Is every human voice and fingerprint really unique?
Barclays, the UK bank, is to replace the password system on its phone banking service with personal voice recognition. "Unlike a password, each person's voice is as unique as a fingerprint," said Steven Cooper, Barclays' head of personal banking. Yet the reality is we have no idea whether either fingerprints or voices are unique at all.
I don't mean to say that it is impossible that each of us has a physical feature that is unique. The problem is that we have no way of knowing whether any of us has such a feature – far less that we all do.
Short of testing all of the 7.4 billion human beings alive today, which is a practical impossibility, we must rely on sampling of some sort. Testing samples of human populations has allowed us any number of vital medical advances. It is great for giving us information about the frequency of particular phenomena within a specified population – blood types, for example.
Yet sampling cannot establish uniqueness. Sampling could no more inform us that one individual has a particular set of fingerprints than it could inform us how that person will vote. And even if you managed to compare a person's fingerprints with every other living human being, it wouldn't establish their uniqueness. What of everyone who is dead? What of all those who are yet to be born?
Neither can you get around this problem with abstract reasoning. If an X will cause a Y then logically whenever there is an X, there will be a Y – or in this case, my fingerprints are caused by a set of factors peculiar to me; therefore everyone has a set of fingerprints caused by their own set of factors. That's logical, yet it does not follow that whenever there is a Y there will be an X – or that a set of fingerprints can only be produced by the one set of unique factors. This is because the same effect might have had different causes on different occasions.
The evidence barrier
The widespread and strongly held belief in the uniqueness of human voices and other physical features characterises and exacerbates a chronic general problem: people misunderstand the nature and significance of quantitative scientific evidence. This is particularly relevant to criminal trials. There is a danger that accused people are not given fair trials because jurors, judges and other court officials put too much weight on certain forensic evidence.
The notorious – later quashed – conviction of Sally Clark, an English solicitor, for the alleged murder of her two babies illustrates well the problem. The case against her turned on the evidence of an expert witness, Sir Roy Meadow, who argued that it was highly improbable that two of her babies could have been the victim of natural cot deaths.
Clark was later exonerated by an appeals court after serving three years in prison, but died four years later. Her family said in a statement that she had never recovered from the miscarriage of justice. Meadow was criticised and temporarily struck off for getting the numbers wrong in his assessment of the statistical improbability, but this misses a more fundamental objection to this sort of argument: using statistical unlikelihood in a general forward-looking way is one thing; using it retrospectively with regard to particular instances is quite another.
If you buy a ticket for the lottery, the chances of winning the big prize are about 14m to 1. You might therefore be justified in regarding that as evidence that you are unlikely to win, and not buy a ticket as a result. Yet after the draw is made and Ms X of Glasgow is announced in the newspapers as the winner, the known unlikelihood of winning is obviously not evidence that she did not win. She did win. Unlikely things do happen.
The way forward
What do these insights mean in practical terms? People might well argue that even with our limited sampling of human voices, we have good reason to suspect we are very unlikely to come across two different people who have identical voices, even if we could never discount the possibility. Fine. Let us say that.
Human voice patterns or iris recognition need not be assumed to be unique to be useful tools for protecting private access to our bank accounts. In the same way, fingerprints need not be assumed to be unique to be useful in courts.
Dangers and misunderstandings occur when too much is claimed for such techniques. They are of limited usefulness. They should not be relied upon totally – as overriding evidence or for security systems on their own. They must be relied upon only as part of a wider case or system of checks.
This story is published courtesy of The Conversation (under Creative Commons-Attribution/No derivatives).