Beyond Access and Impact: The Ultimate Benefit of SkyReading/Writing

Stevan Harnad

Human cognition is not an island unto itself. As a species, we are not Leibnizian Monads independently engaging in clear, Cartesian thinking. Our minds interact. That's surely why our species has language. And that interactivity probably constrains both what and how we think.

Although Wittgenstein's argument that there could be no "private language" -- because language is based on rule-following and rules are shared social conventions -- is probably overstated and refutable, for present purposes it is valid enough: Language is the main medium of interaction of our species and it is fundamentally interactive, dialogical. It did not evolve to leave us lost in subjective, solipsistic thought. In terms of the time we spend doing it, conversing probably exceeds all other forms of human interaction, including feeding, fighting, playing, mating, and the "grooming" that some have argued it has evolved to replace (Dunbar 1993

The origins of language have been the subject of much speculation (Harnad et al. 1976, but perhaps a few things can be said about it with some confidence: Language began hundreds of thousands of years ago, and whether it started as gesture and then moved to speech (Steklis and Harnad 1976, or went straight into speech, the kind of adaptation it was was undeniably an interactive one: Speech accordingly has a characteristic real-time dyadic tempo. There is certainly some variation in the rates at which people speak, the optimal speaking rates they understand, their attention spans, their memories, etc., but, to a close enough approximation, the timing parameters of a contemporary TV chat show are probably representative of our species since very near the advent of language.

There are consequences of this: Speak too fast or too slow, and I won't be able to understand you. A subtler consequence (having to do with the memory and attention-span factor) is that if you speak for too long, I'll have trouble understanding too, and I'll only be able to respond to what you said near the beginning of your speech, or near the end, or to some selected portions that caught my attention in the middle. Chances are that the adaptive value of language in the original environment in which it evolved derived from relatively rapid exchanges of relatively short strings of information, again more like a conversation or a chat show amongst a few interlocutors than a long lecture by one orator to a throng -- that came too, but it came later. Although the adaptive scenarios people have proposed are without exception mere speculations, they are all variants on the idea that the utility of language must have been connected with its use in hunting, tribal defense, tool-making, and/or training others (especially the young) in these or other essential hominid survival skills. With the exception of pedagogy (which was probably a later development), it's hard to imagine these uses of language as consisting of long monologues: Relatively short interactive comment and response were probably the order of the day, performed at about the speeds we perform them today. But even if primal conversations were one-sided rather than interactive, a rate-limiting factor was how fast we could speak and understand, and how big a chunk we could remember long enough for it to have any useful effect (Miller 1956 , Cowan 2001

So it is likely that because of real-time constraints on articulatory rate, the speed of thought co-evolved with the speed of speech; their rates converged on roughly the same order of magnitude (though one hopes we thought a bit faster than we spoke) and were in phase. And that's still the way things stand now, biologically speaking, for, after the advent of language, the rest of the developments in the linguistic arena were technological (and cultural) rather than biological -- feats of "cognitive engineering," if you like: We invented the new medium of writing, so words, and the thoughts they conveyed, could then be transmitted beyond the reach of any individual human's voice, ears or memory. (The oral tradition had done this in part, but imperfectly, and only through the mediation of a vocal internuncio.) Then we invented the medium of printing, so words and thoughts could be transmitted beyond the reach of any individual human's pen or paper (Harnad 1991

Writing and print were not only ingenious ways of preserving and distributing thoughts, but they also freed us from many of the immediate constraints of memory and attention span, because written words could be read and re-read, allowing messages to be longer and more complex than anything one could hope to convey orally. They could also be written and rewritten, allowing messages to be more careful and disciplined than anything the rapid pace of spontaneous conversation could ever generate. But this lapidary property of the written word was purchased at a price: The interactivity of speech was gone, or at least it was slowed down to a pace that hardly seemed worthy of the word "interactive" at all -- considering the speed, commensurate with speech, of which human thought had already proved itself capable in the oral era: So literality did, in a sense, make us more like monads conducting monologues. To be sure, we were writing letters to one another, and replying, sometimes on the same day, but it was rather like what had formerly been a jig, danced together, turned instead into a sarabande, danced in lugubrious alternation -- or, to pick a more cognitive example, a long-distance chess game in which the players make only one or two moves a day, and spend the rest of the time waiting to learn their opponent's response: There was something profoundly out-of-phase about it, or rather about the thoughts behind it, which in real-time dialogue would have interdigitated instead of proceeding in fits and starts.

