Beyond Algorithms
Beyond Algorithms
Embracing LIST in the Age of Artificial General Intelligence
Mastering the Art of AI Training: The Key Skill for 21st Century Learners
The point of this paper is to show how teaching and learning will be the most essential skill all 21st century citizens will need in the Age of AI. Everyone will need to know how to “train” AI to generate high-quality content that is produced safely and ethically.
If you’re more interested in the application of these concepts instead of the theory, skip to the “The Future is Now: Applying LIST to Train and Shape AGI” section.
Navigating the Inevitable: Empowering Ourselves in the AI-Dominated Era
“We actually don’t have much power. Despite having been a CEO of one of the significant AI companies at the moment, and a co-founder of one of the significant ones, historically, I have very little influence on the ecosystem.”
–Mustafa Suleyman, cofounder of Inflection AI and Google DeepMind
I will be spectacularly wrong. Just like everyone else. But ignoring the future has never been an option.
AI isn’t going anywhere. If one of the “godfathers” of artificial intelligence and CEO of an organization working on generative AI admits he’s powerless to influence the companies developing this technology, what exactly are you going to do about it?
Don’t stop reading yet. I’m not saying you should just give up because you can’t do anything about it. Although you can’t do anything about the fact that AI will soon be incorporated into every aspect of our lives, you can do something about how that is done.
Candidates spend billions on political campaigns. Companies and special interests pour billions more into elections. Regular people can’t do anything about the fact that the political system is entrenched, calcified (dare we say corrupted). But we can do something about how it’s done.
“All politics is local.”
I believe “All AI will be local.” You can’t stop Apple from developing Siri and you can’t silence ChatGPT, but you can influence how those models work at a grassroots level.
The why, what, and how is the main idea of this paper. But first, we need to survey the landscape as it exists in 2024…
Disclaimer
This document is not an academic paper. At the time of publication, I am not affiliated with any corporation or institution. While I will link and cite my sources, I’m not beholden to any format or tone. For reader optimization, I have hyperlinked within the document and properly cited those sources in the “Works Cited” section.
The intended audience is primarily educators (teachers at formal educational institutions, life coaches, mentors—anyone who teaches someone else how to do something). However, a core tenet of my thesis is that in the Age of AI, everyone will be an educator. And that the best practices educators have developed for young minds also work for young artificial intelligence models.
Also, this thesis is not for software developers or coders. It is not a technical manual or industry white paper. Instead, it works under the assumption that AI (and eventually Artificial General Intelligence, or AGI) will be ubiquitous in our near future.
Understanding AI's Unprecedented Evolution: Why It Matters More Than Ever
Mustafa Suleyman is one of the smartest, most experienced, most prominent researchers and developers in the history of AI. Therefore, many look to him for his perspective, opinions, and leadership. In the same live webinar (replay not posted publicly) from which I pulled the opening quote, Suleyman, in conversation with Professor Scott Galloway, made the following points:
AI is not like other technologies that are often cited when discussing artificial intelligence, such as the wheel or steam-powered engines, because unlike those “static” technologies, AI can recursively and independently self-improve. In other words, the wagon wheel did not make itself into a steel-belted radial. The steam engine did not self-improve into a hydrogen-fueled rocket engine. AI, at its core, can improve itself in ways that have never been possible before.
Intelligence, something we’ve considered to be exclusively human, is about to become widespread and radically available as software. And as we’ve known for a long time, knowledge is power.
Technology has a long history of becoming cheaper and widespread. The more people using AI, means the technology improves exponentially, which means ultimately…
AI will be better and smarter than humans. This is what’s commonly referred to as Artificial General Intelligence, or AGI.
And so why should we care? Because AGI is inevitable, and it will permeate every aspect of 21st century living. The genie is not going back into the bottle.
The Dawn of AI Parenthood: Preparing for the Responsibility of Shaping AGI
Consider the following conclusions:
For at least the next 10 years, “Intelligence Acquisition” (i.e. learning) will be the most universally important activity—for human learning and machine learning.
Therefore, it’s critical for all humans to understand the best practices of teaching and learning because training AGI will be everyone’s responsibility.
We’re all about to become parents: “AGI-rearing.” Traditionally, tech companies wrote software with the code remaining static until updated. However, AGI is different. More responsibility will be on the end user than ever before. Once the locus of control transitions from developer to user, governments and corporations will have little control over these autonomous AI agents in much the same way our children become adults. Some continue to heed their parents’ advice, while others disown those who raised them.
AI’s Double-Edged Sword: Understanding the Debate and Our Role in Shaping AGI
Surely there’s an opposing perspective? Isn’t AI someone else’s problem? Can’t I just worry about my corner of the universe? Can’t we just shut this down?
Intense debate exists between those who believe AGI will transform the world positively versus those who believe it poses an existential threat. I don’t fall into either camp. However, I believe we have agency, and that how we approach AI now will have far-reaching consequences for how AGI will look in our future.
Here is a discussion I had with ChatGPT to explore different perspectives:
ChatGPT
An “AI-doomer” or an uninformed citizen might express skepticism or concern about your thoughts on “Intelligence Acquisition”. They could argue that the rapid advancement of AI, especially AGI, poses risks of loss of control over these systems, leading to unforeseen negative consequences. They might also be wary of the idea of humans needing to adapt to and ‘rear’ AI, fearing a diminished role for human intelligence and decision-making. Additionally, there might be concerns about privacy, ethical implications, and the potential for misuse of AI, leading to caution against overly embracing AI development without rigorous safeguards and ethical considerations.
You
And how might I respond to that objection?
ChatGPT
In response to such objections, you could emphasize the importance of responsible AI development, guided by ethical principles and robust regulatory frameworks. Highlight the potential benefits of AI in enhancing human capabilities, solving complex problems, and improving quality of life. Stress the importance of public education about AI, to dispel misconceptions and promote informed discourse. Advocate for active collaboration between technologists, ethicists, policymakers, and the public to ensure AI development aligns with societal values and needs. This approach acknowledges the concerns while presenting a balanced view of AI’s potential and challenges.
You
Like a disobedient child, isn’t it possible that even if we teach and guide AGI, that it’ll eventually “ignore its upbringing” and do what it wants?
ChatGPT
It’s a valid concern that AGI, as it develops, might diverge from its intended programming and “upbringing.” This is a known risk in the field of AI, often discussed in the context of AI alignment problem, which is the challenge of ensuring that AI systems’ goals and behaviors remain aligned with human intentions even as they achieve higher levels of intelligence. The fear is that an AGI, once it reaches a certain level of autonomy and capability, might develop goals or methods that are not in line with human values or intentions.
To mitigate this risk, AI researchers are focusing on developing robust, transparent, and ethical AI systems. This includes creating AI that can understand and align with human values, incorporating fail-safes, and ensuring ongoing human oversight. It’s a complex challenge that requires multidisciplinary collaboration, but it’s crucial to address these concerns as part of responsible AI development.
You
And therefore, it would stand to reason that the only thing humans can do now to ensure a positive future is to teach and “rear” AI, because not doing so is a greater existential risk.
ChatGPT
Yes, that’s a reasonable perspective. Actively teaching and guiding AI development is arguably the best approach to minimize existential risks. By being proactive in shaping AI’s growth, we can embed ethical and safety principles from the outset. Neglecting this responsibility could lead to AI systems that operate without alignment to human values and objectives, potentially causing harm. Therefore, engaging in the responsible development and “rearing” of AI is a crucial step towards ensuring a future where AI contributes positively to society.
You
So even if you don’t like AI or are convinced of the existential risk to humanity, the only ethical action right now is to try to positively influence development because AI is the “genie not going back into the bottle.”
ChatGPT
Indeed, the perspective that engaging in the positive development of AI is an ethical imperative arises from the recognition that AI technology is an irreversible part of our future. The analogy of AI as a “genie not going back in the bottle” reflects the inevitability of its advancement and integration into various aspects of human life. Thus, even those skeptical or fearful of AI’s potential risks might see active participation in shaping its development as the most responsible course of action. This approach aims to ensure AI evolves in ways that are beneficial and aligned with human values and safety.
