Project Epic

This is a project that challenges how far a dog can learn. Along the way, the human learns too. Teaching a dog means studying the principles of learning and cognition, and because education demands consistent criteria and precise timing, the human has no choice but to keep examining their own behavior and reactions. In the end, this project is about finding out how far a human and a dog can grow together.

Proven facts, my own hypotheses, and questions that have no answer yet are kept apart as much as possible, and sourced content is linked so it can be checked.

1. A dog’s learning ability is underrated

Imagine aliens captured humans and kept them as ornamental pets, raised without a shared language. Could those humans have launched rockets, discovered quantum mechanics, or made films? It would have been hard. But that sight alone would not let you conclude that humans lack those abilities. Dog training and learning seem similar. Most training methods are not about what a dog could do, but about keeping the dog out of trouble and getting it to do fixed routines reliably. In human terms, it is kindergarten education for life. Not elementary school, middle school, university, graduate school — education ends at kindergarten, and nothing beyond it is ever taught.

Not everyone understands quantum mechanics, but with decades of study, some people come to understand it and some even create it. Dog education will of course take a long time too — but if we keep raising the difficulty step by step, couldn’t a dog do far more varied things than we currently assume? This project starts from that question.

2. Base assumption — no limits on ability

The things commonly believed about dogs — they can do this, they can’t do that — are in fact extremely hard to prove. Even with scientific tools like MRI, it is hard to tell whether something is beyond a dog that has received proper long-term education, or whether the ability exists but has never had the chance to show itself. If you took a thousand people who had only ever received kindergarten education and studied them, you might conclude that “humans cannot understand quantum mechanics,” “humans know nothing but instinct,” or “humans cannot understand calculus.” The absence of ability and the absence of education are indistinguishable if you only look at outcomes.

The brain of a Homo sapiens from tens of thousands of years ago is not much different from ours. A familiar thought experiment says a baby from that era, raised today, would do fine in modern society. The same brain ends up in completely different places depending on what it was taught and how. Of course, this is a story within one species. A dog is a different species, so what this analogy shows is not that a dog can do as much — only that without education, the limit of an ability cannot be judged. That is why this project takes as its base assumption that no advance limit is placed on a dog’s cognitive ability. This is not a claim that a dog can do anything — it is a research stance of not drawing the limit before confirming it in results.

3. Why no single answer can be proven

Results depend on the combination

Every dog differs in genetics, temperament, and the education and experience it has received so far. Even the same dog responds differently depending on who trains it. And the human variable is not only skill and timing — it includes the impression the person gives the dog: something you might call charisma, their attitude, their presence. More than any single method, it is the combination of dog, human, and methodology that determines the result. So training is less about finding the one method that works for everyone and more about finding the combination that fits the current dog and human. That said, there are basic principles most trainers agree on, like timing and consistency.

A success story does not prove the best method

Training methodologies usually offer two kinds of evidence: cases where raising a dog this way produced the desired result, and explanations that a certain learning theory predicts that result. A case only shows that such a combination is possible. It does not mean the method fits every dog or is the most efficient. As for theory, even when the theory itself is well verified in the lab, the training procedures derived from it have rarely been verified step by step in real training situations.

Even so, cases and theory are all we can lean on, because repeatable controlled experiments on living beings are impossible. To compare recall methods A and B, you cannot teach one dog method A, reset it to its pre-learning state, and then test B. Use a different dog and temperament and history change; change the human and skill and timing change. Nor can you deliberately create a life-threatening situation to validate a recall that lives depend on. So the fact that A succeeded must be kept separate from the claim that A is best. What the relevant research actually shows, and how far, is collected in the training science paper notes and the correction-versus-reward comparison studies.

Working with a living being involves values

Repairing a computer involves little in the way of values. Educating a dog is different. What counts as a good life, how much discomfort and pressure to allow, how much obedience to demand — these produce constraints. The decision not to use coercive methods, and the decision that a dog must obey its human, both come from values. So comparing methods is a question of fact and at the same time a question of values — and this project makes its value judgments explicit through the goals and constraints below.

