Cognitive distillation

✨

From tkwa.me

Music
Learning the Tide · SongLyrics

Cognitive distillation develops selected abilities in one mind through training based on one or more other minds. A teacher can supply examples of how they recognize a pattern, make a judgment or respond to a situation; a learner develops structures they can use in their own thinking. The practice is used both to prepare starting minds and to help existing people acquire abilities.[3][6]

Early post-merger studies included a 2032 river-surge experiment by engineer Asha Dey and AI forecaster Iso. In a 2356 survey of 18,000 adults in selected Earth–Moon communities, 29% reported undertaking a program to acquire an ability by distillation while already active during the preceding 12 Earth calendar months.[2][5]

A learner touches colored bands on a large wooden frame while a human teacher points to a model and a ceramic-faced teacher stands beside the frame.
A learner and two teachers explore a pattern using colored bands and a model.

History#

In 2015 Geoffrey Hinton, Oriol Vinyals and Jeff Dean described distilling the knowledge in a neural network. They trained a model using outputs from other models, including ensembles, and evaluated it on image recognition and an acoustic model.[1]

In 2032, Dey trained with simulated river surges and responses supplied by Iso. Before training, Dey chose the less-flooding of two barrier timings in 64% of 400 simulated cases; afterward, they chose it in 91% of a separate set of 400 cases drawn from the same distribution. Neither evaluation set was used for training. The result measured that particular decision task, and the study also recorded Dey's subsequent work with real monitoring data.[2]

During the 2030s, researchers improved mappings between differently organized minds and learned to select intermediate representations as well as observable responses. These methods helped combine contributions from several teachers. In 2041, neuroscientist Leena Rao, AI researcher Aven and composer Hani Vale included selected trained abilities in the starting mind of their child Neri, alongside an attributed memory from Rao.[3][4]

How training works#

A teaching set can contain situations, the source's responses, explanations and selected internal representations. In one common computational method, the learner produces answers to the cases; training software compares them with the chosen targets and adjusts trainable connections or an additional module. Repeated practice changes how the learner responds to new cases. Teachers can add examples in areas where the learner is uncertain.[3][6]

Cross-substrate calibration helps when the teacher and learner organize perception differently. Shared examples and feedback establish correspondences between their representations; a spatial pattern sensed across a garden, for instance, can be mapped to cues a learner with another body can use. A mapping is tested on further examples and adjusted as the participants discover differences in meaning.[3][7]

Evaluation can use cases withheld from training, as in the Dey–Iso study, and practical work in the learner's environment. Training requires physical computers, stored examples, power and time. An existing person may test a new module before incorporating it into their ongoing cognition; additional practice can adapt the ability to their own body and purposes.[2][3]

People and uses#

The teachers may be human-descended people, AI-origin people or designed people with several sources. Multi-parent synthesis can use distilled abilities from several contributors in a new starting mind. Neri later chose to study coastal sound and maintain a neighborhood hydrophone, developing interests beyond the material their parents supplied.[4][6]

Distillation also serves people who are already living. The 2356 Earth–Moon survey counted each respondent once and included active residents of simulated environments and people of human and AI origin. Its 29% figure covers those who undertook deliberate distillation training during the preceding 12 Earth calendar months, excluding material included in their initial design and ordinary teaching.[5]

An experienced garden person can offer examples of recognizing patterns of shade and moisture across a body. A learner can test the ability in another garden and adjust it to different sensors. Players of a walking orchestra, a shared articulated instrument that makes music as it walks, can train a newcomer on coordinated steps and practice a turn together, using its movement and sound to refine it.[6][7]

In amateur art, people may learn from a favorite musician so they can enjoy playing together, or cultivate an unfamiliar sense through which to compose. Some groups cherish skills developed over years of practice; others share new abilities before an evening’s improvisation and discover what they want to make with them. The pleasure of making something and recognizing another participant’s contribution helps sustain both kinds of gathering.[3][6]

Related forms of sharing#

A copy begins from a record of an existing mind, while training develops selected abilities from material supplied by teachers. Shared memory exchange can accompany training: a person may receive a teacher's recollection of a particular shoreline as well as learn to recognize tides. A memory can retain its contributor's perspective and be joined with the recipient's own later visit.[3][4]

A person living parallel lives can practice variations through several fragments and use selective synchronization to bring selected results together. People vary widely in ability, including minds around 250 on the Epoch Capabilities Index and others far above. Training can be fitted to a learner's capabilities and interests, and their later choices continue to shape what they become.[3][6]

See also#

References#

  1. ↑ Hinton, G., Vinyals, O. and Dean, J. (2015). Distilling the Knowledge in a Neural Network. arXiv:1503.02531.
  2. ↑ a ↑ b ↑ c Dey, A. and Iso (2032). Learning Surge Decisions from a Predictive Model.
  3. ↑ a ↑ b ↑ c ↑ d ↑ e ↑ f ↑ g ↑ h Sato, L. and Okonkwo, R. (2084). Cognitive Integration: Practice and Continuity. 2nd edition. Meridian Academic.
  4. ↑ a ↑ b ↑ c Laghari, S. and Sol, M. (2324). Families of Many Origins. University of Luna Press.
  5. ↑ a ↑ b Earth–Moon Social Survey Consortium (2357). Connected Lives: 2356 Survey Report.
  6. ↑ a ↑ b ↑ c ↑ d ↑ e ↑ f Adewale, R. and Sato, C. (2260). New Persons: An Atlas of Posthuman Beginnings. Meridian Academic.
  7. ↑ a ↑ b Nguyen, T. and Kline, H. (2348). Garden Bodies: Sensation, Growth and Care. University of Luna Press.