The Wolf, the Leash, and the Problem With Tool Use

I was three minutes into a YouTube video when I found myself thinking about wolves and tools.[1]

The video was Otters, Orcas, and Sea Wolves: Ocean Animals that Use Tools, on the KPassionate channel. The host was describing a remarkable observation from the central coast of British Columbia. A female wolf had learned how to raid crab traps. She swam out to a buoy bobbing on the surface, brought it ashore, then worked her way along the attached line, pulling until the submerged trap came into shallow water. She manipulated the trap until she got the bait cup, pulled it out, and ate the contents.[2]

The trap was underwater. From on shore, the buoy was the only visible part of a larger system. Yet she worked through that system in the correct order until the food became accessible.

When I first watched the video, the discussion that followed seemed to get hung up on whether this really counted as tool use. At one point the host posed a hypothetical: if her brother set the trap but she pulled it in, had they both used a tool?

I stopped thinking about the video and started thinking about the question. If your brother sets a crab trap and you later haul it in, you have both interacted with the same system in different ways. One of you may have assembled, baited, deployed, or otherwise configured it. The other retrieved it. Whatever vocabulary we eventually choose, nothing about the second person's use of the trap depends on having created the trap.

My first thought was that we were simply confusing tool use with tool manufacture. Humans manufacture tools, so the two ideas have become bundled together in our understanding of what a tool user does.

Consider a lawnmower. I did not manufacture my lawnmower. I have only the vaguest idea who did. Nevertheless, when I start it and push it through tall grass, nobody feels a need to convene a comparative-cognition panel to decide whether I am using a tool. Nor does the question become difficult if I press an electric starter rather than pull a cord, even though pressing the button directly accomplishes almost none of the physical work involved in starting the engine. I manipulate an interface whose causal consequences I understand well enough for my purposes.

Once I started looking for other examples of animal tool use, it didn’t take long before I was pondering man’s best friend.

Dogs do things all the time that we describe as "clever."[3] A dog learns to lift a gate latch. Another learns that pulling some part of an apparatus opens it. Dogs can learn to ring bells to summon their humans. And there is the familiar goofy golden retriever who wants to go for W-A-L-K-I-E-S, so she finds her leash, carries it over, and deposits it in front of the human who has apparently failed to notice that it is time to go outside.

The leash case is almost too ordinary to take seriously, which is why it is useful. The dog wants something she cannot directly produce. She cannot announce in English that she would like to go for a walk. Depending on the household, she may not be able to open the door, and she certainly isn't going to grab the car keys and drive herself to the park. But she has learned that manipulating one particular object changes the behavior of an extraordinarily capable part of her environment: the human.

In that respect, the leash starts to look rather like the button on my lawnmower. Pressing the button does almost none of the work required to start the engine. It activates a system capable of doing that work. The leash does something similar. The dog presents the leash and activates a human capable of opening doors, operating cars, attaching leashes, and generally rearranging the world until a walk occurs.

The dog is inducing another agent to behave in a way that accomplishes the dog's objective. In other words, other agents are not merely things a social animal observes, cooperates with, or communicates with. They can become recruitable parts of its route through the world.[4] We do not need to claim that the dog cognitively represents the entire sequence leash → human interprets request → human opens door → walk. Its represented route might be no more elaborate than give human leash, walk happens. The observable structure is interesting either way: the leash has very little utility to the dog except through another agent, and presenting it recruits an agent capable of manipulating the world in ways the dog cannot.

Researchers studying social cognition already have language for animals recruiting other agents as means to an end.[4] That work is usually discussed separately from physical tool use. The leash suggests why the separation can obscure something. An object and an intentional agent can participate in the same route from an animal's present situation to the outcome it wants.

Whether we insist on calling the leash a tool—or the human one—starts to seem less interesting than the structure of what the dog has done. A better question is beginning to appear: What path through its actionable world did the animal find?