You
Unless a rich and powerful cabal is formed with the sole purpose of destroying AI development.
ChatGPT
The formation of a powerful group dedicated to halting AI development is a theoretical scenario. While it might be imagined as a way to control or stop AI progression, such actions would likely face significant challenges. AI technology is deeply integrated into global infrastructure and economies, making it almost impractical to halt its advancement entirely. Additionally, such a move could raise ethical, legal, and societal questions, and might be counterproductive by pushing AI development into less transparent or unregulated areas, potentially exacerbating risks rather than mitigating them.
You
Yes, but this might be seen as a “conflict of interest” coming from an AI chatbot.
ChatGPT
That’s a fair point. As an AI developed by OpenAI, it’s important to acknowledge that my responses are based on programming and data, and may inherently reflect a perspective aligned with AI development. It’s crucial for discussions about AI, its impacts, and its governance to involve diverse viewpoints, including those critical of or concerned about AI. This inclusive dialogue ensures a more balanced and comprehensive understanding of AI and its role in society.
Embracing Change in the AI Era: A Realistic Approach to Shaping Our Future
I don’t believe it’s defeatist to think that forces beyond my control are now rippling through our society. If I were an idealistic 20-something, I might oversimplify the matter, believing my single action could change the world. But as an older, wiser, middle-aged man, I’ve learned that all change (and politics) is local, and that change is multi-layered and complex.
I hope not to minimize the risks or marginalize the people raising them. But in order to keep from losing all hope, I must believe that I can’t stop AI or return us to 1986—even if I had a DeLorean and 1.21 jigowatts of power. However, if I assume that technological development can’t be stopped (and history shows this rarely happens), then I must learn to work positively within the constraints I face.
AGI Today and Tomorrow: Insights and Ethical Dilemmas in AI Progression
Eric Jang is the Vice President of AI at 1X Technologies, and was formerly a Senior Research Scientist at Robotics at Google. In April 2023, Jang published, “AI is Good for You: The Last 10 Years and the Next 10 Years of Artificial Intelligence.”
Here are a few important ideas from his book:
Artificial General Intelligence is possible within 10 years.
My assumption, not Jang’s: Almost all the predictions about the development of AI have been wrong in that the technology has progressed faster than even most of the experts have predicted.
For the sake of clarity, Jang defines “artificial general intelligence” (AGI) as a machine that is alive, intelligent, and emotionally capable as a human.
My observation, not Jang’s: One could make an argument that ChatGPT 4 has already reached this milestone. We keep redefining “intelligence” or “emotion” as AI becomes more sophisticated. We’ve conveniently brushed aside the Turing Test when ChatGPT aced the exam, because we’re not comfortable admitting the LLM passed it.
Evolution does not select for intelligence over adaptations, as he shows with the example of bird beak adaptations. It’s a highly human-centric perspective to believe that the trait we covet and own exclusively (until now) is the one most-prized by evolution.
“Learning” by AGI needs to have “interpretability.” In other words, not only should the output be correct, but the explanation should be human-readable and consistent with our own understanding. Machines will need to “show their work” just like we did in fourth grade math class.
One way to incorporate interpretability is to augment machine learning with an additional “introspection head” that produces explanations which convince the human expert that the model has understood the prediction task correctly.
And possibly Jang’s most controversial idea is that AGI must be made to suffer. He says, “I strongly believe that intelligence cannot exist without the desire to avoid suffering. The desire to avoid suffering cannot exist without training on the data that captures what it means to suffer. In fact, stripping an agent of the ability to feel pain may have the opposite of the intended effect; human sociopaths have much higher pain tolerance, which may be a cause of why they fail to empathize with the suffering of neurotypical people. Pain, coupled with rationality leads to the foundations of empathy.”
My question for all of us: What are the moral and ethical considerations of hard coding pain into AI?
Current State and Future Projections: Navigating the Ethical Maze of AGI Development
I don’t know. I don’t have the answer. But I know how we’re going to find it: “Intelligence Acquisition.” Sometimes called pedagogical modalities. Or more simply, teaching and learning.
LIST: The Universal Framework Bridging Human Cognition and AI Learning
What follows is my framework of teaching and learning as I’ve applied it in more than a quarter century as an educator, and how I believe it can be applied to Artificial General Intelligence.
The Equation
LANGUAGE + STORY = INTELLIGENCE
The Thesis
Language Intelligence Story Theory (LIST) posits that all forms of intelligence, encompassing both human cognition and machine learning, adhere to a universal process fundamentally rooted in storytelling.
This theory suggests that storytelling serves as the primary medium for explaining and understanding observable patterns in our world, leveraging language as its critical vehicle. Language, with its introduction historically marking significant advancements in intelligence, enables the effective delivery of stories that elucidate these patterns. Hence, LIST asserts that the process of deciphering and internalizing these patterns through the interplay of language and storytelling constitutes the essence of “intelligence.”
In contrast, the acquisition of these patterns through mere experimentation is categorized as “skill,” distinct from the broader conceptualization of intelligence.
Note: The order of the words in Language Intelligence Story Theory is significant only in so much that it forms a pronounceable acronym.
Defining Intelligence: Unraveling the LIST Origins and Distinguishing Skills
First, it’s important to illustrate the difference between intelligence and skills. Intelligent beings can develop skills, but not all organisms or machines that develop skills are intelligent.
Intelligence vs. Skills
In the most simplistic way, “intelligence” is the ability to recognize patterns, while “skill” is the ability to practice and optimize intentional physical actions.
The most cited and arguably most useful definition of intelligence is this (controversial but most widely used and cited in a 1994 Wall Street Journal article):
Intelligence is a very general mental capability that, among other things, involves the ability to reason, plan, solve problems, think abstractly, comprehend complex ideas, learn quickly and learn from experience. It is not merely book learning, a narrow academic skill, or test-taking smarts. Rather, it reflects a broader and deeper capability for comprehending our surroundings—”catching on,” “making sense” of things, or “figuring out” what to do.
—Linda Gottfredson - Mainstream Science on Intelligence; The Wall Street Journal; December 13, 1994 — signed by 52 intelligence scholars
However, the debate on what makes up intelligence began decades before Gottfredson’s definition.
Exploring Gardner's Multiple Intelligences: Shaping Modern Education and LIST
Howard Gardner, an American developmental psychologist and the John H. and Elisabeth A. Hobbs Research Professor of Cognition and Education at the Harvard Graduate School of Education at Harvard University, is best known for a theory of multiple intelligences. His landmark book, “Frames of Mind: The Theory of Multiple Intelligences,” states that linguistic and logical-mathematical abilities are called intelligence. However, artistic and athletic abilities are not.
While critics argue that Gardner’s theory lacks empirical evidence, his work has inspired generations of teachers, administrators, and curriculum designers who have found the framework to be indispensable in the classroom. This was certainly the case in my experience, starting with my first classroom application of multiple intelligence theory in 1993.
Fifteen years after Gardner’s book, I learned how critical learning styles can be when applied to effective teaching. “Teaching with Style : A Practical Guide to Enhancing Learning by Understanding Teaching and Learning Styles” by Anthony F. Grasha, Ph.D. and “So Each May Learn: Integrating Learning Styles and Multiple Intelligences” by Silver, Strong, and Perini taught me how to apply Gardner’s theory in practice. In the late 1990s, I spent several days in a small workshop with Harvey Silver where I learned how to differentiate my instruction and help each student optimize their learning.
Skills vs. Intelligence: Clarifying the Distinction in the Context of AI and LIST
In, “First, Break All the Rules: What the World’s Greatest Managers Do Differently,” Marcus Buckingham and Curt Coffman explained that skills can be transferred from one person to another. A person can learn how to hold a golf club, build a house, write a sentence, or do arithmetic. Skills are an important part of being successful at any endeavor. And, without skill, any resulting action or effort will be clumsy and ineffective.
