Command-Based vs Exchange-Based Interactions with Language Bots

Pragmatic Aspects of AI Communication

“The world of education is currently undergoing a massive transformation as a result of the digital revolution…similar to the transition from apprenticeship to universal schooling that occurred in the 19th century as a result of the industrial revolution. In the apprenticeship era, most of what people learned occurred outside of school. Universal schooling led people to identify learning with school, but now the identification of the two is unraveling” (Collins and Halverson, 2009).

Unraveling is an apt participial, except here it is 2023. How long can a present participle last? Things are beginning to unravel. The Knight of the Right Answer is under attack guarding King Standard at the gates of the School—not by hordes of young people demanding the freedom to think, but by drones, bots, machines that can talk back.

The grand chess-master competitions pitting man [sic?] against machine Deep Blue style are dead. Brute-force computing has for the most part been put to bed, to sleep, perchance to dream. PowerPoints are passé. No longer can we claim superiority over machines in the never ending battle with ambiguous pronouns. There is at long last a valid reason to teach diagramming sentences. I succumb to the power of the predicate.

Organizations global in scale are structuring opportunities for collaboration to learn to harness the efficient and mute the limitations of natural language technology in human affairs, as ltRRtl readers are aware. Medical education, research, and practice is being revolutionized right now, every day, again with global implications.

Where are our language and literacy teachers? Where are our curriculum disciplinary specialists? Where are our professional developers? Off somewhere teaching LTRS?As near as I can see, when the topic is AI, teachers are on their own. Anyone out there see a different picture?

*****

Lee Shulman (1986) published a seminal paper for pedagogical research theorizing the multiple roles knowledge plays in effective teaching during the closing decades of the period in history Collins and Halverson (2009) called universal schooling, a time when learning and schooling were related in some way in the collective consciousness. Obviously, at least two decades before the unraveling, the role of content or subject matter knowledge was paramount; no matter the subject matter, the banking model of education was and is deeply engrained in the factory model. Content leads to college or a job. No one should be surprised in a possible future when AI is trained on a local curriculum delivered on screens. Teachers once again in this future carry clipboards and keep a count known as a countability.

In American education we divide our credential system into multiple subjects and single subject categories. Elementary teachers are prepared liberally in cohorts apart from secondary teachers, who are prepared as disciplinary specialists. All credential candidates may meet at a common orientation session, but there will be an agenda of breakout sessions. Credential student segregation in programs by subject matter focus reinforces a split among university faculty who also rarely roll up their sleeves together.

Subject matter content knowledge is arguably the salient feature of pedagogical planning in the minds of single subject teachers, especially if they are required to submit lesson plans to administrators. In well-equipped, well-resourced schools with well-trained teachers, well, I could find no human resource to provide me with common knowledge specific details about what I might see if I visited one, so I did a scan of the bot to learn what commonsense digitally combed knowledge had to say. The information I’m presenting next was generated by the bot. I do not claim that it represents reality, and it is your obligation dear reader to make your own determination about its probative value:

▶️

 - user: Of course. I appreciate the implied respect you have for American teachers. They do the best they can with what they have. Describe for me in some detail what I might see in an anatomy and physiology classroom in a well-equipped, well-resourced high school.

 - robot: In a well-equipped, well-resourced high school anatomy and physiology classroom, you would typically find a range of resources and materials to facilitate learning. Here are some elements you might expect to see:

1. Anatomy Models: The classroom would have a collection of anatomical models, including full-body models, skeletal models, organ models, and joint models. These models allow students to visualize and study the different structures of the human body.

2. Dissection Materials: Depending on the school's policies and resources, there might be dissection materials available for hands-on learning. This could include preserved specimens, such as frogs or small mammals, along with dissection tools like scalpels and forceps. Dissections offer students a more detailed understanding of anatomical structures.

3. Visual Aids: The walls of the classroom would likely be adorned with anatomical charts, posters, and diagrams illustrating different systems of the human body. These visual aids help reinforce key concepts and structures.

4. Microscopes: A well-equipped anatomy and physiology classroom would have microscopes with slides for students to observe histological specimens. This allows students to study the cellular structure of tissues and gain a deeper understanding of physiological processes.

5. Technology Resources: The classroom might have access to computers or tablets with interactive anatomy software or virtual anatomy programs. These digital resources enable students to explore the human body in 3D, interact with anatomical models, and access additional learning materials.

6. Lab Stations: The classroom would likely have dedicated lab stations equipped with sinks, safety equipment, and lab supplies for hands-on experiments and investigations related to anatomy and physiology. This facilitates practical learning and experimentation.