The chess analogy is instructive, because, unlike a conversation, a chess game often involves long periods of motionless thought, and being rushed makes one play less well. Yet the game is interactive, and if it were to be played with limitless time between moves, it would no longer be the same game (and would perhaps no longer draw on the same cognitive capacities). Slow-motion tennis would be even more obviously a different game.

These analogies are imperfect, but the point I do want to make is that in written dialogue as well as in slow-motion chess, apart from the extra time one is happy to take in order to reflect more, there is a great deal of dead time too, in which one's thoughts are idling, waiting for the other (interlocutory) shoe to drop.

Nor is the limitless reflection time an unmixed blessing in itself: Necessity is the mother of invention. What would become of the spontaneous wit of a brilliant salon conversationalist if each item of repartee could be put on limitless hold for pondering before transmission? Would dancing ability (in an age when dancing was still interactive) or tennis prowess be the same if each move and each shot could be preceded by hours of deliberation?

Again, chess, being cognitive, is the most instructive case: In principle, given infinite time. every possible move could be tried in advance, and hence the optimal one could be picked. But trying every possibility is not usually the way cognition goes, and certainly not what we regard as "creative" cognition: The cognitive "moves" we regard as brilliant are not the ones that are a result of mechanically going through all the possibilities, but the ones that somehow find a pattern latent in all the dreary combinatorics, a pattern that swiftly and directly generates a solution to a problem that looked hard until the pattern was discovered ( ).

The human mind occasionally discovers such patterns. It's impossible to quantify this -- to say how often this happens, or how improbable and consequential the patterns really are. Sometimes we discover things through long, noninteractive reflection, to be sure. But, given the evolutionary history and temporal parameters of language and thought, it is probably safe to say that it is in real-time cognitive interactions between minds that the resourcefulness of human cognition is most firmly engaged.

This essay is by no means intended as a polemic for a return to an oral culture, however! The power gained from the discipline of slowing thought down to the pace of writing, and preserving it verbatim, answerable for its validity not merely to the persuasive force of one orator on one occasion, but the endless scrutiny of peers and posterity, was probably almost as revolutionary a technological and cultural advance for human thought as the advent of language itself had been. But let us not forget that in exchange for those virtues of the lapidary medium (writing) we sacrificed some of the virtues of the labile one (speaking), particularly the possibility of minds interacting at the speed of thought. Could one have the best of both worlds?

Not in the medium of speech, which vanishes as it is uttered. Recording it is no help, because what one really needs is playback and editing capacity, for both one's own utterances and one's interlocutor's. And by whatever cognitive engineering means one might secure this -- whether the playback/editing is in the phonological medium or the graphemic one -- the need to do two things rather than one (i.e., not just to listen and speak to one's interlocutor, but to monitor and modify the record of what has been said by both parties) rules out what might might have seemed to be the ideal solution, namely, real-time interaction by writing.

It's not just the slow speed of writing that is the problem; even if a speech-recognizer could generate error-free graphemes as fast as we could talk, and even if we could read these as fast as we can hear, this would still leave us back where we were with spontaneous conversation: We would merely have an instant transcript, but no more opportunity for reflection. It is in part for these reasons that -- except for quick, urgent messages -- most people find the real-time Unix "talk" facility so unsatisfying. It's not just the frustration of watching someone else's slow typing, and backspacing to correct typos, or the bottleneck of one's own typing, but that one feels that if this was just a chat, we could just as well have talked by phone, and if careful reading and serious reflection were called for, off-line email would have been better.

So is email just a somewhat accelerated form of ordinary mail (which, as I said, even in the past sometimes had same-day turnaround)? And are we irretrievably severed by the written medium from the interactivity of the real-time dialogue for which our minds are biologically adapted?

