A system based on the human immune response is being developed to tackle Internet fraud.
Just as our own immune systems distinguish between 鈥渟elf鈥� and 鈥渘on-self鈥� to recognise harmful invaders, the electronic version will be able to tell legal transactions from fraudulent ones.
The system relies on thousands of randomly generated 鈥渄etectors鈥�. Each of these little computer programs recognises a specific kind of transaction. The detectors travel around computers and networks looking for rogue invaders, just as white blood cells do in our bodies.
Advertisement
But before they are let loose, the researchers challenge them with all kinds of legitimate online deals.
鈥淲e get rid of anything that detects 鈥榮elf鈥� by matching the random detectors against 鈥榥atural鈥� or 鈥榞ood鈥� transactions,鈥� says Richard Overill of King鈥檚 College London. 鈥淎ny detector that matches a bona fide pattern is killed off, so only those that recognise fraud remain.鈥�
Overill thinks the system could be used to police any interaction on the Internet. Banks could use it to monitor account transfers, and it could help government departments tackle benefit fraud, he says.
Internet fraud costs 拢20 billion a year in Britain alone, he says. 鈥淣o one wants to admit how big the problem is.鈥�
Ian Brown, a cyberfraud expert at University College London, thinks the system could be useful for monitoring credit card transactions. But he thinks encryption to make electronic business more secure in the first place will also have to be part of the solution.
When a detector finds a match, the alarm is raised and a human looks at the transaction to check whether it really is fraudulent. Detectors that recognise legal patterns are removed, while any detector that identifies fraud is multiplied.
We don鈥檛 get measles twice because antibodies against the infection stick around in our bodies. In the same way, the artificial immune system (AIS) is ready for a rapid response if the same approach is ever tried again.
鈥淲e want to build our immune system so that once it has seen fraud it will remember it,鈥� says Overill.
The detectors are also allowed to mutate and evolve over time so that the system can always adapt and update itself to combat new types of fraud.
Conventional fraud detection systems 鈥� including those based on neural networks 鈥� rely on programmers being able to describe what a fraudulent transaction looks like. But the differences between legal and illegal actions can be subtle and hard to define.
The advantage of an AIS is that it starts off with a random population of detectors. Once faulty detectors have been removed, those remaining should be able to recognise everything else 鈥� including types of fraudulent transaction that haven鈥檛 even been thought of yet.
Just as our bodies will fight many new infections, the AIS should be able to mount an 鈥渋mmune response鈥� to most kinds of attack that criminals can think of.
Overill will develop the technology for detecting fraud over the next three years, initially using historical data from Britain鈥檚 Post Office. The Post Office will be the first user of the new system.
A full-length feature on artificial immune systems will appear in New 杏吧原创 magazine.