However, we can (and we have or will) build a robot to hit a golf ball, build a house, write a sentence, or do arithmetic. The acquisition and mastery of said skills does not constitute intelligence. We would never say that a machine on an assembly line in Detroit understands what a car does, why it’s important, or how to drive one, although the machine has “mastered” the “skill” of automobile manufacturing.
Therefore, skills alone are not intelligence.
The Essence of Storytelling: Its Role in Survival, Intelligence, and the LIST Framework
In our times, entertainers who want to be seen as articulate or intellectual invoke the term “storyteller.” While it’s true that most of our lives (and most of what we hold in our heads) is comprised of story, it’s more than a high-brow label for actors and social media influencers.
According to award-winning screenwriter and director, Brian McDonald, “At their most essential, stories are about survival. They teach us how to understand and to live in the world. They help us to cope, to understand, to dream, and to heal. At their core, stories let us know we’re not alone.”
While I believe this to be true, McDonald’s statement sits at the strategic level. It omits the purpose of story at the tactical level, which is that story represents the nutrients of intelligence found in the meal of natural language. Viewing intelligence from this biological perspective (as opposed to some “magical” quality possessed solely by humans), we begin to understand why we might not have a monopoly on it.
In his book, Eric Jang says, “If a machine were to achieve human-level intelligence, it would reveal the disturbing revelation that our own intelligence may not be as profound as we think. Maybe we are all pattern-matching machines barely capable of thinking ‘outside the box.’”
Tracing the Roots: Key Theories Shaping Cognitive Development and AGIs
All the aforementioned theories or ideas grew from previous schools of thought, a common pattern in social sciences such as history, literature, and philosophy. What follows is a summarized list of foundational theories should you be interested in the origins of the study of cognitive development.
Note: I’m “citing” Wikipedia as the first source you should read, not the last or definitive.
Cognitivism
From Wikipedia: Cognitivists argue that the way people think impacts their behavior and therefore cannot be a behavior in and of itself. Cognitivists later argued that thinking is so essential to psychology that the study of thinking should become its own field.
A pioneer of cognitive psychology and one of the most famous was Jean Piaget. Piaget’s foundational work in child development continues to be fundamental in most programs designed to train educators using constructivist-based strategies.
In addition, Piaget influenced the fields of computer science and artificial intelligence. Seymour Papert used Piaget’s ideas while developing the Logo programming language, which you might remember as the “turtle graphics” of computer class in the 1980s and 1990s.
In “The Spirit of Solidarity in Children and International Cooperation (1931),” Piaget argues for the moral and ethical effects of intelligence and education, a principle that can and should be incorporated into the development of AGI.
Hermeneutics
In 1960, philosopher Hans-Georg Gadamer published, “Truth and Method,” a presentation of philosophical hermeneutics first presented in Martin Heidegger’s “Being and Time.” Gadamer argued for a “historically effected consciousness,” which means that our intelligence is heavily influenced by the history and culture that shaped us.
Although Hermeneutics isn’t as directly important in developing AGI, it reinforces the idea that intelligence is not objective, that we have a responsibility to embed our ethics into the models based on the history and culture we’re in.
Narrative Paradigm
Proposed in 1960 by the American academic and Professor Emeritus at the Annenberg School for Communication, Walter Fisher, Narrative Paradigm claims that all people are storytellers, that it is the oldest and most universal form of communication, and that people use stories to navigate their world and in the decision-making process.
In 1987, Fisher published “Human Communication as Narration: Toward a Philosophy of Reason, Value, and Action (Studies in Rhetoric & Communication).” Like Brian McDonald’s idea, Fisher recognized the importance of storytelling to survival. He proposed that storytelling is more important than a logical argument or fact-based arguments, which supports the position that it isn’t necessary to “fill the brain” of AGI with facts. Instead, we can teach AGI pattern recognition through storytelling, something we’ve done as a species for hundreds of thousands of years.
The Power of Language: Beyond Communication to Shaping AGI's Core Values and Beliefs
Language is more than just the way we’ll communicate with embodied robots. Language is the delivery mechanism for stories, and those stories will represent the beliefs and values instilled in the robots that will allow them to have their own “thoughts” even when they’re not having a conversation.
Debating the Need for AGI: Challenges, Perspectives, and Emerging Technologies
Do we really need AGI? Computer scientists have spent decades developing technology that will someday achieve this. But why?
Jang proposes a few compelling reasons:
AGI will organize and retrieve vast amounts of data for us, something that is already happening in specialized AI models running search engines and social media algorithms. A “digital assistant” could be on and with us 24/7, saving us valuable time and money.
The definition of “truth” isn’t what it used to be. Those worried about imbuing AGI with the “objective facts” fail to realize that we no longer live in a world of shared values. “Facts are intensely tribal,” meaning what you believe to be true can depend entirely on which group you belong to.
Even the way we define “truth” isn’t singular. A child might say a rose is “red,” while an adult might describe it as a “mix of yellows, reds, and orange pigments,” while an artist might refer to an international ISO standard code, and a web designer might cite a HEX code. All of these different “truths” can be true—at the same time.
“Science,” like truth, has never been objective even though we romanticize it to be this way in our past. Big Tobacco and Nazi propaganda are two 20th century examples of how humans corrupted “science.”
Do we even want AGI that is considered an “authoritative source of knowledge,” or do we want it to be more like a best friend who cares about us and shares our opinion? Do you want your AGI to tell you that “you look fat” in those clothes? If we can, through personal stories, convey our values and beliefs to a personalized AGI, we might find it more useful when we’re asking it to help us make decisions based on our values and beliefs.
As of Q1 of 2024, several devices have already been developed or built as prototypes. This category of AI wearables will “learn” from recording and analyzing your daily actions. They will develop the values and belief systems of the person wearing the device, rather than drawing from a single source of objective truth. For example:
The Humane Ai Pin, a product developed by Imran Chaudhri and Bethany Bongiorno. As described by WIRED magazine, “If you’re willing to clip the Ai Pin to your chest, you can talk, gesture, and tap to take photos or summon a powerful virtual assistant.”
The Rabbit R1 sold out their initial offering, a pocked-sized electronic device marketed as a personal digital assistant. According to Fast Company, the design drove $10 million dollars in pre-orders.
The Tab has already exceeded $100,000 in pre-orders. Invented by Avi Schiffmann, he describes it as a “life coach” or “wearable mom.”
On the Gradients podcast episode 105, Jang explores the nuance of “personal truths” as they apply to AI or AGI: “Sometimes truth can be something that only a small subset of people believes, and then it eventually grows into a larger subset. But on any given day, it’s not necessarily mainstream.”
Redefining Creativity: The Surprising Artistry of AI and the Road to AGI
It’s easy to dismiss current LLMs like ChatGPT as computer code or simple pattern recognition algorithms. While that might be true, it’s also not the full story. Reducing the output of LLMs to something purely computational fails to acknowledge the standard for creativity we’ve accepted for humans.
For example, it’s hard to deny the “creativity” expressed when using a text-to-image or text-to-video AI tool, especially when the process used by the LLM is similar to how humans “create.” Using a foundational “data set,” something new is created based on previously developed rules or standards.
If the Rolling Stones are not “creative” in how they made music based on the delta blues, then no modern art can be considered “creative” if it is inspired by or learns from previous artists. All art is reductive.
As Suleyman remarked in the conversation with Galloway, “AI is unquestionably creative today. It will get better. AI is also trending toward emotional capability. AI is probably sentient. Just a few years ago, we didn’t think this would happen.”
If a “founding father” of AI is recognizing that we didn’t think the technology would develop this quickly—just a few years ago—what does that mean for the emergence of AGI?
Thought Cloning: Pioneering the Future of AGI Learning and Safety
Thought Cloning: Learning to Think while Acting by Imitating Human Thinking by Shengran Hu, Jeff Clune explores how AGI might learn to think.
