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. Yet 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 people 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.