Evolving Improvisational Intelligence
William H. Calvin
University of Washington
Department of Psychiatry and Behavioral Sciences
Seattle WA 98195-1800 USA
email@example.com, reprints via
Piaget (1929; 1952) said that intelligence is what you use when you don't know what to do.
Most of the time, improvisation isn't necessary as the individual has encountered similar
situations before, has a repertoire of actions, and simply chooses. But novel situations often
demand novel responses, such as most of the sentences we speak aloud. Every time we
contemplate the current collection of leftovers in the refrigerator, trying to figure out what else
needs to be fetched from the grocery store before fixing dinner, we're exercising an aspect of
intelligence not seen in even the smartest ape. This article addresses how novel plans can be
generated and tested inside the brain before acting, and how such abilities might have evolved in
our hominid ancestors from the base exhibited by primate intelligence, social behaviors, and
Behavioral Complexity Not a Guide
Elaborate, complex behaviors initially seem like reasonable signs of intelligence. But many
complex behaviors in animals are innate (a nice short word meaning without apparent learning:
Gould and Gould 1994, Byrne 1994); such relatively stereotyped movement patterns exhibit no
more insight or understanding of purpose than does a computer program. Both innate (whale
song and insect nest building) and individually learned behaviors can be long and complex
without exhibiting improvisational intelligence, e.g., the idiot savant may demonstrate enormous
detailed recall but a poor ability to break up the pattern into meaningful parts and recombine
Versatility via chaining behaviors, as when whales and birds chain together song sequences, is
also misleading. The most mindless of behaviors may be segued, the completion of one calling
forth the next: courtship behavior may be followed by intricate nest building, then a segue into
egg laying, then incubation, then the stereotyped parental behaviors. Indeed, the more complex
and "purposeful" the behavior is, the further it may be from intelligent behavior, simply because
natural selection has evolved a sure-fire way of accomplishing it, with little left to chance.
Learning, after all, usually is focused on far simpler things than the complex chains of all-important behaviors.
Most animals in most contexts don't appear to have much need for understanding, in our sense of
appreciating the underpinnings of events, and they don't attempt innovations except by modest
variations and a slow learning process. It's as if thinking were a little-used backup, probably too
slow and error-prone to be used regularly.
The best indicators of intelligence may be found in connections with the simpler but less
predictable problems that confront animals, novel situations where evolution has not provided a
standard response and the animal has to improvise by using its intellectual wherewithal. While
we often use the term intelligence to encompass both a broad range of abilities and the efficiency
with which they're enacted, it also implies flexibility and creativity, an "ability to slip the bonds
of instinct and generate novel solutions to problems" (Gould and Gould 1994, p. 70).
The size of the response repertoire is one important factor in improvisational intelligence. Dogs
have many species-typical behaviors such as herding and alarm barks; they can learn many more.
Speed of learning is also related to intelligence; one reason that dogs and dolphins can achieve a
wider repertoire of behavior with training is that they learn faster than cats usually do (Coren
The selection of the appropriate behavior may be the key to sorting out animal intelligence
claims. In many of the "aren't-they-clever" animal stories, the animal may not be thinking for
itself but merely responding to a command. Piaget's element of creativity, in the face of an
ambiguous task, is usually missing except during play.
The scientific literature on nonhuman intelligence tries to cope with innovation (Griffin 1984),
but since an innovation is not a repeated action, it's hard to avoid a series of anecdotes. The
usual scientific hazards of anecdotal evidence can be somewhat avoided by emphasizing
comparisons between species. For example, most dogs can't untangle a leash from around a tree
but a chimpanzee seems to have what it takes. A dog-leash-style snap fastener on the cage door
will suffice to keep most small monkeys inside, even if they can reach the latch to fiddle with it
(D. M. Rumbaugh, personal communication 1995); the great apes can figure this out, so padlocks
are needed on their cages.