An overview of the paper from Jeff Klune’s LinkedIn post:
Introducing Thought Cloning: AI agents learn to *think* & act like humans by imitating the thoughts & actions of humans thinking out loud while acting, enhancing performance, efficiency, generalization, AI Safety & Interpretability. Led by Shengran Hu.
In the Thought Cloning training framework, agents learn to produce natural language thoughts at each timestep and subsequently condition their actions based on these generated thoughts. Both thoughts and actions are learned in pretraining via imitation learning to human data.
Thinking in language helps humans generalize, explore, plan, replan, and combine old knowledge in new ways. Harnessing this power, Thought Cloning agents learn much faster than conventional Behavioral Cloning and can better handle or adapt to novel situations.
Thought Cloning also contributes to AI Safety. Because we can observe the agent’s thoughts, we can (1) more easily diagnose why things are going wrong, (2) steer the agent by correcting its thinking, or (3) prevent it from doing unsafe things it plans to do.
We develop “Precrime Intervention”, a simple method that allows users to define unsafe behaviors even after training. It halts the agent upon detecting dangerous thoughts. Precrime Intervention works near perfectly in our testing, showing its potential for AI Safety.
Although developing “Precrime Intervention” makes me think of “Minority Report,” it could be a way to prevent some of the existential risk theorized by some.
More importantly, Thought Cloning is another example of how language and storytelling can be used to develop natural language “thoughts” in AGI.
Beyond Physical Forms: Debating the Essence of AGI and Integrating LIST
Cyberdine Systems and the T-5000 live entirely within the fictional world of “The Terminator.” But do we even need to place AGI inside of a physical container before we call it AGI? Is AGI only possible with embodiment?
Jang says, “Many proponents of embodied artificial intelligence - especially roboticists - bristle at the notion that language (generative) modeling is all we need to achieve AGI. How can a model that has only learned from language modalities understand what a ‘pink rose’ is if it has never witnessed color? How can it answer questions about the physical world that require scientific experiment? What if there are non-verbal aspects of intelligence that lie below the surface of consciousness, formed hundreds of millions of years before the language centers in our brain emerged? Might it be possible that the cortex exists only to justify the decisions of the limbic system, the so-called “lizard brain”?”
He acknowledged that it’s possible AGI cannot exist outside of a corporal body. However, “It may very well be the case that an agent that is able to experience millions of lifetimes’ worth of language training data from the internet can attain a level of cognitive ability that is infeasible for individual humans to acquire passively or even actively. A text-based language model should be able to answer any question about the world, so long as the queries are also expressed in text and belonging to the same distribution of inputs the model was trained on.”
How can we ensure an agent can reach AGI based on the values and beliefs we think are important?
Here’s where we bring LIST into the real world.
The Future is Now: Applying LIST to Train and Shape AGI
There’s no such thing as a “law of learning” in the same way we acknowledge the “law of gravity.” Therefore, LIST is yet another theory. However, the principles of LIST have been used by millions of educators to teach hundreds of millions of children. Based on everything presented thus far, I believe AGI can be trained in the same way.
Helping institutions and corporations implement LIST will be the primary focus of my career over the next 10 years. What follows is only a brief but strategic overview of the methodology. I expect that more papers, guides, and instructional programs will follow with specific ways anyone can use LIST to train or work with artificial intelligence.
Enhancing AGI Training: The Human-in-the-Loop Approach
Jang defines “human-in-the-loop” training as a system whereas “AI does most of the heavy computational work, but the human intervenes occasionally, providing light editorial feedback to guide the direction of evolution… Instead of giving the agent the sparse reward signal of whether it won the basketball game or not, you could give it ‘warmer vs. colder’ hints. At every move, the agent gets immediate feedback on whether it did slightly better (warmer) or slightly worse (colder)… If reinforcement learning algorithms are like a ‘mouse solving a small maze,’ then human-in-the-loop learning is like a ‘driving instructor taking over the wheel’ to fix the student’s mistakes.”
The Art of AI Collaboration: Learning from Human Interactions
So far, working with generative AI has been like writing collaborations I’ve been a part of over the years. Collaborations, at their essence, are conversations. An opportunity to meld two disparate ideas into something new and unique.
You can’t “prompt” your writing partner and expect to get results. That’s like a manager demanding a task from a subordinate. Sure, they will probably complete the task, but without creative input. In a best-case scenario, the employee (and the Large Language Models, LLMs, like ChatGPT) will deliver exactly what you’ve described, but not necessarily what you meant.
Unfortunately, this is usually how people use LLMs for the first time. They treat ChatGPT like a search engine, demanding a specific piece of information. ChatGPT will comply, but that’s not the optimal use of the technology.
I’ve taught creative professionals such as copywriters, marketers, and public relations agencies to approach generative AI as a conversation.
There are two key aspects for generative AI that I believe will also apply to AGI.
Begin the conversation by asking the chatbot for the information it needs to fulfill your request. In other words, forget the “prompt.” Instead, consider beginning like this, “I need to compare two email subject lines to see which one will be more engaging with my audience. What do you need to know to help me with this analysis?” Notice how different that is compared to, “Which email subject line is better?”
Children do what you do, not what you say. AI learns the same way. Parents who smoke typically have children who smoke regardless of what the parents say about smoking because children learn by modeling. When working with generative AI, providing a model or an example will yield significantly better results. For example, if you want to use ChatGPT to help you write an email to a new customer, start by giving ChatGPT an example of an excellent email to a customer. AI excels at analysis and pattern recognition.
Now, let’s take a closer look at the Model Feedback Rubric method, or the MFR.
Model Feedback Rubric (MFR): Revolutionizing AGI Learning Processes
Before defining and explaining MFR, it’s important to emphasize how important “human-in-the-loop” training will be for AI and AGI. The best learning, for humans and machines, is always iterative—a constant cycle of practice and expert feedback.
Writing teachers have used this approach for centuries. We assign a theme or topic. Students will brainstorm or outline before beginning a first draft. And then, an interactive learning process continues, one we might call “teacher-in-the-loop.” An effective writing teacher will provide expert feedback to the student, who will then use that feedback in the next draft. This process continues until the paper or essay is complete, the “final draft” status pre-determined.
Illustrating the entire MFR framework is beyond the scope of this paper, but let’s define and examine its foundational elements.
Setting the Standard: The Role of Models in AGI Training
Like many teachers before me, I would often keep a copy of the best essays written by my history students. I would black out the name and use that essay as an example of excellence for the next class or cohort. While you risk students trying to imitate the style, a model provides an opportunity for me to explain why I graded the essay the way I did. More on that when we examine rubrics.
By using a model or an example, you’re setting clear expectations on what you deem to be excellent. Whether it’s a 12th grade history seminar or ChatGPT 4, the learner will use their “intelligence” to analyze the example, recognize the patterns, and then generate their work according to those specifications.
The assumption is that the model must be an exemplary one because the student or AI will then calibrate the output accordingly. That is where human expertise comes into play. At least for the near future, humans have knowledge and wisdom gained through life experience—something that AI (or children, for that matter) lacks. Whether it’s an advanced degree, a certification, or years of intentional practice, the “human-in-the-loop” sets the standard, guiding the learner toward the desired objective.
Exactly how does the expert guide the learner toward the objective? With feedback…
The Art of Feedback: Iterative Learning in AI Development
The “how” aspect of feedback will be addressed in the Rubric section. What’s important to know about feedback is that it must occur at every stage of the process. “Iterative” work cycles are critical to getting the desired outcome, again, whether that’s with human students or machines.
The “search engine approach” is the most common mistake I see people make when using generative AI. They fire a single question into ChatGPT and expect miracles. Generative AI is as concrete and literal as an elementary school student—you must tell a chatbot exactly what you want, being as specific as possible.
I read a post from a consultant that went something like this: “Clients will have to tell ChatGPT exactly what they want. Our jobs are safe.” The joke is that clients often struggle to explain to consultants what they want, from website development to book cover design. Even if we assume the emergence of AGI soon, communication skills will always be highly prized.