4. What makes a dog happy

I think happiness is not a sustained state where everything is satisfied, but a state of continuous challenge, difficulty, and overcoming them to achieve. This is not a proven fact — it is my view, and the working hypothesis that shapes the entire design of this project.

Like Maslow’s hierarchy of needs, I think a dog too can climb higher and higher as the lower needs are met. I know Maslow’s hierarchy itself is not a well-supported theory empirically; what I borrow is not the strict stages but the direction — the more the lower levels are filled, the more the higher ones are wanted. The human hierarchy cannot be transplanted as is, but a dog’s version would look roughly like this ladder. A dog whose survival and stability are met wants relationship; a dog whose relationship is secure wants challenge and achievement; and above that sits growth — learning new things. That is why this project’s goals (Part 4) are set on learning and achievement, not on maintaining comfort.

5. Why it learns and obeys should shift

A hypothesis, and a direction, about what the reason a dog learns — and obeys even when it does not want to — should become.

• It can start with food, toys, and punishment.
• But over time, it moves toward learning because learning and accomplishing things is itself fun and fulfilling.
• And the reason it obeys even when it does not want to, even against other desires, should become the relationship with its human — not wanting to disappoint, and following as a leader.

Leader here does not mean a rank produced by coercion; it means someone whose judgment the dog follows because it trusts them. I know that whether dogs perceive human disappointment is itself contested — there is research (Horowitz 2009) showing the “guilty look” is a response to the owner’s cues, not guilt. So this item is not an observed fact but a hypothesis this project sets out to test. The direction itself has support, though: in a study that scanned the brains of awake dogs (Cook 2016), many dogs showed reward responses to a cue predicting praise as strong as to one predicting food. That praise and relationship can be real rewards is established; disappointment is the hypothesis beyond it. External rewards are the tool that starts the engine of learning; the destination is intrinsic motivation and the relationship. Whether this shift actually happens, and how far it goes, is what this project observes.

6. The way of learning climbs too

The way of learning follows the same climb: starting from luring and capturing, where the human directly prompts the behavior, to shaping, where the dog builds the behavior itself, then on to imitation learning and language-and-concept learning — moving ever higher is the goal. Imitation and language-and-concept learning share one rung: both transfer a great deal at once, through a demonstration or a sentence. Luring and shaping are both procedures of operant conditioning; the dividing line here is not conditioning versus something else, but who produces the behavior and how much transfers at once. The higher the method, the more one demonstration or one sentence can transfer. So the higher methods are exposed early and often, from puppyhood — even before they are mastered, so that the very existence of higher methods enters the dog’s repertoire.

7. Learning beyond conditioning — references

The hypotheses above stand on the evidence that learning outside conditioning is real: proof that some learning cannot be explained by classical and operant conditioning alone, and the systems that carried it into animal training.

Compared with human learning

Humans learn by conditioning too. Becoming afraid of dogs after a bite is classical conditioning; doing a praised behavior more often is operant conditioning. School, with its grades and praise, is full of reinforcers as well. But conditioning is not what carries the content. Infants, with no reward at all, statistically find word boundaries just by listening to speech; children learn by watching and copying adults; and the center of a lesson is teaching by explaining in words. Grades can make a student study, but they cannot put calculus into a head.

Set against dog training, the difference is stark. In human education, conditioning is only the base layer — yet typical dog training almost entirely begins and ends on that base layer. What this project attempts is to find out whether dog education can be lifted to the upper floors human education uses: imitation, and language and concepts.

