Slashdot Mirror
Researchers Built an 'Online Lie Detector.' Honestly, That Could Be a Problem (wired.com)
A group of researchers claims to have built a prototype for an "online polygraph" that uses machine learning to detect deception from text alone. But as a few machine learning academics point out, what these researchers have actually demonstrated is the inherent danger of overblown machine learning claims. From a report: When Wired showed the study to a few academics and machine learning experts, they responded with deep skepticism. Not only does the study not necessarily serve as the basis of any kind of reliable truth-telling algorithm, it makes potentially dangerous claims: A text-based "online polygraph" that's faulty, they warn, could have far worse social and ethical implications if adopted than leaving those determinations up to human judgment.
"It's an eye-catching result. But when we're dealing with humans, we have to be extra careful, especially when the implications of whether someone's lying could lead to conviction, censorship, the loss of a job," says Jevin West, a professor at the Information School at the University of Washington and a noted critic of machine learning hype. "When people think the technology has these abilities, the implications are bigger than a study."
"It's an eye-catching result. But when we're dealing with humans, we have to be extra careful, especially when the implications of whether someone's lying could lead to conviction, censorship, the loss of a job," says Jevin West, a professor at the Information School at the University of Washington and a noted critic of machine learning hype. "When people think the technology has these abilities, the implications are bigger than a study."
If this app was put online labeled as "Fred's AMAZING online truth teller" with the usual ads for bikinis, penis enlargement, crockery, the latest Chevy, I don't think you have anything to worry about in terms of it causing problems.
If it's part of the Google home page or comes up automatically when submitting documents to the IRS, I think there is a great deal of concern regarding whether or not people believe the results are accurate.
Mimetics Inc. Twitter
All ML algorithms have an error rate. Its baked into the design. ML researchers talk about error rate all the time. There is even a term, 'irreducible error' in ML that refers to data points that can never be classified correctly by a specific algorithm. Its a mistake to completely trust what a computer database tells you because the wrong data could have been input or bugs could have changed that data in a weird way. Its all those risks plus the error that comes from the ML algorithm. The way to get around this is to have multiple algorithms "vote" but even then there is still an error rate. The error rate can be double digit % or lower than 1% but its always there. And all of this is on top of the risk of bad data just like a DB gives you. Garbage in/garbage out is a real principle. Trusting this stuff is tricky but the bar isn't perfection. Its better than a skilled human. And since I don't really trust a "skilled human" in lie detection, why on earth would I trust a ML algorithm that at best is only marginally better than that and could be far worse.
"Those that start by burning books, will end by burning men."
But when we're dealing with humans, we have to be extra careful, especially when the implications of whether someone's lying could lead to conviction, censorship, the loss of a job,
We already hit people with all that stuff just for typing stuff online that we don't like, to say nothing of whether it's true or not!