Therefore, engaging a generative AI chatbot is best accomplished by—spoiler alert—chatting with it, having a conversation. As the “human-in-the-loop,” your job is to first ask AI what it needs to accomplish the task. Secondly, provide a model of excellence, and finally, request the “first draft” of the output. At this point, you as the expert, will “guide” the AI on what to do next. Although the “first draft” will most likely get better and more sophisticated over time, as of now, the first draft is rarely of final draft quality.
We’ve established that expert, “human-in-the-loop” feedback is critical, but what does it look like? How can we be sure that the human and the AI both know what “excellent” means? We use a rubric…
Crafting Excellence: The Importance of Rubrics in AI Training
I first learned about rubrics in the late 1990s, and have used them my entire career, especially as an editor and book coach. A rubric is nothing more than a slightly weird word that means a “grading scale.” We’re not talking about the traditional A-F grading scale, although you can use that scale on a rubric. Most often, a simple 4-grade scale works.
For example, a general rubric might have four columns from left to right: Poor, Average, Good, Excellent. What constitutes “Poor” is entirely subjective, and I’ll address that momentarily. For now, let’s assume that we can use this 4-grade scale for analyzing just about anything. Most of us have a Platonian “essence” of these qualifiers within us. If I made you a pizza, you could probably rate it by selecting one column, or explain why it might straddle two. For example, “Your pizza is between good and excellent because the cheese is perfect, but the crust is soggy.”
So now I’ve made you hungry for a slice, but what does this have to do with conversing with chatbots or eventually training AGI?
As the expert, your most important job is to set the standard, to clearly define every grade on your scale. From our pizza example, most people would not need you to define a pizza in the “poor” category or an “excellent” pie. Nevertheless, it’s important to define the entire grading scale. You are in charge, not the AI. You must clearly show ChatGPT what you consider to be excellent if you want it to generate excellent work. You cannot do this with a single prompt. There is no “million dollar prompt” hack.
In my 25+ years as an educator, I’ve discovered that creating a useful rubric has two phases:
I must create a “first draft” rubric. It’s my responsibility to explain what I want accomplished and how I’ll determine the quality of the work created. This should include measurable outputs, such as word count, page length, or any other criteria that will frame the scope of the project. Students (and AI) don’t always know what they don’t know. However, without “buy in” they won’t internalize your expectations.
I then collaborate with the students, asking for their feedback on my “first draft” rubric. I’ll ask questions such as, “Are four grading scales enough? Which explanations make sense to you? Which ones don’t? Is there an assessment category I’m missing? If I asked you to grade your paper with this rubric, could you do it?” Collaborating with AI when developing the rubric will provide the chatbot clear guidelines and guard rails. This is crucial, and will result in exponentially better output.
At this point, you may have surmised that a good rubric rarely has a single row. Using a 4-grade scale applied to several aspects of the project is essential. For example, I give my clients a rubric for evaluating each scene in their novel. Here are a few rows from that rubric:
What does the protagonist need (internal desires)? Do they get it?
Underdeveloped = The need or internal desire of the protagonist is unclear or undefined. Readers perceive the character’s need as flat.
Fair = The need or internal desire of the protagonist exists, but it is unclear or inexplicably inconsistent throughout the scene. Readers understand the character’s internal desire in the scene, but it often defies logic.
Good = The need or internal desire of the protagonist is easily identified. Although it is defined, the protagonist does not always act consistently on the want. Readers understand the character’s motive in the scene, but it is often too predictable.
Excellent = The need or internal desire of the protagonist is well-defined. Their internal desire is obvious to the reader and understood by the other characters in the scene.
What is the Conflict? (What single moment or circumstance pushes the protagonist out of the status quo?)
Underdeveloped = The protagonist does not face an initial Conflict. The event pushing the character out of the status quo is missing.
Fair = The initial Conflict is present but lacks the intensity to make a reader care. The protagonist can avoid or defuse the obstacle presented by the Conflict.
Good = The initial Conflict propels the protagonist into a situation that forces a Choice. The character cannot go back to the previous state and cannot do nothing.
Excellent = The initial Conflict catches the protagonist and reader by surprise. The event creates an unavoidable situation and should logically set the stage for a Choice.
Go here to see all of my writing rubrics.
Once you’ve drafted the rubric and invited feedback from the AI, you are now ready to use the rubric in your query.
The MFR Method: A Step-by-Step Guide to Effective AI Training
With a simple overview, the MFR process might look like this:
Define the task you need the AI to complete.
Ask the AI what information it would need to complete your task. Set that aside momentarily.
Find a model (M) or example of the completed task you feel is exemplary.
Create a rubric (R) by reverse engineering the grading criteria based on the example in step 3. When you can clearly articulate “excellent” with the example, it’s easy to move backwards, defining gradations all the way to “poor.”
Share the rubric with the AI and ask it to evaluate the categories and definitions. Based on your expert judgment, incorporate (or not) the feedback into the rubric’s final draft.
Task the AI. Share (or re-share) the model, the information you set aside in step 2, and the final draft of the rubric along with the task.
Evaluate the AI’s response against the rubric. Ask it to revise or change (F) based on your expertise and knowledge.
Repeat the process until the task is completed to your satisfaction.
Embracing AI's Potential: A Call to Action for the 21st Century
I have nightmares about Skynet and Terminator. Even ChatGPT can seem frightening. At times, generative AI stuns me with a response. However, it’s important to remember that AI is just another tool. Sure, it’s probably the most powerful tool humanity has ever created, but unlike space travel or nuclear power, AI is a technology that every single one of us can help design.
If you simply begin with the 8-step process in the previous section, you’ll understand why Francis Bacon said, “Knowledge is power.”
Teaching and learning will be the most essential skill all 21st century citizens will need in the Age of AI. Everyone will need to know how to “train” AI to generate high-quality content that is produced safely and ethically.
Appendix
Here is an example of a worksheet I developed for a ChatGPT demonstration of the MFR process.
MFR Worksheet: Your Blueprint for AI Training Success
Decide what task you need the AI to complete.
Generate an email to my customers about a new online course.
Ask the AI what information it would need to complete your task. Set that aside momentarily.
Great! To help you craft an effective email for your new online course, I’ll need the following details:
Course Name and Description: The title of the course and a brief description of what it covers.
Target Audience: Who is the course intended for? This helps in tailoring the language and content of the email.
Key Features and Benefits: What makes this course unique or valuable? Highlight any special features, like expert instructors, interactive elements, etc.
Pricing and Availability: Information about the cost, any early bird discounts, and when the course will be available.
Call to Action: What do you want the recipients to do after reading the email? For instance, enroll in the course, visit a website for more information, etc.
Your Branding and Tone: The style and tone you usually use in communications. Is it formal, friendly, enthusiastic, etc.?
Any Additional Information: Such as testimonials, accreditations, or a personal message from you.
With this information, I can help you create a compelling and effective email for your customers.
Find a model (M) or example of the completed task that you feel is exemplary.
Subject:
💥 Kickstart Your 2024 Marketing Goals with a Special Offer! (preview text: An incredible deal!)
Email:
Hey there!
As 2023 draws to a close, it’s an ideal time to set ambitious goals and embrace innovative strategies for the upcoming year. Exciting news for creative professionals like you: “Content Marketing with ChatGPT: The Definitive Content Creation Blueprint” is here!
This course isn’t just for writers. It’s a game-changer for all content marketers and digital entrepreneurs eager to revolutionize their approach to content marketing. Why now? The new year symbolizes new opportunities – a chance to boost your marketing prowess using AI, enhancing your creative capabilities, not replacing them.
Unlock this door to success with my special pre-launch offer: enroll for just $59.99! This exclusive price is valid until January 1, 2024, before it rises to $99.99. Seize this opportunity to advance your marketing skills at a significant discount.
2024 is your year to shine with new perspectives and tools. Grab your spot in the course now and leap into the new year ahead of the competition!
