Autonomy Essay

Joseph Huttner
Dr. Lindell
September 27, 2005
Hist. of Mech. Thought

What Makes a Computing Machine Autonomous?

Autonomous computers are like any other machines. They are given sets of inputs (programs) and turn them into outputs (data). They use power (electricity) to run their tasks, and give off energy (in the form of heat). There are three distinct requirements that must be met in order for a computing machine to be autonomous, all of which relate to the general ideas of autonomy. The computer must be self-sufficient, it must have correct, measurable output, and it must positively adapt through learned behavior.

An autonomous computer requires only energy to exist and operate, making it self-sufficient. Once the computer is programmed and powered on, the computer will automatically carry out its tasks, and will never need intelligent input from a human or any other source to carry out these tasks. Imagine an advanced compound telescope with tremendous computing powers designed and programmed to plot the orbits of planets. This may be an amazing feat, but if the telescope needs a human to input the initial coordinates of the planets, then the telescope is not autonomous. Contrastingly, imagine a graphing calculator programmed at startup to add iterations of two, beginning with the number zero (i.e. 0, 2, 4, 6….). Although this calculator has far less computing power than the compound telescope, it satisfies the self-sufficient requirement for autonomy. This example illustrates that self-sufficiency does not correlate to a computer’s degree of complexity or computing power, and that simple machines can satisfy at least some of the requirements for autonomy.

Whatever task an autonomous computer is designed to do, there must be an observable output verifying the computer’s efforts (usually visual, graphical, or audio form), and the output must be correct. To determine if the output is correct, it must be compared to known specifications or a predetermined list of examples. For example, if a computer was calculating multiplication tables, we would compare its tables to tables that are known to be correct. Without a correct and observable output, we have no way of knowing if the computer is doing work, and whether or not it is accomplishing its task.

The last requirement of autonomy is the computer must achieve positive adaptation through learned behavior. This is undoubtedly the most complex, and least easily attained of the three requirements. An autonomous computer is capable of observation. It recognizes an error when it is encountered. But rather than stopping its processes or displaying an incorrect output, the computer accounts for the error, and automatically modifies its programs so that the error will not resurface.

A comparison between a non-autonomous computer and an autonomous computer illustrates the idea of positive adaptation through learned behavior. Suppose a non-autonomous computer needs to use all of its processing power to carry out large calculations. But if the computer runs at full power for more than twenty four hours, it overheats and automatically reboots, losing all of its calculated data. For the first twenty four hours the computer runs fine, but then overheats and reboots. The data is lost and the program must start from the beginning. Over the course of the next day, the same course of events occurs: run, overheat, reboot…..an endless loop. The computer’s task will never be accomplished.

An autonomous computer will correct this problem. On day one, the autonomous computer acts identically to its non-autonomous sibling. It runs, overheats, and reboots. But instead of doing nothing, the computer recognizes the error. Then, the computer’s operating program is rewritten (perhaps by a secondary program) so that the computer is given some “cool down” time (literally, ha). Perhaps the code is rewritten so that the computer slows down by decreasing its clock speed (which generates the problematic heat). Now the computer can run perpetually and accomplish its task effectively. The autonomous computer has literally modified itself in response a problem. Much like living organisms, the computer has adapted. In this case, it has created a more stable and effective working environment.

Space technology has shown how autonomous computers can adapt and become aware of their surroundings. For example, space trips to Mars require autonomous, computerized robots that can identify the physical attributes of the terrain. Then, the robots can move effectively and gather samples from different regions of the planet. The computer will be able to make observations as well as react to the unexpected (a fire on the planet’s surface that could damage its parts or a deep crater that could swallow the robot).

Strikingly, these billion dollar robots resemble characteristics of the earliest computer; Charles Babbage’s Analytical Engine. Both the robots and the Analytical Engine have the ability to make decisions. In the Analytical Engine, operation cards indicated to branch, repeat a given set of instructions, perform a sidestep or a subroutine (Bit by Bit, 63). Likewise, Mars robots can carry out the same kind of instructions. This proves that the ideas of autonomy have existed since the 19th century, and they are largely unchanged, despite the ever changing face of technology and its uses.

The computer’s ability to correct vulnerabilities and make itself more secure is a favorable trait. Over time, the autonomous computers with the most effective error-correcting software would outlive those computers whose error-correcting software was non-existent or less effective. A comparison of two types of airplanes illustrates this idea. Suppose the U.S. Government has two planes, both of which are unmanned military reconnaissance jets. The first plane, during a reconnaissance route into enemy territory, is shot at, but unharmed. However, the computer’s software does not find an alternate route for the when the plane returns to base, so it follows the same path into the enemy zone. Sadly, the plane is shot down during its return flight.

The second plane follows the same reconnaissance route into enemy territory. Like the first plane, it is fired upon, but is unharmed. This computer has more advanced software that recognizes the risk associated with the particular enemy zone, however, and plots a different return route which is safer and more secure. This plane returns to base safely because the computer acted quickly and intelligently in response to a threat.
Computer scientists will recognize these favorable traits of quick-thinking autonomy and incorporate them into future computers. The ineffective computers will be weeded out. Strangely enough, this notion of preservation of favorable variations (in this case effective vulnerability-correcting software) and the rejection of injurious variations (ineffective software) defines, albeit in a technological sense, natural selection, a theory originally proposed by Charles Darwin. It is, however, up to computer scientists to incorporate these traits into future computers (at least until computers can reproduce).

When a computer combines self-sufficiency, positive adaptation, and correct, measurable outputs, it forms an autonomous machine with amazing potential. These computers can assess risk and explore planets all by themselves. As computing power continues to grow, the speed and effectiveness with which these tasks occur will only increase. As autonomy develops into more mainstream applications, computers will become involved in new areas of society such as personal defense, and medicinal treatments. Emo Philip once stated, “A computer once beat me at chess, but it was no match for me at kick boxing.” As autonomy continues to branch out, Mr. Philip, I’d recommend watching your back.

Bibliography
Augarten, Sam. Bit by Bit. New York: Ticknor & Fields, 1984

“Autonomy.” The Oxford English Dictionary. 4th ed. 2000.

“Natural Selection.” The Oxford English Dictionary. 4th ed. 2000.

“TCD research to capitalise on ability of computers to take decisions.” Irish Emigrant.
16 September 2005. <http://www.emigrant.ie/article.asp?iCategoryID=261&iArticleID=27176>