scientific study of human evolution and development.

4. The process of converting human experiences into written form.

Which of the following best describes the term “bias?”

1. A person’s innate ability to understand and respond to the emotions of others.

2. The systematic error introduced into sampling or testing by selecting or encouraging one outcome or answer over others.

3. The psychological tendency to perceive oneself as being better than others.

4. The capacity of the human brain to store, retain, and recall information over time.

What is a “hallucination?”

1. The process of deep thinking and reasoning about abstract concepts.

2. The perception of something not present that is created by the mind, such as seeing, hearing, or feeling something that is not there.

3. The ability of the brain to compensate for physical injury by reorganizing and forming new connections.

4. The psychological state of being aware of and able to think about one’s own existence and experiences.

What is the term “temperature” used to represent?

1. The measurement of the amount of matter an object has.

2. The measure of the intensity or amount of heat energy in a substance.

3. The ability of an object or substance to transfer heat to another object or substance.

4. The measure of the average kinetic energy of particles in a substance or system.

[ChatGPT4]

Appendix C

Prompt:

Create a multiple-choice quiz that defines the term and then use the word bank of terms anthropomorphism, bias, hallucination, and temperature as possible answers. Please create the quiz in relation to large language models. Indicate the correct answer for each question.

Question 1:

This term refers to the attribution of human traits, emotions, or intentions to non-human entities, such as artificial intelligence or animals.

1. Bias

2. Anthropomorphism

3. Hallucination

4. Temperature

Correct Answer: B. Anthropomorphism.

Question 2:

This term refers to the measure of hotness or coldness expressed in terms of a specific scale. In the context of machine learning, it is used in techniques such as simulated annealing to help balance exploration and exploitation.

1. Anthropomorphism

2. Bias

3. Temperature

4. Hallucination

Correct Answer: C. Temperature.

Question 3:

In machine learning, this term refers to the tendency of a model to consistently learn the same wrong thing. It can be a result of overfitting, underfitting, or an issue in the training data itself.

1. Hallucination

2. Anthropomorphism

3. Bias

4. Temperature

Correct Answer: C. Bias.

Question 4:

In the context of language models, this term refers to generating outputs that aren’t based on input or training data—essentially, the model is “seeing” things that aren’t there.

1. Anthropomorphism

2. Hallucination

3. Temperature

4. Bias

Correct Answer: B. Hallucination

Appendix D

Figure D1.

AI–Large language model quiz [Zzish–quiz].

Appendix E

Figure E1.

Quizalize–Large language model quiz.

Appendix F

Figure F1.

Quizalize - Helpful tools for educators.