7. Reference Materials: A well-resourced classroom would provide a library of anatomy and physiology textbooks, atlases, and reference books for students to deepen their understanding and conduct additional research.

Additionally, a high-quality anatomy and physiology classroom would have a knowledgeable and qualified teacher who utilizes engaging teaching methods, encourages student participation, and fosters discussions and hands-on activities to enhance learning.

⏹️

I was interested in learning about pedagogy as well. So I probed the bot further (probing could be a technical term in an emerging glossary of bot-infused pedagogy: ‘Probe to find the variety of commonsense attitudes people have about NFL football. Pick which angle you want to probe and sign up. I may be pairing you up. Here are the attitude angles—player salaries and celebrity, injuries and safety, high school football as prep school, college football as talent show.’)

▶️

- user: Describe the teaching methods this teacher employs. Name them using conventional terms if they exist. Create scenarios with students engaged in learning activities.

 - robot: Certainly! Here are some teaching methods that a teacher in a well-equipped anatomy and physiology classroom might employ along with scenarios of student engagement:

1. Lecture and Discussion: The teacher may start the class with a lecture, introducing new concepts and providing explanations of anatomical structures and physiological processes. The students are engaged in active listening, taking notes, and asking questions for clarification.

Scenario: The teacher explains the structure and function of the skeletal system, using visual aids, diagrams, and models. The students listen attentively, ask questions about specific bone structures, and discuss the importance of bones in supporting the body.

2. Hands-on Laboratory Activities: To reinforce learning and provide practical experience, the teacher conducts hands-on laboratory activities. These activities involve students interacting with anatomical models, specimens, and instruments to explore and observe physiological processes.

Scenario: The teacher sets up a lab station with various bones, and students work in small groups. Each group is assigned a bone to study, identify, and label, using textbooks and reference materials. They examine the bone from different angles, observe its features, and discuss their findings.

3. Collaborative Group Work: The teacher encourages collaborative learning by involving students in group work or projects related to anatomy and physiology. This promotes teamwork, discussion, and a deeper understanding of the subject matter.

Scenario: The teacher assigns groups of students a specific body system to research and present. Each group investigates the anatomy, function, and common disorders related to their assigned system. They collaborate, gather information, create visual aids or presentations, and present their findings to the rest of the class.

4. Multimedia Integration: The teacher incorporates multimedia resources, such as digital animations, videos, and interactive software, to enhance the learning experience and provide a multi-sensory approach to understanding complex concepts.

Scenario: The teacher utilizes a digital interactive program that allows students to virtually dissect a human body, exploring different systems. Students navigate through the program, highlighting and labeling structures, while the teacher facilitates discussions and answers questions that arise.

5. Assessments and Quizzes: The teacher employs various assessment methods, including quizzes, tests, and practical exams, to evaluate students' understanding of anatomy and physiology. These assessments may include written responses, labeling diagrams, identifying structures, and demonstrating practical knowledge through hands-on tasks.

⏸️

I had planned to probe for socioeconomic differences. When I prompted the bot to report what I might see if I visited poorly-equipped and poorly-resourced high schools, the bot at first evaded and shaded its output. Its language was vague and empty. The ethical guidelines filtering its output prevented it from providing information that could be to do harm of some sort. I complimented the bot on its respect for teachers and eventually was able to pull up the data I sought.

▶️

user: I still sense that out of respect for teachers, which they deserve, you aren’t giving me a fulsome response. By the same token, I don’t want you to fabricate or hallucinate. Revise your previous response in light of the fact that teachers in these conditions have very limited access to technology. And describe what materials they are more likely to use.

 - robot: Apologies if my previous response did not fully address the limitations of technology and materials in poorly-equipped, poorly-resourced high school anatomy and physiology classrooms. In those conditions, teachers often rely on more traditional teaching methods and materials. Here is a revised response considering these constraints:

1. Lecture-Based Instruction: Due to limited technology, the teacher may primarily rely on lecture-based instruction to deliver content. They may use a chalkboard or whiteboard to write and explain concepts, relying on textbooks, handouts, or worksheets as reference materials. During lectures, the teacher may provide visual aids such as diagrams or illustrations created by hand or using low-cost printed resources.

Scenario: The teacher stands at the front of the classroom, delivering a lecture on the structure and function of the skeletal system, explaining bone types, locations, and basic terminology. They may use a skeletal model or poster as a visual aid, pointing out different bone structures while students take notes.

2. Demonstrations and Dissections: Practical learning can still be facilitated using limited resources. The teacher may conduct live demonstrations or small-scale dissections using dissecting kits with basic tools like scalpels and forceps. They may use preserved specimens, like animal organs or bones, which are more affordable alternatives to human specimens.