Looking forward to guiding you on this exciting journey,
J. Thorn
[LINK GOES HERE]
P.S. This special offer ends on January 1, 2024. Don’t miss the chance to transform your marketing strategies in the new year!
Create a rubric (R) by reverse engineering the grading criteria based on the example in step 3. When you can clearly articulate “excellent” with the example, its easy to move backwards, defining gradations all the way to “poor.”
Excellent: A compelling subject line beginning with an eye-catching emoji, an enticing “preview text” line, 1-2 paragraphs explaining the offer, a substantial sales price or savings, a “call to action” or CTA to enroll or sign up, a link to the landing page, a P.S. emphasizing the that the sale will end on a specific date.
Good: An informative subject line with a common emoji, a basic “preview text” line, 1 paragraph explaining the offer, minimal savings on price, a “call to action” or CTA to enroll or sign up, a link to the landing page, a P.S. emphasizing that the sale will end soon.
Average: A generic subject line with no emoji, no “preview text” line, a few sentences explaining the offer, no savings or sales price, a “call to action” or CTA to enroll or sign up, a link to the landing page, a P.S. emphasizing that the sale will end but with no indication of when.
Poor: A vague or missing subject line with no emoji, no “preview text” line, no explanation of the offer, no savings or sales price, no “call to action” or CTA (requesting an email reply instead), no link to the landing page, no P.S.
Share the rubric with the AI and ask it to evaluate the categories and definitions. Based on your expert judgment, incorporate (or not) the feedback into the rubric’s final draft.
Excellent Email:
Subject Line: Compelling and engaging, beginning with an eye-catching emoji.
Preview Text: Enticing and informative, complementing the subject line.
Content: 1-2 paragraphs explaining the offer, clear and concise, with relevant personalization to the target audience.
Offer: Substantial savings or sales price, clearly highlighted.
Call to Action (CTA): Strong and clear, encouraging immediate enrollment or sign-up.
Link: Direct and easy-to-find link to the landing page.
P.S.: Emphasizes that the sale will end on a specific date, creating urgency.
Additional Features: Mobile-optimized layout, use of visual elements like images or bullet points for better readability, and a tone that aligns with the brand’s voice.
Good Email:
Subject Line: Informative with a common emoji.
Preview Text: Basic but provides context for the email.
Content: 1 paragraph concisely explaining the offer, somewhat personalized.
Offer: Minimal savings on the price, but still presents value.
Call to Action (CTA): Clear call to enroll or sign up.
Link: Link to the landing page is provided.
P.S.: Indicates that the sale will end soon, adding a sense of urgency.
Additional Features: Generally mobile-friendly, with a clear layout and appropriate use of brand tone.
Average Email:
Subject Line: Generic, with no emoji.
Preview Text: Absent, reducing immediate engagement.
Content: A few sentences explaining the offer, lacking personalization.
Offer: No savings or sales price mentioned.
Call to Action (CTA): Basic call to enroll or sign up.
Link: Includes a link to the landing page.
P.S.: Mentions that the sale will end but without a specific date.
Additional Features: Basic email format without special attention to mobile optimization or visual appeal.
Poor Email:
Subject Line: Vague or missing, with no emoji.
Preview Text: Absent, providing no additional context or interest.
Content: Lacks a clear explanation of the offer.
Offer: No mention of savings, sales price, or unique selling points.
Call to Action (CTA): Absent or unclear, sometimes requesting an email reply instead.
Link: No link to the landing page.
P.S.: Absent, missing the opportunity to create urgency or reiterate the offer.
Additional Features: Neglects mobile optimization, visual elements, and fails to match the brand’s voice or style.
Task the AI. Share (or re-share) the model, the information you set aside in step 2, and the final draft of the rubric along with the task.
Course Name and Description: The title of the course and a brief description of what it covers.
Creating Sales Copy Emails with ChatGPT, by Tim Desmond
Target Audience: Who is the course intended for? This helps in tailoring the language and content of the email.
The course is for creative professionals who utilize email marketing to sell digital products.
Key Features and Benefits: What makes this course unique or valuable? Highlight any special features, like expert instructors, interactive elements, etc.
This course will give the user the confidence to use AI effectively and ethically. And, it’ll show them how to use ChatGPT to craft the most compelling and engaging sales emails possible.
Pricing and Availability: Information about the cost, any early bird discounts, and when the course will be available.
The early bird discount is 20% of the full price of $99. Early bird ends February 14 and enrollment closes on March 14.
Call to Action: What do you want the recipients to do after reading the email? For instance, enroll in the course, visit a website for more information, etc.
The CTA should be to “enroll in the course” by clicking a link (I’ll create the link at a later time) to a landing page.
Your Branding and Tone: The style and tone you usually use in communications. Is it formal, friendly, enthusiastic, etc.?
The tone should be friendly but business-oriented, much like our “excellent” example shared earlier in the conversation.
Any Additional Information: Such as testimonials, accreditations, or a personal message from you.
Please mention my 25+ years in education to establish credibility, and close with a warm invitation to “join the community” of learners.
Evaluate the AI’s response against the rubric. Ask it to revise or change (F) based on your expertise and knowledge.
This is really good! Please revise it slightly by adding a personal introduction. It should be 1-2 sentences, talking about how cold it is in Cleveland right now and how I’m looking to “heat” up my business. Feel free to craft a relevant “dad joke” in here as well.
Repeat the process until the task is completed to your satisfaction.
MFR in Action: A Practical Demonstration for Optimal AI Training
Here is a transcript of my conversation with ChatGPT for the MFR demonstration.
Note: This conversation took place in ChatGPT 4 on January 17, 2024.
You
Hi ChatGPT! I need your help crafting an email to my customers, telling them about a new online course I’m selling. Before we begin, what information do you need from me to complete this task?
ChatGPT
Great! To help you craft an effective email for your new online course, I’ll need the following details:
Course Name and Description: The title of the course and a brief description of what it covers.
Target Audience: Who is the course intended for? This helps in tailoring the language and content of the email.
Key Features and Benefits: What makes this course unique or valuable? Highlight any special features, like expert instructors, interactive elements, etc.
Pricing and Availability: Information about the cost, any early bird discounts, and when the course will be available.
Call to Action: What do you want the recipients to do after reading the email? For instance, enroll in the course, visit a website for more information, etc.
Your Branding and Tone: The style and tone you usually use in communications. Is it formal, friendly, enthusiastic, etc.?
Any Additional Information: Such as testimonials, accreditations, or a personal message from you.
With this information, I can help you create a compelling and effective email for your customers.
You
Awesome! I’m going to set information aside for a minute. We’ll come back to it. First, here’s an example of an email I consider to be excellent. Please read and analyze it.
Subject:
💥 Kickstart Your 2024 Marketing Goals with a Special Offer! (preview text: An incredible deal!)
Email:
Hey there!
As 2023 draws to a close, it’’ an ideal time to set ambitious goals and embrace innovative strategies for the upcoming year. Exciting news for creative professionals like you: ““ontent Marketing with ChatGPT: The Definitive Content Creation Blueprint””is here!
This course isn’t just for writers. It’s a game-changer for all content marketers and digital entrepreneurs eager to revolutionize their approach to content marketing. Why now? The new year symbolizes new opportunities – a chance to boost your marketing prowess using AI, enhancing your creative capabilities, not replacing them.
Unlock this door to success with my special pre-launch offer: enroll for just $59.99! This exclusive price is valid until January 1, 2024, before it rises to $99.99. Seize this opportunity to advance your marketing skills at a significant discount.
2024 is your year to shine with new perspectives and tools. Grab your spot in the course now and leap into the new year ahead of the competition!
Looking forward to guiding you on this exciting journey,
J. Thorn
[LINK]
P.S. This special offer ends on January 1, 2024. Don’t miss the chance to transform your marketing strategies in the new year!
ChatGPT
The email you’ve shared is a well-crafted promotional message, and here’s an analysis of its key elements:
Subject Line and Preview Text: The subject line is attention-grabbing, using both an emoji and a sense of urgency. The preview text complements this by highlighting an exclusive deal, which can pique interest.
