Love them or hate them you can’t escape the fact that you rely on computers every day of your life. But can you trust the information that they give you?
Sam Relton, a Postdoctoral Research Assistant in the School of Mathematics; doesn’t necessarily think so:
Computers are not as smart as we think. Whenever you do anything with a computer it makes small errors.
But computers make fewer errors than humans, so is it really a problem?
Take, for example, the case of the wonky photocopier that used the wrong algorithm and occasionally transformed a 6 into an 8*. Surely this is largely insignificant? But what if the numbers on a construction plan were changed in this way? The effects of a single digit variation could be devastating.
One example of where computational errors can have a profound significance is the recent eHealth trend: making decisions that affect someone’s chance of surviving cancer, for example, are not easy. The doctor has got to take into account the patient’s age, medical history and current state of health. To research the ins and outs of different treatments, to look at the medical records of patients with similar cancers, and then compare all the possible outcomes with the possible treatments and the trajectory of the particular cancer becomes just too complicated: there is a mass of information and too many variables. Computers become necessary to do anything meaningful with such enormous datasets.
Mathematicians like Sam, within the field of numerical analysis, help to design and solve mathematical models, providing answers to complex and intangible real world problems. In this way, a computer can be used to give the doctor an accurate view of the best course of treatment for their cancer patient.
But that is not the end of the story; because both computers and humans can increase the margin of error in a dataset, as Sam explains:
I am doing some work with a nuclear physicist who is involved in the safety of some proposed nuclear reactors. He does some modelling too see if somebody walking around one of these nuclear reactors for prolonged periods of time is going to be at risk of radiation poisoning etc. As part of this he has a large table of data (a matrix) and each element is gathered by a laboratory experiment, meaning that these values are not exact. That is where my kind of research comes in: to make sure that these small errors that people are making in the laboratory aren’t going to change the outcome of the model.
So back to the original question; can you trust your computer?
Yes you can trust your computer because there are researchers like Sam; whose work removes any ambiguity there might be about what the computer tells the doctor, the nuclear physicist and the engineer.
* The error was finally spotted by an eagle-eyed PhD researcher who documented it on his blog.