bet365 live casino games Sandal said，When people talk about the credibility of artificial intelligence，They talk about the performance of artificial intelligence and it reflects reality、Ability to show the truth。Only developers use the correct label、Good quality data repeatedly train artificial intelligence algorithms，It has the opportunity to gain the trust of the public。In fact，Social concerns about artificial intelligence credibility，More concerns about the data used in training artificial intelligence algorithms。

The quality evaluation criteria for clarifying the training data from non -professionals are not an easy task。Sandal and others think，Improve the transparency of the data labeling process、Discover the accuracy of the data label，It will help the public to establish an objective artificial intelligence credibility evaluation system。

To prove this assumption，Researchers recruited bet365 Play online games 430 experimental participants，Let them interact online with an artificial intelligence system。The participants were informed，This artificial intelligence system can distinguish the facial expression in social media images。The development team collected nearly 10,000 facial images，Joy is posted to each picture, respectively、Sad、Angry、Fear、Surprise、Tags of disgust or neutral emotions。Last，The development team will take the labeled data set to train the artificial intelligence system。Participants are also informed，More than 500 people participated in the labeling work。other，Researchers at the experiment deliberately modified the label，two different labels have been produced: in one case，Tags accurately describe the facial emotions in the image; in another case，Half of the number of labels does not match the facial emotions displayed by the image。

bet365 Play online games Next，Experimental participants were randomly assigned to three artificial intelligence performance testing team: artificial intelligence without performance、Performance with prejudice、There is no prejudice。Participants in the latter two groups observed the entire process of artificial intelligence processing mission。where，Seeing the second group members，The artificial intelligence system completes the emotional classification task of white images with 100%accuracy，and the emotional identification accuracy of black images is 0%。When they realize that the artificial intelligence system has racial prejudice，The degree of trust in it has also been significantly reduced。but，Sandal and others also observed，The emotional connection between the second group members and the artificial intelligence and the willingness to continue to use the system in the future did not decrease。

Sandal and others invented bet365 best casino games the word "training data credibility" to describe whether the data used to train artificial intelligence models is credible、Reliable。Sandal thinks，The development team has the responsibility to let users make their own judgment on the credibility of the artificial intelligence system，instead of persuading them to blindly trust the artificial intelligence system。Researchers also suggested that developers and designers continue to disclose the specific process of marking data to users，Cultivate users' ability to identify high -quality labels，Establish a sense of supervision of artificial intelligence。This requires developers to conceive a more creative training data sharing mode，While improving user participation, reducing the user burden，Avoid misleading users。

(Liu Yuwei/Compilation)