Improvisation En Route versus Planning
For most of your movements, such as raising a coffee cup to your lips, there is time for
improvisation en route. If the cup is lighter than you remembered, you can correct its trajectory
before it hits your nose. Thus, a complete advance plan really isn't needed; a goal and periodic
piecewise elaboration will suffice. You get started in the general direction and then correct your
path, just as a moon rocket does. Perhaps we should reserve the term planning for something
that requires multiple stages of the move to be assembled in advance of action, rather than
organizing the later stages after getting the initial moves in motion, which goal-plus-feedback can
There is surprisingly little evidence for this kind of multistage planning, even for the great apes,
even for frequent behavior options. None of the termite-fishing chimps, as Bronowski (1967;
1978, p. 33) pointed out, "spends the evening going round and tearing off a nice tidy supply of a
dozen probes for tomorrow." Multistage planning is best seen in ballistic movements (e.g.,
throwing, hammering, kicking, clubbing, spitting) and in an advanced type of social intelligence:
making a mental model of someone else's mental model, then exploiting it. Chimps can deceive
one another (they can guess what another animal is likely to be anticipating, and exploit it); most
monkeys don't seem to have the mental machinery to pull off a deception (Bryne & Whiten
1988). Imagine a chimp that cries "food" in a place where there is no food, and then quietly
circles back through the dense forest to where she actually saw the food earlier. While the other
chimps beat the bushes at the site of the food cry, she gets to eat all the food rather than having to
So multistage planning for novel situations is surely an aspect of intelligence, indeed one that
appears greatly augmented in the transition from the ape brain to the human brain. It seems
closely related to insight. Russell (1927) noted that "Animals studied by Americans rush about
frantically, with an incredible display of hustle and pep, and at last achieve the desired result by
chance. Animals observed by Germans sit still and think, and at last evolve the solution out of
their inner consciousness." But if chance operates offline during preparation for insightful
action, and is shaped by memorized environments in the same way as realtime environments
shape darwinian processes in the immune response and species evolution, it may be quite
Barlow (1987) says that intelligence is all about making a guess that discovers some new
underlying order. "Guessing well" neatly covers a lot of ground: finding the solution of a
problem or the logic of an argument, happening upon an appropriate analogy, creating a pleasing
harmony or witty reply, or guessing what's likely to happen next.
Versatility as a Virtue
When the chimpanzees of Uganda arrive at a grove of fruit trees, they often discover that the
efficient local monkeys are already speedily stripping the trees of edible fruit (Ghiglieri, 1988).
The chimps are much more versatile animals: they could turn to termite fishing, or perhaps catch
a monkey and eat it, but in practice their population is severely limited by that competition with
the monkeys, despite a brain twice the size of their specialist rivals. At the moment, versatility
isn't helping those chimpanzees very much.
A good memory would also seem to be a virtue, but it can also get you in trouble, if you can't
escape old rules to try out new ones. Rumbaugh (personal communication, 1995) notes that
prosimians and small monkeys often get trapped by the first set of discrimination rules they are
taught, unlike rhesus monkeys and apes which can learn a new set of rules that violates the old
ones. We, too, can overlay a new category over an old one, but it is sometimes difficult:
categorical perception is presumably why some Japanese have such difficulty distinguishing
between the English sounds for L and R. The Japanese language has a phoneme which is in
between the two in sound space, and both L and R are captured by that traditional category; those
Japanese-speakers who can't hear the difference will have trouble pronouncing the two sounds
Language and Improvisational Intelligence
A major boost in improvisational intelligence may have come with the improvement in the neural
structures needed for language. Words themselves, as Gregory (1981) points out, are social tools
that confer intelligence upon the user. For example, all the little grammatical words (Bickerton,
1990) help to position objects and events relative to each other on a mental map: they can
express relative location (above, below, in, on, at, by, next to) and relative direction (to, from,
through, left, right, up, down). Then there are the ones for relative time (before, after, while, and
the various indicators of tense), relative number (many, few, some, the -s of plurality), relative
possibility (can, may, might), relative contingency (unless, although, until, because), possession
(of, the possessive version of -s, have), agency (by), purpose (for), necessity (must, have to),
obligation (should, ought to), existence (be), nonexistence (no, none, not, un-), and so forth.
