Cheaper than printing it out: buy the paperback book.

Out of Control

If nature transmitted information in both directions within organisms, it would allow the possibility of Lamarckian evolution, which requires two-way communication between gene and its products. The advantages of Lamarckism are awesome. When an animal needs faster legs to survive, it could use body-to-gene communication to direct the genes to make faster leg muscles, and then pass that innovation on to its offspring. Evolution would accelerate madly.

But Lamarckian evolution requires an organism to have a working index to its genes. If the organism met a harsh environment -- say extreme high altitude -- it would notify all the genes in its body able to influence respiration and ask them to adjust. The body of an organism can certainly communicate that message to other organs in the body by hardwired hormone and chemical circuits. And it could communicate the same to the genes if it could pinpoint the right ones. But that is the bookkeeping chore that is missing. The body does not keep track of how it solves a problem, so it cannot pinpoint which genes to pump up the muscle on the blacksmith's biceps, or which genes regulate respiration and blood pressure. And because there are millions of genes producing billions of features -- and one gene can make more than one feature and one feature can be made by more than one gene -- the complexity of accounting and indexing could exceed the complexity of the organism itself.

So it isn't so much that information can't be transmitted in the body to gene direction, it's more that communication is blocked because messages have no distinct destination. There is no central gene-authority to direct traffic. The genome is the ultimate decentralized system -- rampant redundancy, massive parallelism, no one in charge, no one looking over the shoulder of every transaction.

But what if there is some way around this? Genuine two-way genetic communication would light up an interesting bunch of questions: Would there be any biological advantage if such a mechanism were possible? What else would it take to have a Lamarckian biology? Could there have been a biological route to such a mechanism at one time? If it is possible, why hasn't it happened? Could we outline a working biological Lamarckism as a thought experiment?

In all probability, Lamarckian biology requires a type of deep complexity -- an intelligence -- that most organisms can't reach. But where complexity is rich enough for intelligence, such as in human organisms and organizations, and their robotic offspring, Lamarckian evolution is possible and advantageous. Ackley and Littman showed that computers programmed by humans could run Lamarckian evolution.

But in the last decade, mainstream biologists have acknowledged an observation a few maverick biologists have preached for a century: that when an organism acquires sufficient complexity in its body, it can use its body to teach the genes what they need to know to evolve. Because this mechanism is a hybrid of evolution and learning, it has great potential in artificial realms.

Every animal's body has a built-in but limited power to adjust to different environments. Humans can acclimatize to life at a significantly higher elevation. Our heart rate, blood pressure, and lung capacity must and will compensate for the lower air pressure. The same changes reverse when we migrate to a lower elevation. But there is a limit to the degree to which we can acclimatize. For us, it's around 20,000 feet above sea level. Beyond this altitude, the human body cannot stretch itself for long-term habitation.

Imagine a settlement of people living high in the Andes. They have moved from the plains into a niche where they are not exactly best suited -- the air is thin. For the thousands of years they have lived there, their hearts and lungs -- their bodies -- have had to work overtime to keep up with the altitude. If a "freak" should be born in their village, one whose body has a genetically more proficient way to handle the stress of high altitudes -- say, a better hemoglobin variety rather than faster heartbeat -- then the freak has an advantage. If the freak has children, then this trait could potentially spread through the village over generations because it is an advantage to lower stress on the heart and lungs. By the usual Darwinian dynamics of natural selection, the mutation of altitude acclimation comes to dominate the village gene pool.

On the surface there appears to be nothing but classical Darwinism at work here. But in order for Darwinian evolution to take place, the organism first had to survive in the niche for many generations without the benefit of genetic change. Thus it was the flexibility of the body that kept the population surviving long enough for the mutation to arise and fix itself in the gene. An adaptation spearheaded by the body (a somatic adaptation) is assimilated over time by the genes. Theoretical biologist C. H. Waddington called this transfer "genetic assimilation." Cyberneticist Gregory Bateson called it "somatic adaptation." Bateson likened it to legislative change in society -- first a change is made by the people, then it is made law. Writes Bateson, "The wise legislator will only rarely initiate a new rule of behavior; more usually he will confine himself to affirming in law that which has already become the custom of the people." In the technical literature, this genetic affirmation is also known as the Baldwin effect, after J. M. Baldwin, a psychologist who first published the idea as a "New Factor in Evolution" in 1896.

