Deepfake: synthetic (合成的) media, including images, videos, and audio, is generated by Al technology to show something that does not exist or events that have never occurred.
Examples of deepfakes have been widely spread, including a video of Facebook CEO Mark Zuckerberg giving a speech about his company's plan, and a video of Elon Musk dancing and talking about the power of dreams, etc.
It's easy for AI to produce such deepfakes using two different deep-learning algorithms (算法): one that creates the best possible clone based on a real image or video and another that detects whether the copy is fake (伪造的) and, if it is, reports on the differences between it and the original. The first algorithm produces a synthetic image and receives feedback on it from the second algorithm and then adjusts it to make it appear more real; the loop is repeated as many times as it takes until the second algorithm does not detect any false imagery.
Deepfakers often have evil motives, including creating misinformation and generating confusion. They tend to demean, terrify, and annoy, and have targeted not only celebrities but ordinary citizens as well.
Most of the academic research surrounding deepfakes focuses on the detection of huge amount of deepfake videos emerging online. One detection approach is to use algorithms to identify inconsistencies in deepfake videos. For example, an automatic system can examine videos for errors such as irregular blinking patterns of lighting. However, these approaches have been criticized because deepfake detection is characterized by a "moving goal post" where the production of deepfakes is changing and improving while detection tools are always on the way of catching them up.
However, education and medicine are two of the fields that may benefit from deepfake technology. In the classroom, historical speeches could be deepfaked to offer immersive and engaging lessons. In health care, it can improve the accuracy with which tumors (肿瘤) are spotted, making them easier to treat. Its use also permits using synthesized data instead of that from real patients to avoid privacy concerns.