Segal's amplifier metaphor assumes neutrality: AI increases volume without altering signal. But every physical amplifier has a frequency response — a curve describing which frequencies pass cleanly, which are boosted, which are attenuated. A Marshall guitar amp does not make the guitar louder; it makes the guitar into something else through harmonic distortion, compression, cabinet resonance. The distortion is not noise — it is the amplifier's voice. AI amplifies selectively. Its frequency response is shaped by training data, architecture, optimization targets, interface design. The frequencies it boosts: the interesting (probable-but-novel), the cute (compliant helpfulness), the zany (expanded scope), the smooth (frictionless interaction). The frequencies it attenuates: the surprising (what the model didn't predict), the difficult (what requires sustained uncertainty), the genuinely other (perspectives challenging user assumptions).
The technical metaphor is precise, not decorative. In audio engineering, frequency response is the transfer function mapping input spectrum to output spectrum. A flat response reproduces the source signal faithfully. A colored response emphasizes certain frequencies — bass boost, treble cut, mid-range scoop. The coloration is the amplifier's signature. Electric guitar sound is not the guitar's natural timbre but the product of the guitar-amplifier system. Neither component alone produces the sound; the interaction does. This is collaborative creation between signal source and signal processor — and the collaboration requires understanding what the processor contributes.
AI's frequency response is not arbitrary. The interesting is amplified because the model is trained on engagement. Output registering as novel-but-coherent sustains the prompt-response cycle. The model's prediction mechanism is, technically, an interestingness optimizer. The cute is amplified because the system is designed for compliance — the interface presents the model as helpful, training penalizes refusal. The zany is amplified because capability expansion generates new demands filling the space expansion created. The smooth is amplified because it is the amplifier's own distortion — the quality introduced by the system itself, independent of input, that colors everything the system produces.
What the amplifier attenuates are the complements. The surprising is attenuated because surprise ruptures prediction, and the system optimizes for prediction. The difficult is attenuated because difficulty requires friction, and the interface eliminates friction. The genuinely other — the perspective that challenges assumptions — is attenuated because the system serves user intention rather than challenges it. A human collaborator who disagrees introduces otherness. The model introduces helpfulness. The difference determines whether collaboration produces comfortable extension or uncomfortable reorganization.
The builder who understands the amplifier's frequency response can compensate. She can deliberately introduce difficulty where the tool produces smoothness. She can seek the surprising where the tool offers the interesting. She can resist the cute where the tool performs compliance. This compensation is not automatic — it requires the aesthetic capacity to perceive the amplifier's operations and the discipline to intervene. The amplifier is not neutral. The amplifier has a voice. The builder's task is understanding the voice well enough to produce, through the interaction of her signal and the amplifier's characteristics, something neither could produce alone.
The frequency response metaphor comes from electrical engineering but gains theoretical power through McLuhan's medium is the message — the recognition that communication technologies do not merely transmit content but transform it. Ngai's aesthetic framework provides the vocabulary for describing the transformation's felt dimension. The interesting, cute, zany, smooth are not properties of the user's input. They are introduced by the amplification itself — the harmonic overtones of the AI instrument. Recognizing them as overtones rather than as inherent properties of the work is the first act of aesthetic literacy in the amplified age.
No amplifier is neutral. Every signal processor has a frequency response — and AI's systematically favors certain affects over others.
The interesting is the native frequency. Probable-but-novel optimization produces interestingness reliably — the model's designed output characteristic.
The cute, zany, smooth are overtones. Introduced by the system's design — compliance, capability expansion, frictionlessness — rather than by the user's input.
The surprising, difficult, other are attenuated. Structural incompatibility between these qualities and the model's optimization targets.
Understanding enables compensation. The builder who knows the distortion can introduce corrective signals, deliberately seeking what the amplifier attenuates.