I disagree. (I realize that you were primarily referring to the immediate interface, but consideration of the higher levels may well be important in the implementation of the lower levels besides the implications on the amount and type of processing power desired.) Having a dynamic display provides opportunities not available to pen-and-paper. E.g., a tablet could recognize geometric shapes as well as symbols and render them more neatly than they are drawn. A tablet can also provide zoom-in and zoom-out as well as a virtual sheet of effectively unlimited size. Erasure of large areas should also be easier on a tablet (one might want strike-through or crossing-out to implement erasing) as would be cut-and-paste and moving blocks of content (this would include the ability to write content, put it inside a rectangle, e.g., and have it automatically resized to fit). Hyperlinks to other documents and dynamic adjustment of detail (something like inlining footnotes could be useful–I find that when I write down thoughts I encounter side-thoughts which belong with the main thought group but are peripheral, explanatory, or otherwise not appropriate to a higher-level view [in writing this leads to parenthetical comments within parenthetical comments]). Layering could also be used–somewhat like transparencies–to conveniently express certain relationships of information. In addition, changing the color and size of the ‘pen’ should be much easier than for traditional pen-and-paper (as well as flood-fill). (Also providing lines or a grid could be useful, with the ability to hide such temporarily as well as to make objects snap-to-grid.)

One problem with having the computer make the writing more like typing in appearance is that delayed feedback could be annoying (in particular knowing whether a particular writing is legible to the computer is important [having a quick way to correct or even merely constrain the computer's choices could be good]) but early guesses by the computer could be even more annoying. (In some cases, off-line interpretation of writings could be useful as such could use a larger context, use much more computation, and provide feedback about uncertain aspects of a ‘translation’ [such feedback would be distracting during composition but could be preferred over making assumptions when the writer is in an editing/review mindset].)

(Human-computer interfaces could also be improved by exploiting user context [e.g., this user often draws flowcharts, that user often sketches biological structures] and document context, ideally with minimal additional human input and the ability to explicitly correct errors/change modes with minimal effort. Being able to provide vocal input could also be useful, but such is limited by the need to avoid the voicing being heard by others [whether for privacy or to avoid annoyance]. A virtual reality helmet [or even a mask over the mouth] would probably be too cumbersome and unfashionable for most users and sound cancellation would probably not be practical technologically in the near term. [A helmet would have the additional advantage that a set of relatively large antennae could be included. Combining the protective elements of a traditional helmet with the above I/O aspects could be useful--there may already be motorcycle helmets that would facilitate listening to a radio and talking on a cell phone while riding a motorcycle. {On a science fictional note, a constant covering of the face could remove some social cues, possibly reducing effective racism, and allow more conscious control of one's presentation of oneself--coloration and patterning could be dynamically changed to express mood and voice synthesis could hide or transform vocal cues, one might even have most communication be directed rather than broadcast as current vocal communication is.}]