Because relationships are what analogies usually compare (as in bigger-is-faster), this
positioning-words aspect of grammar could also form a foundation for an important aspect of
Syntax is an advanced aspect of grammar: it is a boxes-within-boxes structuring of relative
relationships in your mental model of things that goes far beyond conventional word order or the
aforementioned "positioning" aspects of grammar. Argument structure supplements it, as when
the verb give sends you searching for three nouns to fit into the roles of actor, recipient, and
object given. With all three aspects of grammar, a speaker can quickly convey a mental model to
a listener of who did what to whom. But this requires sophisticated guessing on the part of the
listener, as Savage-Rumbaugh and Lewin (1994, p. 174) point out:
Comprehension demands an active intellectual process of listening to another party while trying
to figure out, from a short burst of sounds, the other's meaning and intent - both of which are
always imperfectly conveyed. Production, by contrast, is simple. We know what we think and
what we wish to mean. We don't have to figure out "what it is we mean," only how to say it. By
contrast, when we listen to someone else, we not only have to determine what the other person is
saying, but also what he or she means by what is said, without the insider's knowledge that the
Perhaps the mechanisms for foresight are similar to those used in the fancier aspects of mental
grammar, the ones involving long-term dependencies, as when basic word order is replaced by
the alternate forms for those who-what-when questions. Perhaps phrase structure and the
obligatory roles of argument structure (Bickerton, 1990) also have mental mechanisms that are
useful for foresight in a more general way.
Chunking and Sequencing
Jensen (1992) notes that the two strongest influences on intelligence's general factor g are speed,
such as how many questions you can answer in a fixed amount of time, and the number of items
that you can mentally juggle at the same time. Juggling a half-dozen things at the same time is
one of those abilities measured by multiple-choice tests, particularly analogy questions: A is to B
as C is to [D,E,F]. It also shows up in our ability to remember phone numbers long enough to
dial them. Many people can retain a seven-digit number for 5-10 seconds, but will resort to
writing it down if faced with an out-of-area number or an international one of even greater
length. When getting close to your limit, you try to collapse several items into one chunk, so as
to make more room (Simon, 1983).
Evolving a Better Brain
While hominid evolution surely involved improvements in toolmaking and hunting, we should
not expect to find a "toolmaking area" of cortex bulging out. Though olfaction is an exception,
the general rule (Finlay & Darlington, 1995) seems to be "enlarge one cortical area, enlarge them
all." But that's anatomy, not function. Not all functions enlarge equally: functional maps of
human temporal lobe, for example, are not simply an enlargement of the maps of temporal lobe
in the monkey (Calvin & Ojemann 1994). If other areas enlarge "for free," any natural selection
for neocortical size subserving one function may, pari passu, benefit other functions.
Also, we must bear in mind that the underlying brain mechanisms might serve multiple
functions, any one of which could be driven by natural selection, and so incidentally benefit the
others. Bundling (paying for one thing, but getting something else "free") is a familiar marketing
strategy; might good guessing or planning come bundled with some other ability, simply because
core facilities can be multipurpose?
One multipurpose candidate is a neural mechanism for stringing things together in structured
ways, ones that go far beyond the sequences produced by other animals. Besides combining
words into sentences, we combine notes into melodies, steps into dances, and elaborate narratives
into games with procedural rules. Might the production of structured strings be a core facility of
the brain, useful for language, storytelling, planning ahead, games, and ethics? Might natural
selection for any of these abilities augment the common neural machinery, so that improved
grammar incidentally serves to expand plan-ahead abilities and improvisational intelligence?
Because of this multifunctionality, planning ballistic movements (Calvin, 1983, 1990, 1993) may
have once promoted language, music, and intelligence. Apes have elementary forms of the rapid
arm movements that we're experts with - hammering, clubbing, and throwing - and one can
imagine hunting and toolmaking scenarios that, in some settings, were important additions to the
basic hominid gathering and scavenging strategies. If the same "structured string" core facility is
used for the mouth as is used for ballistic hand movements, then improvements in language
might promote manual dexterity. It could work the other way, too: accurate throwing opens up
the possibility of eating meat regularly, of being able to survive winter in the temperate zone -
and of talking better as an incidental benefit, a "free lunch."