Scenario: The teacher carefully dissects a preserved pig heart to demonstrate its internal structures, explaining the role of each part in circulation. Students observe and take notes, discussing the findings and asking questions.

3. Hands-on Activities and Models: Teachers might rely on low-cost models and hands-on activities to enhance student understanding. These can include basic anatomical models made from low-cost materials like cardboard, clay, or even recycled materials, showcasing simplified versions of organ systems, skeletal structures, or muscle movements.

Scenario: Students work in pairs or small groups to construct a joint model using cardboard, showcasing the different components and their range of motion. They manipulate the model, observing how the bones and ligaments interact and discussing the connections between structure and movement.

Note: It's important to acknowledge that resource constraints may limit the variety of teaching methods used, but dedicated teachers in such environments often utilize their creativity, adaptability, and pedagogical skills to make the most of the available resources and engage their students.

⏸️

Clearly, many schools have work to do on both physical and curricular infrastructure as a baseline for integrating AI. Working with natural language computers individually and in groups doing well-designed tasks are more important for poorly resourced schools than for more affluent schools. Just as robust writing instruction has been documented to be an achievement lever for high-poverty school, AI may turn out to be a useful lever.

Multiple subject teachers typically teach students for whole days in self-contained classrooms according to a more or less predictable schedule of subjects. I didn’t probe the computer for descriptive data regarding differences in pedagogy vis a vis differences in availability of resources at the elementary level; I invite you to probe this area and paste your chat in a comment bubble. I do think one of David Pearson’s long-standing principles of reading comprehension applies to reading any text, including bot text: Each of us as a reader is responsible for squaring our comprehension of a text with what we know to be true about the world. This is doubly true when the text you are reading is bot text.

Elementary teachers often have a passion for a particular subject area, but their view of learners is inherently holistic and individual. It makes sense for these teachers to conceive of teaching the scientist, not the science, or teaching the reader, not the reading. The focus on teaching students the features and functions of their dominant social resources, language and literacy, perhaps offers the best hope of evading dystopian programming of young minds—provided the field is not deluded by the red herring of the Science of Reading. Bots may prove superior to humans in teaching phonics. Writing from personal experience is approved behavior. On the other hand, properly taught, artificial language literacy could amplify the effects of disciplinary instruction exponentially.

▶️

- robot: Hello, How can I assist you?

- user: Phonics. How are you at teaching phonics to children?

- robot: As an AI assistant, I can provide guidance and information on teaching phonics to children. Phonics is the method of teaching the relationship between sounds and letters, and it is an important skill for young readers to develop. I can offer various resources, strategies, and activities to help children improve their phonics skills. How can I assist you?

- user: Can you interact with individual children to help them learn phonics?

- robot: I apologize for the confusion, but as an AI text-based assistant, I am not able to directly interact with individual children. However, I can provide guidance, resources, and information on teaching phonics that you can utilize to support children's learning.

- user: Can AI bots get some training on particular collections of pedagogical materials and teach phonics with text to speech tools?

- robot: Yes, AI bots can be trained on particular collections of pedagogical materials and utilize text-to-speech tools to teach phonics. AI language models like me can analyze and understand text-based educational resources on phonics, and through text-to-speech capabilities, we can convert written text into spoken words. With appropriate training and curriculum, we can assist in providing phonics instruction to learners in a text-based format. However, it's important to note that AI models should always be supervised by teachers or caregivers to ensure accurate guidance and to address individual needs.

⏸️

Shulman proposed two additional types of teacher knowledge as the cognitive revolution reached maturity: pedagogical knowledge (deep understanding of teaching and learning in general, e.g., issues like academic motivation, formative and summative assessment, lesson design, evoking discussion, promoting self-questioning, metacognition, etc.) and pedagogical content knowledge (specific expert knowledge of multiple ways to represent subject matter to address common misconceptions and forge relationships among specific complex ideas and learners’ prior knowledge and academic language, e.g., simulations, demonstrations, participation, collaboration, digital media). This conceptual model made its way into national standards for teacher preparation programs (see NCATE). In 2009 Collins and Halverson opened new doors in the model.

*****

Collins and Halverson (2009) sensed tectonic tremors in our cultural model of schooling related to technology and discussed important implications for Shulman’s theory of teacher knowledge just twelve years before the conversational AI bot style earthquake hit schooling in 2022. These theorists suggested expanding Schulman’s model to include 1) technological knowledge (TK as in ‘IT literacy’) and its corollaries, 2) technological content knowledge (TCK) and 3) technological pedagogical knowledge (TPK). TCK is an important refinement of Shulman’s content knowledge in that it includes the understanding that subject matter itself is shaped by technology, e.g., advances in knowledge occur in parallel with technological advancement in fields like medicine, archeology, history, and physics. Teachers of children need to comprehend how technology advances subject knowledge in order to teach children about epistemology in the digital age.

