Manhasset, N.Y. - A research program in artificial intelligence has been refocused to simulate real biological organisms in hopes of gleaning an understanding of biological diversity. The goal is to create a living "road map" that encapsulates the history of evolution in an electronic petri dish called Avida. The researchers are investigating how complex organisms evolve from simple ones.
"These simulations will help direct research on living systems and will provide understanding of the origins of biocomplexity," said Richard Lenski, a researcher at Michigan State University who runs the program in tandem with Chris Adami at the California Institute of Technology. "Darwinian evolution affects DNA and computer code in much the same way, which allows us to study the process of evolution in an electronic medium," said Lenski.
Engineers have long sought to harness the principles of Darwin's survival-of-the-fittest theory to promote the self-organization of higher-level functions from a foundation of simple rules. "Simple functions, each unremarkable if viewed in isolation, can string together into a long series of mutations that natural selection weeds out to evolve higher-level functions," said Sam Scheiner, program director in the division of environmental biology at the National Science Foundation, which funded the research.
The overall aim of the Avida project is to use the road map being created to catalog the steps that must be taken to ensure that biodiversity is not thwarted by overdevelopment of natural resources. Researchers have argued for a long-term longitudinal study, since mutations that often seem counterproductive in their first generation later appear to steer the species in a new direction that, after a few generations, leaves it better off than before the random mutation.
"Some mutations look like really bad events when they happen, but because each simple function contributes to the formation of complex functions, sometimes things that look bad turn out to have important benefits to the population over time," said Lenski.
The artificial organisms inside Avida reproduce, mutate and live longer by performing mathematical calculations. Along the way they reproduce, leaving copies of themselves wherever they go.
Because time is speeded up inside Avida, what would take thousands of years of evolutionary time can occur in a matter of minutes or hours. In addition, every step an organism made along its evolutionary way is tracked in a sophisticated audit trail.
One of the most useful functions in Avida is the ability to roll time backward to see what happened. By looking back, each microstep along the way can be reviewed and key steps catalogued. Because there is so much detail, the team has developed tools that zero in on what is important. "We've found that new functions borrow from old structures that can be tweaked here and there to eventually create a new function, which is a lot easier to understand than how something entirely new could have been invented," said Lenski.
Darwin himself felt that the existence of complex organs such as the eye was a serious challenge to his theory. And he suspected that a complex organ must gradually emerge through a long chain of small improvements. In addition, the intermediate forms might have entirely different functions than the final version. The problem for evolutionists has been the lack of detail in the fossil record. While major stages in the development of an organ or organism appear, a large amount of critical detail about the in-between steps is simply missing.Replication and competition
Evolution itself is based on three simple functions: A system must be able to replicate itself, random variations in the replication must occur and there must be competition that eliminates the less-fit outcomes of replication. With the advent of computers, those simple conditions could be easily encoded, leading to the fields of genetic-algorithm design and artificial-life systems.
Adami leads the Digital Life Laboratory at Caltech, where electronic petri dishes have been shedding light on artificial-life studies for years. For him, the computer is an ideal breeding ground because its principles match those of low-level life structures. "Often evolution can design things that solve problems better than we do using our own intelligence," said Robert T. Pennock, a member of the department of philosophy at Michigan State.
Avida is based on a concept of organism borrowed from computer architecture. In this model, an organism is a simple CPU with a circular sequence of instructions, representing DNA, along with an instruction-processing unit and two stacks. There are 26 basic instructions such as nop-A, if-less, pop, push or swap-stk.
The organism's metabolism consists in the endless execution of the sequence of instructions. Energy from the environment, or "food," is modeled as single-instruction processors (SIPs) that are fed to the CPU. The number of SIPs that a CPU receives is proportional to the length of its tape. Thus, as the CPU becomes more complex in terms of the length of its instruction tape, it is able to get more food from the environment, giving more-complex organisms a competitive advantage.
SIPs introduce new instructions to the CPU, allowing it to grow as well as to reproduce. Reproduction is achieved by copying the instruction tape. That produces an asexual reproduction similar to cell division. Random errors are introduced into the copying process to simulate mutations. Most offspring are at the same level as, or slightly less fit than, their parents, but a small number are more effective.
The purpose of each Avida organism is to evolve more-complex logic instructions starting with NAND, the most primitive. The realistic aspect of this scheme with regard to evolution is that a successful logical operation, built up from NAND primitives, must emerge simultaneously from a set of NANDs that are executed in a specific order along with I/O operations. So a single mutation will not in itself produce an evolutionary step.
Growing more complex
The researchers performed evolutionary runs starting with individuals that could replicate themselves but could not perform any logic operations except the simple NAND. The project was looking for an organism that had evolved the ability to take two 32-bit strings and compare them to see if they were the same-the EQU function, which, in relation to the complexity of the organisms, represented a highly complex sequence of basic instructions. The scientists determined the smallest program that could perform this operation and found that it consisted of 19 steps. To achieve that precise sequence simply through chance is extremely unlikely.
The CPUs were rewarded on an exponential scale for producing more-complex logic functions. The expectation was that once an offspring got to the complex EQU function, it would therefore become the dominant type. The Avida runs confirmed this. The EQU function first appeared at the 27,450th step in the program, and went on to become the dominate type. The final evolved organisms that were deemed most fit also were able to perform most of the simpler logical operations leading up to EQU, making them able to absorb the maximum amount of energy in the form of SIPs.
Examining the prior logic functions that evolved into the final EQU operation, the group found that, as in evolution, these prior evolved forms were not specifically related to the EQU function in terms of the types of logical operations they handled. Repeated runs revealed that the steps the organisms took to achieve this complex function were highly variable. In short, a wide variety of evolutionary strategies all can reach the same result.
The study has concluded that complex functions evolve from simple mutations of useful functions that have previously evolved, which is what Darwin supposed. Often the previous functions are lost in the mutation, leaving no trace in the final organism. Only by rolling back the program could the researchers show that evolution progresses via incremental improvements in effective organs. They verified that insight by not rewarding the formation of the intermediate forms. When this was the case, they found that the complex EQU instruction could not evolve.
-Chappell Brown contributed to this story.
See related chart