Social Intelligence and Sexual Selection
Social intelligence is another aspect of intelligence, not merely because mimicry adds to the repertoire but
because of the daily challenges that social life (living in groups) poses for innovative problem solving.
Humphrey (1984) considers social intelligence to be of primary importance because more frequent and
complex than tool-using cleverness:
Social primates are required by the very nature of the system they create and maintain to be calculating
beings; they must be able to calculate the consequences of their own behaviour, to calculate the likely
behaviour of others, to calculate the balance of advantage and loss - and all this in a context where the
evidence on which their calculations are based is ephemeral, ambiguous and liable to change, not the least
as a consequence of their own actions. In such a situation, 'social skill' goes hand in hand with intellect,
and here at last the intellectual faculties required are of the highest order. The game of social plot and
counter-plot cannot be played merely on the basis of accumulated knowledge. It asks for a level of
intelligence which is, I submit, unparalleled in any other sphere of living.
Natural selection for social intelligence would not involve the usual staying-alive aspects that are
commonly stressed in adaptationist arguments. The advantages of social intelligence would
instead manifest themselves primarily via what Darwin called sexual selection. In harem-style
mating systems, only a few males get the chance to mate, after having tried to outsmart or
outpush the others. In female choice mating systems, acceptability as a social companion is
likely to be important for males, e.g., good at grooming, willing to share food (de Waal, 1995),
etc. I argue elsewhere (Calvin, 1993) that social acceptability for extended periods would have
been an excellent setup for improving language abilities, were a female to insist on male
language ability at least as good as her own.
Environmental Selection for Intelligence
Almost every example of intelligent behavior has an adaptationist rationale associated with it (for
those associated with language and toolmaking, see Gibson and Ingold 1993). Yet of the
millions of branches on the tree of species, only a few other limbs have actually evolved
intelligence approaching mammalian levels (octopus, raven). The occasional advantages
conferred by an intelligent guess or improvisation must fit into an evolutionary picture dominated
by species stability, climatic change, and tradeoffs.
Versatility is not always a virtue, and more of it is not always better. As frequent airline travelers
know, passengers who only have carry-on bags can get all the available taxicabs while those
burdened by three suitcases await their checked luggage. On the other hand, if the weather were
to become so unpredictable and extreme that everyone had to travel with clothing ranging from
swim suits to Arctic parkas, the "jack of all trades" would have an advantage over the "master" of
Whether or not versatility is important during an animal's lifespan depends on the relative time
scales: for both the modern traveler and the evolving ape, it's a matter of how fast the weather
changes and how long the trip lasts. I argue elsewhere (Calvin 1990) that the abrupt climate
shifts that are seen superimposed on the ice ages of the last 2.5 million years played a major role
in preventing efficiency from dominating versatility.
Darwinian Processes for Improvisation
The notion of trial and error was developed by Bain in 1855. Safety is the big problem with trial
combinations, ones that produce behaviors that have never been done before. Even simple reversals in
order can yield dangerous novelty, as in "Look after you leap." Craik (1943, p. 61) proposed that:
the nervous system is... a calculating machine capable of modeling or paralleling external events...If the
organism carries a "small-scale model" of external reality and of its own possible actions within its head, it
is able to try out various alternatives, conclude which is the best of them, react to future situations before
they arise, utilise the knowledge of past events in dealing with the future, and in every way to react in a
much fuller, safer and more competent manner to the
emergencies which face it.
Humans can simulate future courses of action and weed out the nonsense off-line; as Popper
said, this "permits our hypotheses to die in our stead." Creativity - indeed, the whole high end
of intelligence and consciousness - involves playing mental games that shape up quality.
The best examples of creative processes that evolve quality are darwinian: species
evolution in millennia and the immune response on the far-shorter time scale of days to weeks.
William James was talking about mental processes operating in a darwinian manner as early as
1874, only 15 years after On the Origin of Species.
