The Stack On Software and Sovereignty · 第 278 页

The suite of assistants they designed were not meant for one imaginary User for whom Siri ’ s personality is “ universally ” appropriate. Instead they developed idiosyncratic and culturally specific agent personalities. Their menagerie included agents for specific ethnic experiences, familial niches, curatorial perspectives, and User predicaments. Each would be finely tuned to a certain User but irrelevant to most others. Second, the agents were in different ways unreliable. Their personalities were defined by contradictions and mixed motives, much like human personalities. They may be manipulative or insecure, prejudiced or forgetful. The design issue here is not that broken tools are better, but that in designing interfaces that resemble a familiar mind, it may well be that vulnerable and limited artificial personalities invite and support deeper identification and loyalty from human Users than generic and predictable ones . ⁴⁴ The fifth edition of Diagnostic and Statistical Manual of Mental Disorders may prove essential reading for programmers of the next decades full of Siris. ⁴⁵ If Siri is just the tip of a AI interface iceberg, then to presume that “ she ” should be so defined by C-suite master-slave small talk is already to presume too much. The full spectrum of weirdness, based on actors passing information into and out of one another, is a richly reticulated space, and approaching it as such makes those clich é d relationships moot. We need to ensure that AI agents quickly evolve beyond the current status of sycophantic insects, because seeing and handling the world through that menial lens makes us, in turn, even more senseless. Something else needs to sideline the vestigial, embarrassing anthropocentrism that brings the false flattery of passive (or passive-aggressive) assistants. In fact, the stage management of the human User as somehow commanding the work of the App is already, we may conclude, an unnecessary alibi protecting the essential opposite effect (that the mammal User is only a provisional mechanism for dragging gigaflop tracking devices through the avenues of cities and for remonetizing these routes as the spatial career of algorithmic capital and its successors).

Put differently, bots are Users already, and we are already well past the point in which more than 50 percent of Internet traffic is generated by nonhuman users. ⁴⁶ As far as The Stack may see it, humans are a privileged User, but a shrinking minority, even as the next billions of them come online over the next decade. Similarly, the role for HFT algorithms is indeterminable, even though they do not speak at us they do speak for us, and we can assume that if similar forms of weaponized mathematics become more normative, then the plural and partial nature of any individuated human User subject interest and position may be that much harder to keep straight. In the ongoing technologicization of intelligence, we see cute slippages of position between humans and machines. For example, the original mechanical turk in the eighteenth century was a chess-playing machine, apparently an automaton exhibiting human intelligence, but in fact operated from afar by a human User. Today Amazon Mechanical Turk restages this arrangement, not just for chess but for any menial task the User can devise. Behind the browser are at least half a million “ workers ” who complete piecemeal tasks for micropayment. ⁴⁷ We see not AIs appearing as if they were human, but humans appearing as if they were AIs. As it ’ s been since Karel Čapek ’ s “ universal robots ” introduced on the stage in 1921, the robot not only mimics the human, but provides a portrait of the human as an object viewable to itself from the outside, and with the human negotiates an ongoing dance of reciprocal idealization. And in itself, this may signal less a mechanistic subsumption than a wider berth for incentive mutual identification between co-Users.

278

User Layer