I think not. Although fast paperless mail was what email may originally have been intended for, it has turned out to have some unexpected consequences, opening up some revolutionary possibilities. First, let us not under-rate the speed factor. In principle, for a message of just about any length, it can reach my interlocutor the instant I complete it. Second, it can at the same instant be branched to multiple interlocutors (in principle, to everyone). These two factors, of speed and scale are without precedent, but they are still noninteractive ones, insofar as the speed of thought is concerned. Rhat is not all, however. Instantaneous and flexible text-capturing, quoting and commenting capabilities allow a form of highly focused and selective off-line interaction with the text that does engage the real-time speed of thought, and engages it interactively, yet in the lapidary medium, and with precisely the playback/editing facilities that were missing in real-time dialogue (verba volant, scripta manent).

Recall the memory and attention-span constraint on the length of a particular utterance in oral dialogue: Run on too long, and your interlocutor will lose continuity, forget, and misinterpret. One is tempted to say (to a long-winded interlocutor): Why don't you just write me a letter? Long-winded conversations, if they do not turn into one-sided monologues (which are a fortiori noninteractive), are more likely to be divergent duo-monologues -- each interlocutor in turn launching off from some point in the primacy/recency memory curve for the preceding peroration that they have just endured impatiently, waiting their turn -- rather than convergent dialogues, which require each interlocutor to be relatively brief and to the point (so it can be ensured that it is the same point they are both addressing).

Well, in the quote/comment capability that email has made possible, a long-winded passage can be given full attention (if it deserves it); it is preserved verbatim, free of memory constraints, to be re-read as often as one wishes; and, most important of all, it can be selectively edited down to the specific points one decides to address in replying, and the reply can then be focused on those passages, quoting them so as to provide the full requisite context. One's own reply, too, has the benefit of the playback and editing capability, and can be written and rewritten till one feels one has gotten it right. Moreover (and it is for this reason that I have dubbed this form of interaction "skywriting"; Harnad 1990, in engaging in this form of real-time cognitive interactivity with electronic text, one can also keep in mind that it is not only one's interlocutor who will see one's quotations and comments, but all the others to whom the message was branched: This is like the benefit of a trial by jury without the real-time pressure and stage fright of oral testimony; or like a public debate conducted in writing; or like a symposium and discussion likewise conducted in writing; but writing in a new key: at electronic speed and scale, and with the powerful playback/editing capability just described. Most important of all, a permanent, public record of the interaction is preserved in the form of a "Hypermail" Archive:

I realise that these resources are quite familiar to all of you, but sometimes a thing has to be "made strange," as Schopenhauer termed it, and looked at as if for the first time, if one is to see its true properties and potential, particularly if it is something relatively new that has become a familiar commonplace too quickly, as email has done. The text-capturing quote/comment tools and conventions that have rapidly developed in the past decade were not the work of cognitive engineers, experimenting with and optimising emerging interactive resources. They were simply co-invented out of expediency by emailers. I'm not sure where the "> " convention for setting off quoted passages started, but it quickly becomes unworkable with multiple levels of quoted text, even when supplemented by preceding the ">" with each interlocutors initials (see ). It was probably born on Usenet and carried over to Unix mailers, or vice versa. It is certainly not a successful piece of cognitive engineering, yet it is an absolutely remarkable capability that deserves to be closely analysed and developed, because it is the means by which the best of both worlds -- labile speech and lapidary writing -- can be realised. But before closing, I would like ... [to describe] two anecdotes from my own experience that I suspect will resonate with experiences many of you have had too.

The first anecdote concerns my own first exposure to "skywriting": In 1980, there had appeared in Behavioral and Brain Sciences (BBS), the journal I edit, an extremely controversial critique of Artificial Intelligence (AI) by the philosopher John Searle called the "Chinese Room Argument" ( Most people, including me, thought the Argument was wrong; after umpiring several years of critical commentary on it in BBS, I was writing my own critique of it in the mid '80s when it was drawn to my attention that on "" -- the Usenet chat group devoted to discussions of AI -- a discussion of Searle's Argument had been going on for several years.

I tuned in to see what had been said, and perhaps add my own critical voice to the throng, but quickly discovered two things: First, the critics were mostly not cognitive scientists but computer programmers and students. Second, all of their Counterarguments to Searle were wrong. Now I myself considered Searle's Argument to be wrong at that time, but before posting my own critique I wanted to dispel the clouds of invalid arguments that were in the air, so I took them on, one by one (though often they were just variants of the same wrong reasoning or assumptions or conclusions), and I consciously did so as if my contributions were formal commentaries in a learned journal (even if the postings I was commenting on had not been that scrupulous), except that I adopted the Usenet quote/comment convention.