What might it take to operate a darwinian processes on the milliseconds to minutes time
scale of thought and action? When I try to abstract the essential features of a darwinian
process from what we know about species evolution and the immune response, it
appears that a Darwin Machine (Calvin, 1987) must possess six essential properties, all
of which must be present for the process to do anything interesting:
From all this, one gets that surprising darwinian drift toward patterns that almost seem
designed for their environment. In the cardboard version of darwinian, particular parts
such as "natural selection" are often confused with the entire darwinian process, but no
one "essential" by itself will suffice. Without all six essentials, the process will shortly
grind to a halt.
- It involves a pattern. Classically, this is a string of DNA bases called a gene. As
Dawkins (1976) pointed out, the pattern could be a melody, and he usefully
coined the term meme for such patterns. It could also be the brain patterns
associated with a thought.
- Copies are somehow made of this pattern. Cells divide. People hum or whistle a tune
they've overheard. Indeed, the unit pattern (that's the meme) is defined by
what's semi-reliably copied, e.g., the gene's DNA sequence is semi-reliably
copied while whole chromosomes or organisms are not.
- Patterns occasionally change. Point mutations from cosmic rays may be the best known
alterations, but far more common are copying errors and (as in meiosis) shuffling
- Copying competitions occur for occupation of a limited environmental space. For
example, several variant patterns called bluegrass and crabgrass compete for my
- The relative success of the variants is influenced by a multifaceted environment. For
grass, it's the nutrients, water, sunshine, how often it's cut, etc. We sometimes
say that the environment "selects" or that there is selective reproduction or
selective survival. Darwin called this biasing by the term natural selection.
- The next generation is based on which variants survive to reproductive age and
successfully find mates. The high mortality among juveniles makes their
environment much more important than that of adults. This means that the
surviving variants place their own reproductive bets from a shifted base, rather
than the original center of variants at conception (this is what Darwin called an
inheritance principle). In this next generation, a spread around the currently
successful is again created. Many new variants will be worse than the parental
average but some may be even better "fitted" to the environment's collection of
Building Blocks of Intelligence
In our search for suitable brain mechanisms for guessing intelligently, we now
have 1) those trees of syntax that structure strings, 2) argument structure with all its
clues about probable roles for words, 3) those mapping words such as near-into-above, 4)
the limited size of scratchpad memory and the consequent chunking tendencies, and 5)
common-core facilities for fancy sequences, with quite a lot of need for extra copies of
the neural patterns used to produce ballistic movements. Our sixth clue, from
darwinian processes, now appears to be a whole suite of features: distinctive patterns,
copying them, establishing variants via errors (with most of them coming from the most
successful), competition, and biasing copying competitions via a multifaceted
environment. What's more, it looks as if the multifaceted environment is partly
remembered and partly current. I discuss elsewhere (Calvin 1994) the neocortical
circuitry for implementing darwinian processes.
Darwinian processes allow random beginnings to be bootstrapped into
something of quality, such as a good guess. But a Darwin Machine is probably only the
high end of intelligence's building blocks. We could compare two promising species (or
artificial creations) by enumerating how many building blocks of intelligence that each
had managed to assemble, plus how many stumbling blocks they'd managed to avoid.
My current list would emphasize:
- A wide repertoire of movements, concepts such as words, and other tools. But even
with a large vocabulary from cultural sharing over a long lifespan, high
intelligence still needs additional elements in order to make novel combinations
- A tolerance for creative confusion that allows an individual to occasionally escape
old categories and create new ones.
- More than a half-dozen simultaneous workspaces ("windows") per individual -
enough to pick and choose between analogies, but not so many as to obviate our
acronymlike tendency to chunk and create new vocabulary.
- Ways of establishing new relationships between the concepts in those workspaces,
relations fancier than the is-a and is-larger-than which many animals can grasp.
Tree-like relationships seem particularly important for our kind of linguistic
structures. Our ability to compare two relationships (analogy) enables
operations in a metaphorical space.
- Off-line shaping up before acting in the real world, somehow incorporating the six
darwinian essentials (patterns that copy, vary, compete judged by multifaceted
environment, with the more successful providing the center for the next round of
variants) and some accelerating factors (equivalents of recombination, climate
change, islands), with shortcuts so that the darwinian process doesn't have to start
from primitive representations but can operate at a high level, comparing ideas
rather than movements.
