Skip to content

A Pipeline to Truth? Fighting Absenteeism with Voice Analysis

The Daily Mail warns that bosses want to
use over-the-phone lie detectors to catch out workers pulling a sickie
. The issue is the new generation of voice
analysis software that listens in when someone calls in sick, and prompts the person
talking the call on whether the person is suspiciously stressed. Y
et another step towards
1984, a
great way of
saving money and improving the truthfulness of people, or a double deceit?

Whether lying can ever be justified has
long been a favourite discussion topic in ethics. Famously Kant said that lying
is wrong under all circumstances, even when a murderer at the door asks if the
innocent victim is in your house
. Other philosophers have been more accepting
of deception under some circumstances, for example when it would have
beneficial effects or reduce harm. But let us for the sake of argument assume
that it is always wrong to claim the benefits of a few sick days when not
actually being sick.

It would seem that any increase in the
ability of the people at the job to detect deception would be good since it
would enforce virtuous behaviour. However, the problem is not whether it is
right to detect lies, but whether it is OK to lie in order to reduce lies.

Voice stress analysis is based on the idea
that infrasounds too low to be heard in the human voice will vary depending on
how stressed the person is; this generally does not work over phone
lines, which clip sounds too much. Other approaches try to find other patterns
to detect emotions. There are much debate about whether the methods actually are
accurate, with enough vested interests among polygraph associations and voice
analysis companies to keep it going for years.

In general the accuracy does not appear to
be that great outside narrow experiments
where it is hard to get enough stakes
to make the lying plausible. In a real-world test with recently arrested
who were asked about recent drug use and then subjected to a urine test,
two of the most popular voice analysis programs failed miserably. Even if the accuracy of the test was good there would also be a sizeable
number of false positives where innocent people would be accused of lying.

Besides these problems the key uncertainty
of what causes the stress remains; just like the polygraph is unable to tell
whether somebody becomes stressed because they are lying or just stressed by a
nasty accusation these systems are likely to be sensitive to other factors –
and different people become stressed by different things. A study of real
estate sales agents asked about ethical code guidelines
showed that some become
stressed while following the code, others did not.

However, it is known that people lie less
when connected to a fake lie detector, the “bogus pipeline effect”. If you think someone has a “direct
pipeline” to your mind, you will likely lie less since you think you will be
found out. As noted in the above jail paper, from the perspective of the police
the device “works” if they can get more confessions (by telling subjects that the
machine knows they are lying). It does not matter much to the police if they get more
false confessions, a relatively common occurrence.

The real ethical problem with analysing
voice to reduce absenteeism is that it is likely to work due to the deception
that the analysis will reveal lies. The users of the system may or may not
believe in it, but it would be in their and the manufacturers interest to
spread the impression that it is effective.

If we agree that lying in order to claim
the benefits of a few sick days when not actually being sick is immoral, it
would seem that spreading the deception that lies can be accurately detected
would be equally or more immoral. Falsely claiming to be sick is akin to
breaking a promise to the job. Falsely claiming to be able to reliably detect
internal mental states changes the relationship between people to be built upon
deception – and this occurs even in the majority of phone calls where the caller
is genuinely sick. A Kantian would say the system disrespects the human dignity
of callers. A utilitarian would likely conclude that the reduction in lying
about sick days is not worth the reduction of respect and trust between
employees such a system would cause.

What would happen if the system were
actually 100% accurate? At this point all bets are off, since society would
have to function entirely differently (see this fictional treatment); presumably it would be moral to use it to
catch absenteeism. But for most statements a 100% accurate detection of deceit
is not possible even in principle. Our memories and self-images are fallible
and self-serving. The borders of our categories (such as “sick” and “healthy”)
differ from person to person. Unless forced by strictly defined questions to
answer in exactly agreed ways, reducing us to little more than cogs in the
truth enforcement machine, deception detection will never be 100% accurate. The
perfect lie detector would be more akin to a mother, who when her child claims
to be too sick to go to school both see through them and understand them well
enough to reach a proper decision that respects them.

Share on