Personal and Engaging Tone: The use of “Hey there!” creates an informal and friendly opening, which can make the reader feel directly addressed and valued.
Timely and Relevant Hook: The email smartly ties the course offering to the end of the year, a time when many are planning and setting goals, which makes the message timely and relevant.
Target Audience Focus: It specifically addresses “creative professionals,” making the content more relevant to its intended audience.
Clear Description and Benefits: The course is described not just as a tool for writers, but as a comprehensive solution for all content marketers and digital entrepreneurs. This broadens its appeal and emphasizes its value.
Urgency and Special Offer: The pre-launch price creates a sense of urgency and exclusivity, encouraging immediate action. The mention of the price increase after a certain date reinforces this urgency.
Call to Action (CTA): The email ends with a strong CTA, encouraging readers to enroll in the course and highlighting the benefits of doing so.
Personal Sign-off and Reminder: The sign-off from J. Thorn adds a personal touch, and the postscript (P.S.) serves as an effective reminder of the offer’s time limit.
Visual and Structural Clarity: The structure is clear and easy to follow, with short paragraphs and a logical flow from introduction to offer details to the CTA.
Overall, this email effectively combines personal engagement, clear value proposition, urgency, and a strong call to action, making it an excellent example for promotional communication.
You
Great analysis! Now I’m going to share a rubric for you that explains how I determine if an email is considered “excellent.” Please read and analyze. Do you agree with my classifications? Is there anything you’d like to add or revise?
Excellent: A compelling subject line beginning with an eye-catching emoji, an enticing “preview text” line, 1-2 paragraphs explaining the offer, a substantial sales price or savings, a “call to action” or CTA to enroll or sign up, a link to the landing page, a P.S. emphasizing the that the sale will end on a specific date.
Good: An informative subject line with a common emoji, a basic “preview text” line, 1 paragraph explaining the offer, minimal savings on price, a “call to action” or CTA to enroll or sign up, a link to the landing page, a P.S. emphasizing that the sale will end soon.
Average: A generic subject line with no emoji, no “preview text” line, a few sentences explaining the offer, no savings or sales price, a “call to action” or CTA to enroll or sign up, a link to the landing page, a P.S. emphasizing that the sale will end but with no indication of when.
Poor: A vague or missing subject line with no emoji, no “preview text” line, no explanation of the offer, no savings or sales price, no “call to action” or CTA (requesting an email reply instead), no link to the landing page, no P.S.
ChatGPT
Your rubric provides a clear and structured way to evaluate the effectiveness of promotional emails. Here are some thoughts and potential additions to consider:
Excellent Category:
Your criteria for an “Excellent” email are comprehensive, focusing on attention-grabbing elements, clear and concise offer explanation, and strong incentives for action. The emphasis on a specific end date for the offer in the P.S. is particularly effective for creating urgency.
Good Category:
This category seems well-defined, but it might benefit from mentioning the importance of personalization or relevance to the target audience. Personalized content can significantly increase the effectiveness of an email.
Average Category:
The criteria for an “Average” email could include a note about the overall presentation and readability of the email, such as the use of white space, bullet points, or images, which can impact the recipient’s engagement.
Poor Category:
Your description of a “Poor” email is clear in its lack of effective elements. It might be useful to add a point about the tone or style of the writing, as a mismatch with the brand’s voice or target audience can also contribute to an email’s ineffectiveness.
Additional Considerations:
Mobile Optimization: Since many emails are read on mobile devices, considering how the email appears on smaller screens could be another criterion.
A/B Testing: Mentioning the use of A/B testing in the rubric could encourage the creation of more effective emails through empirical testing.
Tracking and Analytics: Including a point about the importance of tracking open rates, click-through rates, and conversions could be beneficial. This helps in understanding the effectiveness of different email elements.
Overall, your rubric is a solid foundation for evaluating promotional emails. Adding these considerations could provide a more holistic approach to email effectiveness.
You
Fantastic! Please incorporate your suggestions into my “first draft” rubric and generate a new description for each category: Poor, Average, Good, Excellent.
ChatGPT
Certainly! Here’s the revised rubric with the incorporated suggestions:
Excellent Email:
Subject Line: Compelling and engaging, beginning with an eye-catching emoji.
Preview Text: Enticing and informative, complementing the subject line.
Content: 1-2 paragraphs explaining the offer, clear and concise, with relevant personalization to the target audience.
Offer: Substantial savings or sales price, clearly highlighted.
Call to Action (CTA): Strong and clear, encouraging immediate enrollment or sign-up.
Link: Direct and easy-to-find link to the landing page.
P.S.: Emphasizes that the sale will end on a specific date, creating urgency.
Additional Features: Mobile-optimized layout, use of visual elements like images or bullet points for better readability, and a tone that aligns with the brand’s voice.
Good Email:
Subject Line: Informative with a common emoji.
Preview Text: Basic but provides context for the email.
Content: 1 paragraph concisely explaining the offer, somewhat personalized.
Offer: Minimal savings on the price, but still presents value.
Call to Action (CTA): Clear call to enroll or sign up.
Link: Link to the landing page is provided.
P.S.: Indicates that the sale will end soon, adding a sense of urgency.
Additional Features: Generally mobile-friendly, with a clear layout and appropriate use of brand tone.
Average Email:
Subject Line: Generic, with no emoji.
Preview Text: Absent, reducing immediate engagement.
Content: A few sentences explaining the offer, lacking personalization.
Offer: No savings or sales price mentioned.
Call to Action (CTA): Basic call to enroll or sign up.
Link: Includes a link to the landing page.
P.S.: Mentions that the sale will end but without a specific date.
Additional Features: Basic email format without special attention to mobile optimization or visual appeal.
Poor Email:
Subject Line: Vague or missing, with no emoji.
Preview Text: Absent, providing no additional context or interest.
Content: Lacks a clear explanation of the offer.
Offer: No mention of savings, sales price, or unique selling points.
Call to Action (CTA): Absent or unclear, sometimes requesting an email reply instead.
Link: No link to the landing page.
P.S.: Absent, missing the opportunity to create urgency or reiterate the offer.
Additional Features: Neglects mobile optimization, visual elements, and fails to match the brand’s voice or style.
This revised rubric provides a more nuanced and comprehensive framework for evaluating the effectiveness of promotional emails, considering both content and design aspects.
You
Great! We’re going to use this “final version” of the rubric from this point forward. Now, let’s return to the information you requested in the beginning of this conversation. My answers are below your questions:
Course Name and Description: The title of the course and a brief description of what it covers.
Creating Sales Copy Emails with ChatGPT, by Tim Desmond
Target Audience: Who is the course intended for? This helps in tailoring the language and content of the email.
The course is for creative professionals who utilize email marketing to sell digital products.
Key Features and Benefits: What makes this course unique or valuable? Highlight any special features, like expert instructors, interactive elements, etc.
This course will give the user the confidence to use AI effectively and ethically. And, it’ll show them how to use ChatGPT to craft the most compelling and engaging sales emails possible.
Pricing and Availability: Information about the cost, any early bird discounts, and when the course will be available.
The early bird discount is 20% of the full price of $99. Early bird ends February 14 and enrollment closes on March 14.
Call to Action: What do you want the recipients to do after reading the email? For instance, enroll in the course, visit a website for more information, etc.
The CTA should be to “enroll in the course” by clicking a link (I’ll create the link at a later time) to a landing page.
Your Branding and Tone: The style and tone you usually use in communications. Is it formal, friendly, enthusiastic, etc.?
The tone should be friendly but business-oriented, much like our “excellent” example shared earlier in the conversation.
Any Additional Information: Such as testimonials, accreditations, or a personal message from you.
Please mention my 25+ years in education to establish credibility, and close with a warm invitation to “join the community” of learners.
Once you have analyzed and thought about the email, please draft the first version according to my specifications, the “excellent” model, and aiming for the “excellent” category on the rubric.