Let's say there is this other village in the mountains, this time in the Himalayas, in a valley called Shangri La, whose residents' bodies are able to acclimatize up to 30,000 feet -- 10,000 more than the Andes folks -- but who are also able to live at sea level. Over generations a mutation spreads to hardwire this talent into the villagers' genes, just as it did in the Andes. Of the two alpine villages, the Himalayan population now has a body type that is more stretchable, more flexible, and therefore, in essence, more evolutionarily adaptable. It may seem like a textbook example of Lamarckism, but giraffes who can evolve the most stretch in their necks can stake out an adaptation with their bodies long enough for their genes to catch up. As long as they keep their hides adjustable to a wide range of stresses, they'll have a competitive advantage in the long run.

The evolutionary moral is that it pays to invest in a flexible phenotype. It makes better sense to keep an adaptable body in service than to have a rigid body wait around for a mutation to pop up anytime an adaptation is needed. But somatic flexibility is "expensive." An organism cannot be equally flexible everywhere, and accommodating one stress will decrease its ability to accommodate another. Hardwiring is more efficient, but it takes time; for hardwiring to work, the stress must remain constant over a long period. In a rapidly changing environment, the tradeoff favors keeping the body flexible. An agile body can foreshadow, or more accurately, try out possible genetic adaptations, and then hold a steady line to them, as a hunting dog holds to a grouse.

But the story is even more radical than it appears because it is behavior that moves the body. The giraffe had to first want (for whatever giraffey reasons) higher leaves, and then had to reach for them over and over again. The humans had to choose to move to more alpine villages. By behavior, an organism can scout its options, and explore its space of possible adaptations.

Waddington said genetic assimilation, or the Baldwin effect, was about converting acquired traits into inherited traits. What it really comes down to is the natural selection of traits controls. Genetic assimilation bumps up the reach of evolution a notch. Instead of being able to tune the dial to the best trait, somatic and behavioral adaptation gives evolution quicker control over what the dials are and how far and in what direction they turn.

Behavioral adaptation works in other ways, too. Naturalists have verified that animals are constantly roaming out of their adapted environment and taking up homes in areas where they "don't belong." Coyotes creep too far south, or mockingbirds migrate too far north. And then, they stay. Their genes endorse the change by assimilating an adaptation which began, perhaps, as a vague desire.

What begins as vague desire can skate dangerously close to the edge of classical Lamarckism when it reaches individual learning. One species of finch learned to pick up a cactus needle to poke for insects. By this behavior the finch opened up a new niche to itself. By learning -- perceived as a deliberate act -- it altered its evolution. It is entirely possible, if not probable, that its learning will affect its genes.

Some computerists use the term "learning" in a loose, cybernetic sense. Gregory Bateson described the flexibility of the body as a type of learning. He saw little in its effect to distinguish the kind of search the body performed from the kind of search that either evolution or mind did. By this reckoning, a flexible body learns to acclimatize to stresses. "Learn" means adaptation within a lifetime instead of over lifetimes. The computerists make no real distinction between behavioral learning and somatic learning. What matters is that both types of adaptation search the fitness space within the lifetime of an individual.

An organism has great room to reshape itself within its lifetime. Robert Reid, at the University of Victoria, Canada, suggests that organisms can respond to environmental change with the following types of plasticity:

  • Morphological plasticity
    (An organism can have more than one body form.)

  • Physiological adaptability
    (An organism's tissues can modify themselves to accommodate stress.)

  • Behavioral flexibility
    (An organism can do something new or move.)

  • Intelligent choice
    (An organism can choose, or not, based on past experiences.)

  • Guidance from tradition
    (An organism can be influenced or taught by others' experiences.)

Each of these freedoms is a front along which the organism can search for better ways to refit itself in a coevolutionary environment. In the sense that they are adaptations within a lifetime which can later be assimilated, we can call these five options, five varieties of inheritable learning.