• Insight — Köhler’s The Mentality of Apes (1925). Watching the chimpanzee Sultan combine boxes and sticks to solve problems, Köhler interpreted it as insight without accumulated trial and error. That interpretation is still debated.
• Latent learning — Tolman’s maze experiments (1930). Rats that merely wandered without reward still built cognitive maps.
• Observational learning — Bandura’s Bobo doll experiment (1961). Behavior is learned just by watching, without direct reinforcement.
• Language is not explained by reinforcement — Chomsky’s review of Skinner’s Verbal Behavior (1959).
• Lexigram combinations — the bonobo Kanzi (Savage-Rumbaugh). Carried out first-time combined instructions.
• Inference by exclusion — the Border Collie Rico (Kaminski, Science 2004). Hearing an unknown name, he fetched the never-seen object.
• 1,022 object names — the Border Collie Chaser (Pilley 2011). Understanding verb–noun combination sentences came in a follow-up study (2013).
• Category, color, and number concepts via Model/Rival — the African grey parrot Alex (Pepperberg).
• Concept training — Ken Ramirez. A curriculum that taught dogs match-to-sample, modifier cues, mimicry, and even counting.
• Imitation learning — Fugazza and Miklósi’s efficiency comparison of Do As I Do versus shaping (2014). Internal notes at Do As I Do.
• Naming and explaining — Kayce Cover’s SATS. Teaches through names and explanation instead of markers.
• Button language — the They Can Talk study. In one study (Bastos 2024), dogs responded appropriately to words like “play” and “outside” even when the speaker and context changed. Internal notes at FluentPet.

8. What we teach

If the hypotheses above hold, there are five strands to teach.

• Concepts — from basic to abstract, building up from single concepts to combinations.
• How to learn, itself. Not only conditioning like capturing and luring, but Do As I Do, concept training, and FluentPet — reaching imitation learning and learning through language in particular. Learning beyond conditioning.
• Two-way communication, made more precise and more varied. That is what makes teaching possible.
• Keeping motivation at its peak throughout the learning process.
• The ability to regulate one’s own emotions.

9. Principles for choosing methods

• However difficult, teach through imitation learning and language learning as much as possible.
• Teach language and concepts.
• Language is divided into commands and information — and taught that way.

Commands and information are separated because their functions differ. A command requests behavior, and whether it is performed has consequences. Information conveys states and facts without compelling any behavior. If the boundary blurs, either every word starts to sound like a command, or the dog learns that no word carries any obligation.

10. The language and concepts to teach

Language to teach

• Commands — words that request behavior. Sit, down, stand, come (recall), heel, wait, drop it, fetch, go in, get up, touch, copy me (Do it), stop.
• Information — words that convey states and facts. Markers like yes, no, good, done; names of objects, places, and people; time words like now and later; permission and prohibition like okay and not okay; announcements of what is coming, like food or a walk.

Concepts to teach — from basic to abstract, from one to combinations

1. My behavior creates outcomes — how learning itself works.
2. Objects, places, and people each have names.
3. Categories — names that group many things into one, like toy or ball.
4. Modifiers — distinctions like big and small, left and right.
5. Same and different — picking the one that matches what was shown.
6. Time — the difference between now and later.
7. Imitation — the meta-rule of “copy me.”
8. Inference by exclusion — an unknown name points to the never-seen object.
9. Rules — what is allowed, what is not, what requires permission first.
10. Must — some things are performed without exception.
11. Combination — putting learned words and concepts together to understand a first-time instruction.

The order is the difficulty. The early items can be taught by conditioning, but the later ones — inference by exclusion, the meta-rule, combination — require abilities that conditioning alone struggles to explain. For most of the steps, the references in Part 2 hold evidence that they are possible in animals. But some steps — time, rules, must — have no matching study I have found yet. Those steps rest not on existing evidence but on what this project has to confirm itself.

11. Current priorities

The whole curriculum cannot run at once, so effort currently goes in this order.

1. Teaching language — using words constantly in daily life to teach them, and having the dog express itself through FluentPet.
2. Learning methods — the combination of imitation learning and teaching through language.
3. Teaching a variety of rules and concepts.
4. Emotional control — returning quickly to a rational state even after getting excited.

12. Thought experiment — three ways to teach “go in the crate”

The same behavior can be taught at different levels.

1. Teach it by conditioning — build the going-in behavior with luring or shaping, then attach a cue once the behavior is stable. The most widely used approach.
2. Teach it by imitation — Do As I Do. The human goes into the crate first, then gives the “copy me” cue. The behavior is transferred whole, not in fragments.
3. Teach it by language and concepts — teach what a crate is (a name), what going in is (an action word), and what a command is (something that must be performed), each separately, then chain them together in words and have the dog carry it out.