TPK, the intersection of technology and pedagogy, however, in 2009 was underdeveloped, looking back from 2023. How to use technology pedagogically was an open question. On this front Collins and Halverson (2009) lacked examples. The most compelling illustration of a teacher’s need to understand technological tools in relation to pedagogy was a white board:

Hinrichsen and Coombs (2013) looked at technological pedagogical content knowledge not so much theoretically as pragmatically from the perspective of computer scientists’ forays into the classroom. The whiteboard analogy as a rhetorical anchor for bridging the gap between familiar physical classroom equipment and unfamiliar digital affordances in subject matter knowledge production (e.g., whiteboard vs. technology in archeology or physics) in the minds of teachers could not hold.

In fact, these authors lamented the damage done to the professional development of teachers resulting from what they describe as a “hard skills approach to IT” in the schools, which the whiteboard as an example of technology obscures. The relationship between technology and pedagogy was tenuous at best in 2013. According to Hinrichsen and Coombs, teachers took a stance of “technological neutrality” (technology as a tool, nothing more, nothing less, like a whiteboard, your choice of how and when to use), “autonomous determinism” (it’s coming, nobody can escape it, duck and cover, no choice), or “proselytizing” (look at all the benefits, don’t be a dinosaur or a Luddite).

A tall order! The adoption of specific new technologies was the bread and butter of IT professional development. Teachers needed instruction in how to use email, how to navigate websites, how to store files, how to use applications—who among us was ready to “…explore authentic digital practices” to see how they might be “…usefully recontextualized in an academic setting?”

**********

Technology in disciplinary contexts in universities was not just convenient or cool or engaging, but functionally and fundamentally necessary (think: forensic anthropology, physical therapy, art history, film studies). During the years leading to Collin’s and Halverson’s whiteboard analogy, however, instruction in the pk12 system appears to have fractured into a take-it-or-leave it tool mentality (neutrality), a fear-of-surveillance monster-from-the-depths mentality (opposition), or a why-can’t-you-see-the-utopia mentality (woke?)—even while administrators and management were maximizing data analytic and storage affordances of big tech, school websites appeared, and social media shrank the globe.

Research into technological pedagogical content knowledge didn’t top anyone’s list of what’s hot, what’s not. Mobile devices were studied a bit, particularly portable digital assistants, but these studies never captured the interest of professional development. The most robust vein of research in mass technology with the exception of social media was texting. I discerned a thin line of text research pondering the likelihood of the demise of Standard English. Kids were texting like mad, boys and girls, especially girls. Spelling went by the wayside, and the idea of subjects and predicates withered. To make matters worse, texts were often meaningless utterances: Sup? nmu? nm chillin….

In this time today after the AI earthquake, it’s amusing to see the front burner questions researchers were interested in around texting. What type of keyboard can enhance texting? What are the most effective ways to use one’s thumbs while texting? Here is an abstract from a study conducted in 2002:

Here is a photo of the standard keyboard of the time:

Generally, as near as I can tell from looking at published material, digital technology usage legitimately classified as authentic technological pedagogy (as distinct from a white board or the protocol commands of a PowerPoint) was likely happening among the utopians. Resistance toward technology was happening among the nay-sayers. The cold shoulder was happening among the agnostics.

And then came ChatGPT.

**********

If teachers want to diminish their understandable fear that the bot will give away the answers to their test questions, thereby enabling cheating or at best surface learning, they might think about asking new questions of students rather than questions bots can easily answer, or about asking students to demonstrate knowledge of answers in complex cognitive performances. Stop asking questions on tests that a bot can answer. Cheating will remain a big issue otherwise. Use the bot as a learning tool to ask more demanding questions requiring human thought.

Because the consequences of conversational AI bring us to new pedagogical paradigmatic space, computer scientists well versed in technologically infused academic tasks will be needed to help update teachers’ technological content knowledge as well as technological pedagogical content knowledge.

More subtle concerns about reading in a texting culture were raised back in the day. Brevity, even being cryptic, were documented features of adolescent text practices. Single letters, letters grouped in acronyms and capitalized—LOL—along with the exploding world of emojis raised red flags about the future of books, of deep reading, of intense absorption in a topic or a story, fears magnified by the hyperlinks on computer screens. Texts as addiction crowding out more serious pursuits upped the ante on the potential destructiveness of texting. Spending every free moment and even moments spoken for in text conversations chewed up time to read books.