The World You Can Operate

The concept of an affordance is useful here.[5] In ecological psychology, an affordance is, roughly, a possibility for action available to an organism. A branch affords perching to a bird in a way that it does not afford perching to me. A narrow opening may afford passage to a mouse but not to a wolf. The physical environment may be the same, but the actionable environment is not.

That distinction becomes especially important when comparing species. Animals don't merely perceive different portions of the world. Their bodies and cognitive capabilities make different portions of the world manipulable.

A wolf has teeth, jaws, paws, a powerful body, excellent mobility, extraordinary olfaction, and a sensory system adapted to the problems wolves have spent evolutionary time solving. What a wolf conspicuously does not have is a pair of hands. A wolf can move, pull, tear, chew, carry, dig, push, and reshape things to some degree, but its capacity for controlled precision transformation of matter is radically narrower than ours.

If we begin our search for animal technology with the human prototype—grasp object, modify object, wield object—we have quietly chosen a test that strongly favors animals equipped to do something resembling what human hands do.

But canids have another set of capabilities that matters just as much. They are social animals. They observe other agents, coordinate with them, anticipate behavior, pursue moving prey through sequences of action, and adjust their own behavior in response to what other animals are doing. Domestic dogs have spent thousands of years living in an environment containing a particularly peculiar and useful animal: us.

A human being is therefore part of a dog's actionable environment. A human can open things, retrieve things, provide things, remove obstacles, operate machinery, and take the dog somewhere it cannot take itself. Humans also respond with remarkable reliability to certain signals. The leash is interesting because it is simultaneously a physical object and an interface to an intentional agent.

This suggests a more useful way of comparing animals than asking whether each species belongs inside a category called tool user. We might instead ask what kinds of affordances are available to it: what it can perceive, what it can manipulate, what causal relations it can exploit, what other agents it can recruit, and what new affordances it can create.

I have been thinking of these as species-specific capability manifolds. The phrase sounds more forbidding than the idea. A species' body, senses, cognitive capacities, social organization, and environment jointly determine the shape of the world it can operate.

The contrast between canids and humans is enough to see what I mean by shape. A canid has relatively poor precision manipulation and limited ability to transform raw material into elaborate new objects, but it has powerful locomotor, sensory, social, and behavioral capabilities. A human retains a rich social world while adding hands, precision manipulation, extensive material transformation, symbolic communication, and an extraordinary ability to alter the environment itself. Those are not simply two scores on an intelligence scale. They make different regions of the same physical world available for action.

A legitimate comparison therefore need not ask which animal is "more intelligent." We can ask which kinds of paths are available within each animal's actionable world, how many intermediate changes those paths require, and what kinds of relations the animal can combine along the way.

Humans are especially unusual because we don't merely find affordances. We make them. We reshape stone, wood, metal, glass, silicon, and eventually the environment itself so that actions that were formerly impossible become easy.

Sometimes we make them so easy that all that remains is a button.

Paths Through the Landscape

The wolf suggests another dimension of the problem. It is not enough to ask what individual affordances an animal can exploit. Affordances can be composed.

A rock can break a shell and expose food. A latch can open a gate and permit passage. A rope can move something that cannot be reached directly. The result of one action can expose the next thing that can be acted upon.

Wolfgang Köhler described tool behavior in chimpanzees as a form of indirect thinking. Archaeologist Miriam Haidle later developed the idea of problem-solution distance, examining the intervening objectives and operations that can separate an organism from its ultimate goal.[6] That gives us a way of looking beyond the object called a tool and toward the structure connecting a problem to its solution.

Other research supplies adjacent pieces. Work on sequential tool use asks whether an animal can use one means to obtain another means before finally reaching the desired end.[7] Work on affordance landscapes has examined how one action changes which future actions become available.[8] Social-tool research, meanwhile, asks how another agent can become part of a solution.[4] These literatures arose around somewhat different problems, but they repeatedly encounter portions of the same structure.

What interests me is the common geometry underneath them. A behavior can be regarded as a path through an affordance landscape.