Is the third actually possible? Among non-human animals, the bonobo Kanzi carried out first-time instructions given through lexigram combinations, and the Border Collie Chaser reportedly distinguished over 1,000 object names and verb–noun combination sentences. But how far this goes with one dog in an ordinary home has no answer yet. That is exactly the question this project sets out to answer.

A single success, though, will not be taken as proof of understanding. The test is designed to remove the other explanations one by one.

• First confirm that the names — crate, mat, box — and the actions — go in, get up, touch — are each discriminated on their own.
• Keep the human’s gaze, gestures, and position from giving away the answer; give only the words.
• Slip the never-practiced sentence in among sentences the dog already knows.
• Record the first-trial response separately from success after repetition — success after repetition may mean the sentence was newly learned in between.
• Check again with a never-seen crate, a different place, and a different speaker.

13. Project Epic’s methodology

First, set goals and constraints. Within them, try many methods and watch the results. Because no single answer can be proven scientifically, no method is declared the answer. What we are looking for is not a universal method for all dogs but the combination that actually works when this dog and this handler reach for the goal. Even if we cannot know whether it is optimal, the point is to find out how far we can go and to actually achieve what we believed possible.

Training theories overlap in places and conflict in others. We accept the shared principles but take no theory on faith. We separate what is observed fact from what is interpretation. Conflicting claims are tested under varied conditions after understanding their principles as fully as possible — compared in the same period when we can, in separate periods when applying both at once is impractical.

The standard of judgment is the goals and constraints. People differ in how they want to live with their dogs, so the means differ too. Even a question like “is it acceptable to hit a dog” is not settled as an abstract yes-or-no; we ask what it is meant to prevent, whether other methods exist, what damage remains, and whether it crosses the current constraints. We hold views now, but not as eternal answers — they change with new experience and evidence. The constraints themselves can be revised. But while they stand, they are kept without exception.

I also know this project is an N-of-1 record — one dog, one human. So whatever the results, they cannot show that this method is best for every dog, and no such claim will be made. Instead, the records are kept so that others can verify the process.

• Define success in advance — not “understood,” but which behavior must follow which cue under which conditions.
• Separate what is used for practice from what is used for testing.
• Record first-trial results separately from success after repetition.
• Reduce the hints the human gives without noticing — gaze, hands, body orientation.
• Record not just correct counts but time taken, avoidance, arousal, and recovery.
• Keep the failures too.

14. Goals

Everything starts from goals, and judgments change according to them. Below are the dog’s goals. The introduction said the human grows too, but the human’s growth is not managed as a separate goal: carrying out this process consistently is itself the human’s discipline, so it follows on its own while pursuing the dog’s goals.

1. A life of learning and growing — loving to learn and grow of its own accord, and enjoying doing things together with its human.
2. A life with both joy and hardship — experiencing not only joy but sorrow and adversity, achievement and growth, overcoming them itself and feeling the accomplishment. Not a life that is only ever fun, but a whole one.
3. Autonomy, and commands without exception — deciding and choosing most things itself and living with the consequences. But for the safety-critical recall, heel, and the immediate stop-and-drop down, there are no exceptions. Heel belongs here because before it is an obedience exercise, it is the means of staying safe in environments that demand control — roads, crowds.
4. Understanding higher concepts — finding out how far it can understand abstract concepts like object names, time, rules, and imitation. No limit on learning is set in advance.

15. Method

1. Study the education systems producing real results around the world, and the related research.
   • Agility — Susan Garrett, OneMind Dogs
   • Obedience — Fenzi
   • IGP — Training Without Conflict
   • Ring sports — NePoPo
   • Research-based — ELTE(Do As I Do, imitation and social learning), FluentPet(language communication)
   • Tricks — My Aussie Gal, Do More With Your Dog
   • Perspectives — D.I.N.G.O., Absolute Dogs, Control Unleashed
   • Other — nosework and autonomous problem solving, cooperative care
2. Understand what each methodology aims at and how it works, then directly test the parts relevant to this project’s goals.
3. Training is a performance skill that cannot be acquired through knowledge alone, so practice it directly and repeatedly.