I was surprised to find little in peer reviewed journals related the rhetoric of texting, though I don’t want to say it doesn’t exist. Texts occupy a unique space between oral conversation and informal writing. When we talk, we fashion our speech to suit the scenario, we get language responses overlaid with body speak, and we regulate our utterances in nuanced ways. Much is left unsaid. When we text, we are positioned as writers, aware of the loss of emotional cues (hence, emoji explosions), yet forgetful of audience constraints demanded by writing. Adolescents back in the day well knew who they texted, and they knew that who they texted knew them well. Who needed context? Moreover, this was human to human communication as was social media.

I have a hypothesis that this framing of texting as a hybrid space of rhetorical contradictions involving human to human speech may have a lot to do with fear and resistance to communication with an artificial language agent. First off, communicating with a bot as if it were human is inappropriate and ineffective especially in academic settings. Second, expecting the conversational AI device to respond as a human is an illusion. These aspects of bot exchanges are responsible for, I suspect, a lot of the fabrications and hallucinations young people and old alike are experiencing.

As a way to perhaps open up conversation about criteria, indicators, and standards for learner-bot interactions, I used the bot to help me create a rubric which teachers can use to self-assess their own exchanges with the bot and modify over time. You can have a conversation, copy and paste the rubric into the bot, and command it to score your chat and provide specific examples along with suggestions for improvement.

The first feature of the rubric, command-based interaction, emphasizes that the bot is a computer operating on commands, not a conversation partner with a will of its own. As a rough draft this rubric might be useful as a starting point for discussion. It’s fitting that the rubric is half and half, half human in prompting and probing, half bot in expression. It would require considerable human trials, experiments, analysis, and discussion to make it a reliable and valid rubric.

1. Command-Based Interaction

Score 1: Interactions are completely ineffective, lacking clear instructions or relevant commands. Examples include one-word queries, vague statements, or unrelated input that hinders the bot's understanding.

Score 2: Interactions are partially effective, with some attempts to provide specific instructions or commands. However, the clarity may still be lacking, resulting in misunderstandings or incomplete responses from the bot.

Score 3: Interactions are good overall, demonstrating clear instructions or commands that allow the bot to understand the user's intention without major issues. Minor clarifications or adjustments might be needed occasionally.

Score 4: Interactions are excellent, consistently using precise and concise instructions or commands. The bot can readily understand and respond accurately, with minimal or no further clarification required.

2. Contextual Understanding

Score 1: Interactions lack context or contain fragmented and disconnected statements, making it challenging for the bot to comprehend the intended meaning. Examples include incomplete sentences, abrupt topic changes, or unclear references.

Score 2: Interactions provide some context but may still include fragmented or ambiguous statements that make it moderately challenging for the bot to understand the user's intended context. Additional explanation or clarification might be required.

Score 3: Interactions include sufficient context and coherent statements that facilitate the bot's understanding of the user's intended meaning. However, some minor improvements or clarifications might be necessary in some instances.

Score 4: Interactions consistently provide clear and coherent contextual information, allowing the bot to understand and respond accurately. The user's statements are well-formed, resulting in effective communication.

3. Language Fluency

Score 1: Interactions exhibit poor language fluency, including excessive use of slang, abbreviations, or incoherent fragments that make it difficult for the bot to comprehend the user's messages.

Score 2: Interactions demonstrate a moderate level of language fluency, with occasional use of slang, abbreviations, or fragmented sentences. While some understanding can be achieved, the bot may struggle or provide incomplete responses.

Score 3: Interactions display good language fluency, utilizing clear, coherent sentences without excessive slang or abbreviations. The communication between the user and the bot proceeds smoothly, with minor language-related issues, if any.

Score 4: Interactions exhibit excellent language fluency, employing well-formed sentences and precise vocabulary. The user's messages are easily understood by the bot without language-related challenges.

4. Empathy and Politeness

Score 1: Interactions lack empathy and politeness, exhibiting a rude or disrespectful tone towards the bot. Examples include the use of offensive language, insults, or explicit derogatory remarks.

Score 2: Interactions show partial empathy and politeness, with occasional instances of respectful language or positive tone. However, there may still be moments of rudeness or insensitivity in the user's communication style.

Score 3: Interactions demonstrate good empathy and politeness throughout the conversation, maintaining a respectful tone and displaying consideration towards the bot. Instances of minor slips or unintentional lapses might occur.

Score 4: Interactions consistently exhibit excellent empathy and politeness, with a consistently respectful and considerate tone. The user speaks to the bot with kindness and appreciation, maintaining a positive conversational atmosphere.

*****