When learning either a primary or secondary language this process is incremental and advancements are not made in days but in months and years.  For example, a group of Japanese students was required to memorize a thousand new English words over a period of four months (Hosoda et al.2013).  The rate of transfer of such information from teacher to student was estimated to be 0.0006 bits per second (Tehovnik, Hasanbegović, Chen 2025), which is much slower than the throughput for human language production by an advanced speaker at 40 bits per second (or over a trillion possibilities per second, Reed and Durlach 1998).  Along with this low rate of learning, it was found that the synaptic connections between Wernicke’s area (in the temporal-parietal cortex) and Broca’s areas (in the lateral frontal cortex) were enhanced over the 4-month period, as evidence by increased myelination and synapse formation (Hosoda et al. 2013).

Since learning a language is slow, the feedback when using ChatGPT (or its equivalent) must also be incremental.  For example, if a student that is functioning at 20% correctness for a language (where chance is 1%) is then provided with text-feedback that is close to 100% correctness, there will be no learning—just plagiarism.  Thus, the feedback-text created by chatGPT needs to be closer to the level of the student (for a 20%-correct student the goal should be 25% correctness, for instance) so that he/she can appreciate the incremental change in the text by it being similar to the original, but en route to 100% correctness following many repetitions through rewriting after each bout of feedback (on this point see Footnote 1).  Also, since every human has a unique writing style—Ernest Hemingway can never be confused with James Joyce—the AI algorithm will need to create a %100-correct text that is unique per student, otherwise we will be creating replications of the same communication style, thereby preventing our kids from developing their endowed uniqueness.  Therefore, AI algorithms need to be customized for teaching rather than remaining a general-purpose algorithm for producing the best (but generic) text.

In short, AI as implemented thus far induces a regression to the mean of a massively absorbed data-set. The biological brain, on the other hand, generates excellence as it pertains to an individual’s custom-made data-set (Noble and Noble 2023), such that Einstein can never be Pelé (nor vice versa), but both are extraordinary examples of a life-time of learning.

Footnote 1: When a PhD student would hand in a thesis to Henry Kissinger, Kissinger would say: ‘Is this the best you can do?’ If the student said, ‘With a little more time I can do better.’  Kissinger would then send the student off to do better.  This process was continued until the student (and not Kissinger) was satisfied.  This is how excellence is created (based on EJT’s recollection of information on the late Henry Kissinger).

More Edward J Tehovnik's questions See All
Similar questions and discussions