The results were quite remarkable. The archive of my own discussion quickly reached booklength. I spent countless hours on the Net every day, taking on all comers, patiently replying to different variants of the same bad arguments a different way each time, so it might not bore but inform the silent majority that I assumed were following all this. (I still have no idea how many were "tuning in"; although Usenet's Arbitron statistics estimate the total readership of each group, one does not know how many of them -- and who -- are following a particular discussion "thread.") And though my behavior no doubt had obsessive-compulsive features, I don't regret the time I spent at it at all: Necessity is indeed the Mother of Invention, and in the course of that Skywriting Tournament, defending Searle against invalid criticism, I came up with some positive ideas of my own, including a hypothesis about what the real problem underlying Searle's critique of AI was (the "Symbol Grounding Problem" -- its name was born as the subject of one of the threads of the Searle discussion, and it has since become one of my more important papers; Harnad 1990 as well as a potential solution to the problem (Harnad et al. 1991

It is unlikely that I would have come up with those ideas otherwise. The essential features were the real-time interactivity, the quote/comment capability, and the long series of determined interlocutors. It has since occurred to me that the exercise might have been even more fruitful if my adversaries had not just been students and programmers, but the best thinkers in the field -- and that eventually impelled me to start a refereed electronic journal (Psycoloquy ) and to become a polemicist for the online self-archiving of the refereed research literature (, but that is another story. What's relevant here is that even with that less than optimal demography, the interaction proved to be such a powerful idea-generator for me. Nor was there any doubt in my mind that the quote/comment feature was at the hear of it. Which brings me to my second anecdote, mercifully shorter than the first:

It was around that time that I noticed that I -- a compulsive reprint/preprint collector since the 70's -- completely lost my taste for on-paper texts: If someone sent me a paper reprint, I would email them to ask if they didn't perchance have an on-line version. Why? Not because I find the current generation of VDU's any more appetising to look at than you do, but because of the quote/comment (Q/C) capability. I had become addicted to it as a way of interacting with text, irrespective of whether it came from a "live" posting or a "dead" text (even by a dead author!): Either way Q/C made it alive for me. The technique was the same. Read it on-screen, save a back-up full copy, then start selectively deleting the irrelevant or uncontroversial passages, leaving only the skeleton of what I wanted to address in my "reply". But who was this reply for? Well, in some case I could think of it as being for the author, but usually that was not enough for inspiration. So I set up some discussion groups involving multiple minds, all interested in the topic under discussion. (A population of Skyreaders is essential to the inspirational power of Skywriting.) And as with the symbol grounding discussion, often -- not always, but often enough -- the interaction would generate the germ of my next published article. And meanwhile Hypermail Archives ( could serve as the permanent SkyDocument of record (see also Hayes et al. Flach & Powers 1993 and as well as Okerson & O'Donnell 1995 ) to see how Skywriting can sometimes spawn terrestrial incarnations too.

[... text deleted]

In this essay I have suggested that the on-line text-capturing and quote/commenting and self-archiving capability that has emerged in the last two decades has created the possibility of combining the discipline and reflectiveness of writing with the speed and interactiveness of speech in a form of interactive cognition that is sui generis and without precedent in human discourse and inter-mentation. Whether with a live interlocutor or just an inert text, the interaction can now take place at the brain-friendly, on-line speed of thought, rather than at the lamentably slow, off-line turnaround time that paper communication had dictated. Hence, in the interests of developing a textual corpus on which to base this revolutionary new form of interaction, what better course of action can the scholarly/scientific community take than to make its all-important refereed research record freely accessible online to one and all, for open-ended cycles of skywriting and skyreading, forever?

Updated and revised excerpts from:

Harnad, S. (1995) Interactive Cognition: Exploring the Potential of Electronic Quote/Commenting. In: B. Gorayska & J.L. Mey (Eds.) Cognitive Technology: In Search of a Humane Interface. Elsevier. Pp. 397-414.