16. Constraints

• By default, every training method stays on the table.
• Start with the methods carrying the least shock and burden.
• The current constraint is not to use methods that can cause irreversible physical or mental harm.
• Actions the dog may experience as heavy pressure are used only with the principles and criteria understood. The criteria must be clear enough that, if the dog could speak, it could explain what it was supposed to do and what it failed to do that changed the outcome. This does not mean the dog can actually speak — it is a check on whether the relationships between cue, expected behavior, and consequence have been taught consistently. The dog must never receive pressure merely because the human is angry.
• A stop rule — if avoidance, freezing, or failure to regain composure keeps recurring, it is not rationalized as part of growth. The task is made easier or the method is changed.

On pressure, the research pointing the other way is also known: dogs trained with heavy aversives showed more stress behaviors and higher post-training cortisol in an observational study (Vieira de Castro 2020), and judged ambiguous situations more pessimistically in another (Casey 2021). So even with pressure left on the table, whether behavior improved is not the only check — how the dog’s emotions and welfare change is checked separately.

17. Log

Dated entries record the central tasks and curriculum of that moment. The list below is not a list of completed items — it is the curriculum in progress as of that date.

2026-08-10

So far the focus has been on building a good relationship with humans and stacking positive experiences with the many things of the world. The current center of gravity is teaching many concepts, the ability to regulate impulses, trust and a right relationship with humans, and varied use of the senses and body.

• Concepts
  • How learning works
    1. Learns that behavior produces reward, and that getting what it wants means doing it correctly, quickly, and when asked. Some things are mandatory.
    2. Acquires multiple ways of learning itself — luring, capturing, shaping, learning by imitation.
    3. Knows that when the human demonstrates first, it can learn by copying.
    4. Builds the sense of agency that its own behavior creates outcomes.
  • Signals and language
    1. Distinguishes different verbal signals without seeing the behavior first.
    2. Knows that every object has a name.
    3. Understands signals that report the outcome of behavior — yes and no, good and bad, right and wrong.
    4. Learns the difference in time, like now versus later.
    5. Distinguishes wait signals for objects, places, and body positions.
  • Rules and roles
    1. Learns that play has rules too, and that following them is what makes playing together fun.
    2. Finishes even after failing, and separates the times it may do as it pleases from the times it may not.
    3. Distinguishes what is allowed, what is forbidden, and what requires permission first.
    4. Knows its role and what is not its role.
    5. Can ask for what it wants in the right way, and that request may be granted or refused.
  • Activities
    1. Finds specific scents and solves problems by nose.
    2. Broadens the ways of playing together — tug, retrieve, chase and catch, wrestling, copying, searching, quizzes.
• Emotions
  • Focus and switching
    1. Locks in focus quickly and settles arousal when needed.
    2. Regains focus despite distractions.
  • Joy
    1. Finds fun in shared education and play.
    2. Performs many cues willingly and joyfully, and does not crumble at failure.
  • Recovery
    1. Tries again after difficulty and frustration.
    2. Regains composure even after frightening, hard experiences.
• Relationship
  • Foundations
    1. The base is trusting and following the human. When trouble comes or things get hard, it should be able to ask the human for help.
    2. That does not mean total dependence. It keeps the independence to be fine alone.
    3. Time together is enjoyable for both.
  • Communication
    1. Can ask for what it wants and accepts the process of that request being approved or refused.
    2. Expands the kinds of communication and makes each signal clearer.
• Senses and body
  1. Moves its body in varied ways across varied environments.
  2. Finds many scents and experiences many stimuli — and learns not just to be stimulated, but to return to calm afterward.

