Abstract
Thesis Abstract
Artificial life in the digital medium has been explored by many in an attempt to understand life on Earth, its origins, and the manner in which intelligence developed. This knowledge can then be used to make predictions about life elsewhere in the universe. Theoretically, the problem solving ability of evolution could also be harnessed and used to solve real world problems, although this is still to be aptly demonstrated. Thomas Ray has developed an artificial life simulator called Tierra which has been used by others such as Chris Adami to evolve creatures of increasing complexity and to study evolution and intelligence. However, they have found it difficult to evolve organisms past a basic level of complexity. This is partially due to the problem of maintaining learned abilities while encouraging growth through many levels of complexity.
This thesis investigated the evolution of intelligence - the ability of organisms to perform computations in the Tierra simulator. A novel two-stage fitness function called the "teaching function" was used in an attempt to overcome complexity barriers and evolve creatures that were able to perform a number of mathematical computations. The teaching function permitted the evolving organisms a certain degree of freedom to explore different methods of performing computations while implementing a "building block" mechanism to capture increases in ability. "IQ tests" were used by the teaching function to measure the computational ability of the organisms as well as to determine the appropriate rewards or punishments. Although the organisms did not evolve the ability to perform the mathematical computations, the teaching function was found to be useful in preventing evolutionary regression, when suitably implemented.
Experimentation revealed that the length of time between successive IQ
tests, the strength of the rewards and punishments and the point at which
the teaching function changed modes all contributed to its effectiveness.
Weaknesses in the implementation of the teaching function were discussed
and improvements suggested. Finally, the implications of this project
were considered in relation to evolution on Earth.