A recent study found that AI algorithms identify Deep Fake images quite a bit better than humans. Humans identified them basically at 50%, which is just chance level, while an AI correctly distinguished between the images as high as 97% accuracy. However, humans still identify Deep Fake videos better than the AIs that were used in the study - the AIs were only around chance level, but humans were correct nearly two-thirds of the time.
If one is into computer science, psychology, media science, or some other fields, something along these lines could be a research project. Perhaps a study of identifying real images from deep fakes as a function of age, or gender. Or as a function of race or where one lives (city vs suburb vs rural). Are there any significant differences between who can identify them better, or is it chance level for all groups? Are certain types of images easier to distinguish than others? If one can develop a good sample of real and faked images, perhaps different studies could be developed by students - be creative!
With this being an election year, everyone is anticipating the use of Deep Fakes will spike. We should be working in unison with AI algorithms to hopefully weed-out as many of these as possible, as it will be an ongoing battle as long as these technologies exist. Let's try to maintain as much reality in our lives as possible, as we continue on into the age of mis- and dis-information. It is becoming more challenging to pick out real from fake, to be sure.
Which image below (of a NU professor) is real, which is fake???
No comments:
Post a Comment