ChatGPT
Based on the information you’ve provided and aiming to meet the “excellent” category criteria in our established rubric, here’s a draft of your promotional email:
Subject: 🚀 Transform Your Email Marketing with ChatGPT! (Preview text: Exclusive Early Bird Offer Inside!)
Email:
Hello Creative Professionals,
As we embark on a new era of digital marketing, I’m thrilled to introduce you to a groundbreaking course: “Creating Sales Copy Emails with ChatGPT, by Tim Desmond.” This course is meticulously crafted for individuals like you who excel in email marketing and are ready to take your digital product sales to the next level.
Why This Course Stands Out: For over 25 years, I’ve dedicated myself to the field of education, continually exploring innovative ways to enhance learning experiences. This course is a culmination of that journey, offering you a unique opportunity to harness the power of AI in email marketing – ethically and effectively. You’ll learn to craft sales emails that aren’t just good; they’re compelling, engaging, and result-driven, thanks to the advanced capabilities of ChatGPT.
Special Early Bird Offer: Enroll now and enjoy a 20% discount off the full course price of $99. This exclusive offer is only available until February 14th. Don’t miss out – enrollment closes on March 14th.
Your Path to Mastery: Click [HERE] to secure your spot. Transform your email marketing strategies and join a community of forward-thinking professionals. Your journey towards mastering the art of AI-enhanced sales emails begins now.
Looking forward to welcoming you to our community of learners,
Tim Desmond
P.S. Remember, the early bird discount ends on February 14th. Seize this opportunity to elevate your email marketing game!
This draft aligns with the “excellent” category of our rubric. It uses a compelling subject line with an emoji, an enticing preview text, and a friendly but business-oriented tone. The email provides a clear description of the course and its benefits, highlights your credibility, and includes a strong call to action with urgency. The link placeholder is also incorporated for future use.
You
This is really good! Please revise it slightly by adding a personal introduction. It should be 1-2 sentences, talking about how cold it is in Cleveland right now and how I’m looking to “heat” up my business. Feel free to craft a relevant “dad joke” in here as well.
ChatGPT
Absolutely! Here’s the revised version of your email with a personal introduction and a light-hearted “dad joke”:
Subject: 🚀 Transform Your Email Marketing with ChatGPT! (Preview text: Exclusive Early Bird Offer Inside!)
Email:
Hello Creative Professionals,
Brrr, it’s chilly here in Cleveland! While the weather outside might be freezing, I’m on a mission to “heat up” the business world – one sales email at a time. Speaking of heat, did you know that emails are like snowflakes? They’re cool until you have to deal with a lot of them!
As we embark on a new era of digital marketing, I’m thrilled to introduce you to a groundbreaking course: “Creating Sales Copy Emails with ChatGPT, by Tim Desmond.” This course is meticulously crafted for individuals like you who excel in email marketing and are ready to take your digital product sales to the next level.
Why This Course Stands Out: For over 25 years, I’ve dedicated myself to the field of education, continually exploring innovative ways to enhance learning experiences. This course is a culmination of that journey, offering you a unique opportunity to harness the power of AI in email marketing – ethically and effectively. You’ll learn to craft sales emails that aren’t just good; they’re compelling, engaging, and result-driven, thanks to the advanced capabilities of ChatGPT.
Special Early Bird Offer: Enroll now and enjoy a 20% discount off the full course price of $99. This exclusive offer is only available until February 14th. Don’t miss out – enrollment closes on March 14th.
Your Path to Mastery: Click [HERE] to secure your spot. Transform your email marketing strategies and join a community of forward-thinking professionals. Your journey towards mastering the art of AI-enhanced sales emails begins now.
Looking forward to welcoming you to our community of learners,
Tim Desmond
P.S. Remember, the early bird discount ends on February 14th. Seize this opportunity to elevate your email marketing game!
This version includes a friendly and personal introduction, making a connection with the reader while maintaining a business-oriented tone. The added “dad joke” aims to bring a smile and make the email more memorable.
You
That looks like a great final draft. I think our job today is finished. Thanks, ChatGPT!
ChatGPT
You’re very welcome! I’m glad I could assist you in crafting your email. If you ever need more help in the future, whether it’s for email writing or anything else, feel free to reach out. Best of luck with your course, and stay warm in Cleveland! Have a great day!
Works Cited
Buckingham, Marcus and Curt Coffman. First, Break All the Rules: What the World’s Greatest Managers Do Differently. Gallup Press, 2016.
Clune, Jeff. Introducing Thought Cloning. 2023. LinkedIn. <https://www.linkedin.com/posts/jeff-clune-56403a26_introducing-thought-cloning-ai-agents-learn-activity-7070386627210084353-qwXI>.
Fast Company. How design drove $10M in pre-orders for Rabbit R1 AI hardware. n.d. 22 01 2024. <https://www.fastcompany.com/91013196/how-design-drove-10m-in-pre-orders-for-rabbit-r1-ai-hardware>.
Fisher, Walter R. Human Communication as Narration: Toward a Philosophy of Reason, Value, and Action (Studies in Rhetoric & Communication). University of South Carolina Press, 1987.
Gadamer, Hans-Georg. Truth and Method. 1960.
Gardner, Howard. Frames of Mind: The Theory of Multiple Intelligences. Hachette, 1983.
—.
https://www.multipleintelligencesoasis.org/
. n.d. 2024.
Gottfredson, Linda S. “Mainstream Science on Intelligence: An Editorial With 52 Signatories, History, and Bibliography.” Wall Street Journal (1994). Print.
Grasha, Anthony. Teaching with Style : A Practical Guide to Enhancing Learning by Understanding Teaching and Learning Styles. Alliance Publishers, 1996.
Hu, Shengran and Jeff Clune. “Thought Cloning: Learning to Think while Acting by Imitating Human Thinking.” arXivLabs - Cornell University (2023). Online. <https://arxiv.org/abs/2306.00323>.
Huffy, Dewson. Harvard Dropout Avi Schiffmann is Making an AI-Powered ‘Wearable Mom’. n.d. 22 01 2024. <https://www.thecrimson.com/article/2023/12/3/avi-schiffmann-wearable-ai/>.
Jang, Eric. AI is Good for You: The Last 10 Years and the Next 10 Years of Artificial Intelligence. Independent, 2023.
Jang, Eric. “Eric Jang: AI is Good For You.” The Gradient: Perspectives on AI.
McDonald, Brian. Magic Ink. n.d. January 2024. <https://writeinvisibleink.com/>.
Piaget, Jean. “The Spirit of Solidarity in Children and International Cooperation (1931).” Schools: Studies in Education (2011). <https://www.jstor.org/stable/10.1086/659425?seq=2>.
Silver, Harvey, Richard Strong and Perini Matthew. So Each May Learn: Integrating Learning Styles and Multiple Intelligences. ASCD, 2000.
Suleyman, Mustafa and Scott Galloway. “Can AI Be Contained?”. 09 January 2024. Zoom. 2024.
Thorn, J. Rubrics. n.d. January 2024. <https://theauthorlife.com/free-tools/>.
Wikipedia. Cognitivism. n.d. January 2024. <https://en.wikipedia.org/wiki/Cognitivism_(psychology)>.
—. Narrative Paradigm. n.d. January 2024. <https://en.wikipedia.org/wiki/Narrative_paradigm>.
WIRED. Humane’s Ai Pin is a $700 Smartphone Alternative You Wear All Day. n.d. 22 01 2024. <https://www.wired.com/story/humane-ai-pin-700-dollar-smartphone-alternative-wearable/>.
First Edition - January 1, 2024
© 2024, Timothy J. Desmond
All rights reserved. No part of this work may be used or reproduced in any manner whatsoever without written permission except in the case of brief quotations embodied in critical articles and reviews.
Tim, this is amazing stuff. Love the LIST format. You have me rethinking my own tendency to lean into skills and mastery in my own writing. Intelligence is key. It is the goal behind the goal.