18. Perspectives currently being studied

• Playing together
  • Training Without Conflict’s chase-and-catch and possession games. Not teaching fetch — the human plays first, in earnest, until the dog wants to steal the toy; in possession games the dog wins most of the time.
  • Fenzi’s Personal Play. Play with no food and no toys — the interaction with the human is the game. Get down to the dog’s level and find the style that dog loves.
• Relationship, communication
  • Susan Garrett’s Recallers. Grade the value of reinforcers and the distraction level of environments, and build up through games from low distraction. Recall, in this view, is one giant round of ItsYerChoice.
  • Fenzi’s engagement. From acclimation to the environment, up through human-initiated engagement, dog-initiated engagement, maintaining without reward, and finally work before reward — stage by stage.
  • FluentPet. Start with one word with a clear outcome, modeled by the human pressing the button, and never interpret button combinations before they are repeatedly confirmed.
  • Control Unleashed. Pattern games like 1-2-3 and Look At That, and the Off-Switch. Repeating the switch between arousal and calm until settling itself becomes a reinforced behavior.
  • Absolute Dogs. Diagnose not behaviors but concepts — arousal regulation, disengagement, optimism — and grow the weak axis through games.
• Training methods
  • D.I.N.G.O.’s fair training. The standard is whether the dog is given the choice to start, and it learns on its own from only the “correct” signal, with no signal for “wrong.”
  • Do As I Do. Build the meta-rule of copying — stay, demonstration, then the "Do it!" cue — on behaviors the dog already knows, then extend it to new ones.
  • OneMind Dogs. Handle the line the dog runs, not the obstacles. Dogs read body signals before voice, so every signal element must point at the same line.
• Drive, rules, daily life
  • NePoPo. Apply light pressure with the cue and switch it off the instant the correct behavior begins. Release of pressure and the final reward — two reinforcers stacked on one behavior.
  • Cooperative care’s Bucket Game. Start with looking at the bucket, then holding the gaze, being touched while looking, being touched with tools in hand. How quickly the dog offers the start-button behavior becomes the dashboard of its training state.

19. Common threads confirmed so far

1. A dog does what its innate temperament and what it has learned dictate in the situation it is given. Less a strictly proven observation than a working principle — seeing it this way is what has solved problems.
2. A dog’s behavior is the product of genetics and its life with humans. If the human acts consistently, quite a lot can be taught.
3. Timing and consistency decide the results.
4. Communication must happen in a way the dog can understand.
5. When failure repeats, before blaming the dog, the human changes the criteria, the environment, the reward, or the difficulty.

20. What I think

1. Actual results matter more than the good intentions in the process. Results here mean not just whether a command was performed, but the change in behavior, the emotional state, recovery after pressure, and the relationship with the human.
2. Once relationship and concepts are properly learned, the learning that follows becomes easy.
3. The more varied the experience, the more directions growth can take.
4. No limit is set in advance on a dog’s understanding, learning, and growth.
5. For most learning, the more initiative the dog has, the greater the possibilities — let it act to get what it wants and learn with joy. But domains that demand unconditional performance, like recall, are strictly controlled, and there pressure can be used. Start with low-burden, positive methods and let the results decide the next means.
6. Rather than removing all pressure, teach in advance how to handle it. The dog should be able to process pressure productively without falling apart. Raise difficulty step by step within what it can bear, and lower it when it cannot recover.
7. Do not filter out all of the world’s hardships. The world contains dangers, there are responsibilities to bear, and what must be done must be done.
8. The dog, too, has a role and responsibilities as a member of the household. This is not moral blame but the behavior and self-regulation that living together requires. Its current role is to learn hard, manage itself, take care of its body, and not become an unnecessary burden on the team.

These are the views held now. The goals stay clear, the constraints are kept, and the methods keep being revised by results.

21. Questions still without answers

1. If the dog performs a first-time word combination on the first try, did it really combine the meanings? How do we remove the explanations of object placement, human body language, or a similar behavior learned before?
2. Is “must” learned as one rule that transfers across cues, or did recall and down each simply grow strong through separate, massed practice?
3. Why does the dog keep working when no food or toy is in sight — the fun of learning itself, the human’s reactions, the memory of occasional rewards, or trust?
4. Which behaviors distinguish pressure the dog absorbs and recovers from, from pressure that produces helplessness or avoidance?
5. Of the teaching sequence that worked for this dog and me, which parts could other dogs and humans reuse?

As these questions find answers, this document will be revised.