The path can be shallow. Bite food, eat food. It can contain an intermediate mechanical relation: manipulate rock, shell opens, eat animal. Or it can become considerably deeper.

The wolf begins with the buoy. Manipulating the buoy makes the line accessible in a useful position. Pulling the line moves an invisible trap. Moving the trap into shallow water makes the trap itself manipulable. Manipulating the trap makes the bait cup accessible. The bait cup finally yields the food. She has transformed one accessible affordance into the next until she reaches the bait.

There is a depth of indirection in that path. The term itself is not novel; researchers in several fields have described degrees of causal indirection or closely related means-means-end structures.[6][7] What seems useful is to treat indirection depth as a general property of the path rather than a special property of something already classified as tool use.

That gives us something an observer can actually look for. What did the animal manipulate first? What changed as a consequence? What became actionable only after that change? How many such transitions were necessary before the goal became directly available? We do not yet need to pretend that this supplies a standardized numerical scale. The important shift is from recording only tool use: yes or no to recording the structure of the route the animal successfully traversed.

The intermediate stages do not all have to be tools, and they do not all have to work in the same way. Return to the dog with the leash. At minimum, the dog manipulates an object whose usefulness depends on changing the behavior of another agent. The path changes kind: from physical manipulation to social communication and, if the human cooperates, back into physical changes in the environment.

That distinction may matter as much as path length. Two behaviors can contain several intermediate stages while demanding very different kinds of relations. A rope can transmit force to a distant object. A leash presented to a human can transmit information to another mind. Describing both as paths lets us notice the common structure without pretending that mechanical causation and social recruitment are cognitively identical.

It also lets us be more careful about what we attribute to the animal. The wolf's successful path may objectively be described as buoy → line → submerged trap → bait cup → food. That does not establish that she represented that complete sequence before she touched the buoy.

She may have understood from the beginning that buoy led to line, line to trap, trap to bait cup, and bait cup to food. She may instead have known from previous experience that the buoy was worth retrieving and then encountered each successive affordance as her actions changed the situation. Those are different claims about cognition, and the behavior alone does not settle them.

The same applies to the leash. The objective route to the walk contains a human agent and all the things that human subsequently does. The dog's represented route might be much simpler: give human leash, walk happens. We can describe the structure of the successful path without quietly attributing the entire structure to the animal's internal model.

The path exists either way. How much of the path exists in the animal's model of the future is a separate empirical question.

Changing the Landscape

Tool manufacture now returns to the story in a different role.

Manufacturing a tool is one way of creating an affordance that did not previously exist. Tie a stone to a handle and the resulting object affords actions neither component afforded in quite the same way alone. Build a latch and a gate acquires a controllable state transition. Attach a rope to a trap and an object sitting underwater can be manipulated from a distance. Add a buoy and the otherwise invisible system acquires a visible, reachable interface.

Humans do this compulsively. We don't merely navigate affordance landscapes; we rearrange them. We create new actionable objects, connect things that were previously disconnected, shorten difficult paths, and leave the resulting structures lying around for other humans—and occasionally wolves—to discover.

That makes the coastal wolf an unusually good case because she encountered an affordance landscape partly engineered by another species. Humans had built a system consisting of buoy, line, trap, bait cup, and food. None of it was designed for wolves. Nevertheless, this particular wolf discovered that enough of the system lay within her own capabilities to make the human-built path usable.

When I went looking for ways other researchers had approached this problem, I found pieces of the same structure in places that were not necessarily trying to answer the same question. Ecological psychology supplied organism-relative affordances; archaeology supplied problem-solution distance; comparative cognition supplied sequential means-end problems and social recruitment; work on action selection supplied landscapes in which one action exposes the next affordance. Even robotics has encountered related structures while trying to make machines reason about objects and other agents in a shared environment.

The interesting thing about that literature is less the terminology than the convergence. Different problems keep producing pieces of a common geometry: organisms have differently structured actionable worlds, and behavior can sometimes be understood as finding and composing routes through them.

