untitled design

New technique helps AI detect when humans are lying

Researchers developed a Artificial Intelligence training software which helps detect lies in economic contexts. The objective is to prevent the tools used by these companies from making predictions based on distorted information about users' financial lives.

Assuming that humans can lie when providing personal information that affect their economic life, the program was designed so that AIs learn to detect these frauds.

At times of applying for a mortgage or setting insurance premiums, e.g. users have a tendency to change information crucial for calculating these values ​​and the tools used by companies use statistics that take into account the information provided by users.

The set of parameters developed by the researchers makes the AI ​​teach itself to predict moments when humans tend to lie to obtain financial benefits. To do this, it takes into account the economic incentives that person would have by lying about their personal information.

O program has been made publicly available so that AI developers can use it in their code. The idea is to apply it in contexts where financial decision forecasts are made.

Scholars still want to improve the training tool to detect “small lies” and know how to differentiate them from “big” ones. This would help statistics show even more accurate data about users' financial condition.

“At some point, if we make AI smart enough, we might be able to eliminate these incentives [de mentir] completely”, declared Mehmet Caner, co-author of the study.

Source: CNN Brasil

You may also like

Get the latest

Stay Informed: Get the Latest Updates and Insights


Most popular