- Guessing at long-term strategies as well as short-term tactics, making moves that
help set the stage for a future feat. Forming agendas and monitoring their
progress is even better.
Bain, A. (1855; 1868). The Senses and the Intellect, p. 574. London: Longmans,
Barlow, H. B. (1987). Intelligence: the art of good guessing. In R. L. Gregory
(ed.), Oxford Companion to the Mind., pp. 381-383. Oxford: Oxford University Press.
Bickerton, D. (1990). Language and Species. Chicago IL: University of Chicago
Bronowski, J. (1967 lecture, 1978). The Origins of Knowledge and Imagination. New
Haven CT: Yale University Press, p. 33.
Byrne, R. W. (1994). The evolution of intelligence, pp. 223-265. In P. J. B. Slater,
T. R. Halliday (Eds.), Behaviour and Evolution. Cambridge: Cambridge University Press.
Byrne, R., & Whiten, A., eds. (1988). Machiavellian Intelligence: Social Expertise and
the Evolution of Intellect in Monkeys, Apes, and Humans. Oxford: Oxford University Press.
Calvin, W. H. (1983). A stone's throw and its launch window: timing precision
and its implications for language and hominid brains. Journal of Theoretical Biology, 104,
Calvin, W. H. (1987). The brain as a Darwin machine. Nature, 330, 33-34.
Calvin, W. H. (1990). The Ascent of Mind: Ice Age Climates and the Evolution of
Intelligence. New York: Bantam. Available:
Calvin, W. H. (1993). The unitary hypothesis: A common neural circuitry for
novel manipulations, language, plan-ahead, and throwing? In K. R. Gibson & T. Ingold
(Eds.) Tools, Language, and Cognition in Human Evolution. New York: Cambridge
University Press, pp. 230-250. Available:
Calvin, W. H. (1994). The emergence of intelligence. Scientific American, 271(4),
Calvin, W. H., & Ojemann, G. A. (1994). Conversations with Neil's Brain: The
Neural Nature of Thought and Language. Reading MA: Addison-Wesley. Available:
Coren, S. (1994). The Intelligence of Dogs: Canine Consciousness and Capabilities.
New York: Free Press,
Craik, K. J. W. (1943). The Nature of Explanation. Cambridge: Cambridge
Dawkins, R. (1976). The Selfish Gene. New York: Oxford University Press.
de Waal, F. B. M. (1995). Bonobo sex and society. Scientific American, 272(4), 82-88.
Finlay, B. L., & Darlington, R. B. (1995). Linked regularities in the development
and evolution of mammalian brains. Science, 268, 1578-1584.
Ghiglieri, M. P. (1988). East of the Mountains of the Moon. New York: Macmillian,
Gibson, K. R., & Ingold, T., eds. (1993). Tools, Language, and Cognition in Human
Evolution. New York: Cambridge University Press.
Gould, J. L., & Gould, C. G. (1994). The Animal Mind. New York: Scientific
Gregory, R. L. (1981). Mind in Science. London: Weidenfeld and Nicholson.
Griffin, D. R. (1984). Animal Thinking. Cambridge MA: Harvard University Press.
Humphrey, N. K. (1984). Consciousness Regained. Oxford: Oxford University
James, W. (1874 in letters; 1880). Great men, great thoughts, and the
environment. Atlantic Monthly 46(276):441-459.
Jensen, A. R. (1992). Understanding g in terms of information processing.
Educational Psychology Review 4:271-308.
Piaget, J. (1929; 1952). The Origins of Intelligence in Children. New York:
International Universities Press (translation of Naissance de l'intelligence chez l'enfant),
Savage-Rumbaugh, E. S., & Lewin, R. (1994). Kanzi: The Ape at the Brink of the
Human Mind. New York: John Wiley & Sons.
Russell, B. (1927). An outline of philosophy. London: G. Allen & Unwin, Ltd.
Simon, H.A. (1983). Reason in Human Affairs. Palo Alto CA: Stanford University
|| Home Page || Calvin
publication list || My Science Surf
column || The
Calvin Bookshelf || Revised 6 December 1996