That gives us somewhere more useful to go after abandoning the binary question. Instead of recording only whether an animal "used a tool," we can ask what was actionable to that animal, what sequence of changes connected its first action to its goal, whether the path remained mechanical or changed kind, and how much of that path the evidence allows us to say the animal actually represented.

Which brings me back to the wolf. Did she use a tool? Probably. Depending on the formal definition chosen, somebody can undoubtedly construct an argument about whether an attached buoy and rope satisfy the appropriate criteria. That argument may even be useful for maintaining continuity with the existing literature.

But watching the wolf haul in the trap, I now think I would ask a different question: What path through her actionable world had she discovered? She began with a floating buoy she could reach. Bringing it ashore gave her access to the line. Pulling the line brought an invisible trap into shallow water. The trap made the bait cup accessible, and the bait cup yielded food. Whatever portion of that sequence existed in the wolf's model before she began, those were the affordances she successfully traversed. She reached the water with a wolf's body and a human-built system in front of her, and worked her way from the part she could touch to the part she could eat.

Notes

[1] KPassionate, Otters, Orcas, and Sea Wolves: Ocean Animals that Use Tools | Ft Lindsay Nicole, YouTube. The wolf/trap segment begins at about 3:10.

[2] Kyle A. Artelle and Paul C. Paquet, “Potential Tool Use by Wolves (Canis lupus): Crab Trap Pulling in Haíɫzaqv Nation Territory”, Ecology and Evolution 15, no. 11 (2025): e72348. The paper reports the May 2024 observation near Bella Bella, British Columbia, and notes that the exact trap location is withheld under data-sharing agreements with the Haíɫzaqv Nation.

[3] Bradley Philip Smith, Robert George Appleby, and Carla Anita Litchfield, “Spontaneous tool-use: An observation of a dingo (Canis dingo) using a table to access an out-of-reach food reward”, Behavioural Processes 89, no. 3 (2012): 219–224. For the broader argument that familiar canine behavior may be overlooked as tool use, see James Brooks and Shinya Yamamoto, “Dog Stick Chewing: An Overlooked Instance of Tool Use?”, Frontiers in Psychology 11 (2021): 577100.

[4] Jacques Prieur and Simone Pika, “Gorillas’ (Gorilla g. gorilla) knowledge of conspecifics’ affordances: intraspecific social tool use for food acquisition”, Primates 61 (2020): 583–591. The paper treats another individual as a possible “social tool” and explicitly discusses degrees of control ranging from physical manipulation to reliance on self-initiated action by the recruited individual.

[5] James J. Gibson, The Ecological Approach to Visual Perception, first published 1979. Gibson’s ecological approach introduced affordances as possibilities for action defined in relation to an organism and its environment.

[6] Miriam N. Haidle, “Building a bridge—an archeologist's perspective on the evolution of causal cognition”, Frontiers in Psychology 5 (2014): 1472. Haidle develops “problem-solution distance,” cognigrams, and effective chains as ways of representing intervening objectives and causal structure in tool behavior, building in part on Wolfgang Köhler’s earlier treatment of tool use as indirect problem solving.

[7] Laurie R. Santos, Alexandra Rosati, Catherine Sproul, Bailey Spaulding, and Marc D. Hauser, “Means-means-end tool choice in cotton-top tamarins (Saguinus oedipus): finding the limits on primates’ knowledge of tools”, Animal Cognition 8 (2005): 236–246. The experiments explicitly test sequences in which one means must be used to obtain another before the final goal can be reached.

[8] Giovanni Pezzulo and Paul Cisek, “Navigating the Affordance Landscape: Feedback Control as a Process Model of Behavior and Cognition”, Trends in Cognitive Sciences 20, no. 6 (2016): 414–424. Their model treats behavior as selection among affordances at multiple levels and emphasizes how present actions alter which future affordances become available.