Disrupting Surveillance Capitalism
A practiced overview of around of the techniques we've looked at over the years too next upward on yesterday's "Potemkin AI: Many instances of 'artificial intelligence' are artificial displays of its ability too potential", demystifying AI too showing how it is all the same pretty dumb—compared to what's coming.
From Logic Magazine:
Monkeywrenching the Machine
Silicon Valley’s surveillance-based concern model relies heavily on machine learning. But amongst the right techniques, nosotros tin forcefulness out resist the enclosure of our lives for turn a profit too disrupt the disruptors.
This slice is too available inwards audio from our friends at Curio.io.
makeinternetnoise.com. Because your information is non exclusively used to brand assumptions close you, exactly close other users amongst similar browsing patterns, you lot destination upward interfering amongst the algorithm's conclusions close an entire grouping of people.
Of course, the effectiveness of this tactic, similar all others described here, increases when to a greater extent than people are using it. As the CIA's Simple Sabotage Field Manual explains, "Acts of uncomplicated sabotage, multiplied yesteryear thousands of citizens, tin forcefulness out live on an effective weapon...[wasting] materials, manpower, too time. Occurring on a broad scale, uncomplicated sabotage volition live on a constant too tangible drag on...the enemy."
Attacks of this sort—where nosotros corrupt the preparation information of these systems—are known every bit "poisoning" attacks.
The Pathological too the Perturbed
The other category of adversarial machine learning attacks are known every bit "evasion.” This strategy targets systems that accept already been trained. Rather than trying to corrupt preparation data, it tries to generate pathological inputs that confuse the model, causing it to generate wrong results.
From Logic Magazine:
Monkeywrenching the Machine
Silicon Valley’s surveillance-based concern model relies heavily on machine learning. But amongst the right techniques, nosotros tin forcefulness out resist the enclosure of our lives for turn a profit too disrupt the disruptors.
This slice is too available inwards audio from our friends at Curio.io.
makeinternetnoise.com. Because your information is non exclusively used to brand assumptions close you, exactly close other users amongst similar browsing patterns, you lot destination upward interfering amongst the algorithm's conclusions close an entire grouping of people.Of course, the effectiveness of this tactic, similar all others described here, increases when to a greater extent than people are using it. As the CIA's Simple Sabotage Field Manual explains, "Acts of uncomplicated sabotage, multiplied yesteryear thousands of citizens, tin forcefulness out live on an effective weapon...[wasting] materials, manpower, too time. Occurring on a broad scale, uncomplicated sabotage volition live on a constant too tangible drag on...the enemy."
Attacks of this sort—where nosotros corrupt the preparation information of these systems—are known every bit "poisoning" attacks.
The Pathological too the Perturbed
The other category of adversarial machine learning attacks are known every bit "evasion.” This strategy targets systems that accept already been trained. Rather than trying to corrupt preparation data, it tries to generate pathological inputs that confuse the model, causing it to generate wrong results.
The spam filter attack, where you lot fox an algorithm into seeing spam every bit ham, is an illustration of evasion. Another is "Hyperface," a collaboration betwixt Hyphen Labs too Adam Harvey, a particularly designed scarf engineered to fool facial recognition systems yesteryear exploiting the heuristics these systems purpose to position faces. Similarly, inwards a recent study, researchers developed a brace of spectacles that consistently drive a state-of-the-art facial recognition organization to misclassify faces it would otherwise position amongst absolute certainty....MORE
No comments