While brands and the ad industry may tremble, consumers rejoiced earlier this year when Princeton researchers announced the creation of AI-backed perceptual ad-blocking technology.
The concept uses a machine-learning model that detects the images on pages with ads and assesses which parts of the page are ads and blocks them while leaving the rest of the content intact.
However, web users might want to put the champagne cork back in the bottle as a recently released paper from Stanford and CISPA Helmholtz Center researchers titled “Ad-versarial: Defeating Perceptual Ad-Blocking” (yes, they went there) looks at countermeasures that also use machine learning and how they could set a high a hurdle for ad blocking technology to overcome.
As the abstract to the study states:
“Perceptual ad-blocking is a novel approach that uses visual cues to detect online advertisements,” before going on to warn that “perceptual ad-blocking engenders a new arms race that likely disfavors ad-blockers.”
For their study, the researchers revealed a set of eight techniques to generate just slightly modified ads that could fool the perceptual ad-blocker.
Without getting too wonky, some techniques included something as simple as modifying a few pixels in the logo used by ad-blockers to spot an ad or simply overlaying a transparent image to fool its AI ad-versary.
To further demonstrate the countermeasures, the team showed that the perceptual blocker’s model could even be tricked into to wrongly identifying a website’s non-ad content as itself being an ad – which it would then block – while leaving all of the actual ads unblocked.
Long story short: As the ad-blocking war rolls on, the Stanford, CISPA Helmholtz researchers suggest that perceptual ad blocking technology is in a “precariously disadvantaged position.”
“As long as robust defenses to adversarial examples elude us, perceptual ad-blockers will be dragged into a new arms race in which they start from a precariously disadvantaged position— given the stringent threat model that they must survive.”
You can read the paper here.