Originally published in Kristan Horton’s Oracle, 2001.
Writing Down What Is Said
It could be said that we are witnesses to a technological revolution. To some this revolution is a liberating force, to others it is a threat to humanity, but to most of us it simply represents a series of bewildering changes to almost every aspect of our lives. The reaction from artists is equally varied. The artist Kristan Horton, for one, has thoroughly embraced the revolution, and has wasted no time in incorporating technological innovations into his art production. Nonetheless, Horton has not been intoxicated by it; which is to say, he has not succumbed to the allure of applying technology in art for its own sake. Instead, he has managed to remain aloof in this regard. As such he plays the traditional role of artist as outsider by exploiting the technology to help us see it anew. His Oracle project typiﬁes this approach to art. Below I shall discuss some of the issues that this project raises concerning the nature of the relationship between speech and writing.
A physical description of Oracle is straightforward: the Oracle machine comprises a row of six-tape players each plugged into a relay switch, which in turn is connected to a computer. Once the ﬁrst tape has ﬁnished the machine is programmed to cue the next tape-player to start while simultaneously switching to allow the computer to receive the audio signals from this next tape-player. This process continues until all six tapes have been played, at which point the machine loops back to the ﬁrst tape to start the process over again. Installed on the computer is dictation software, that is, a program designed to interpret the spoken word – converting the audio signals into text. This text is printed off, its pages to be eventually bound. In addition a monitor continually displays the interpreted text as the tape plays.
All six tape-players are used to play a nine-hour recording of a revised version of Homer’s Odyssey read by Alex Jennings1. The choice of this story is centrally relevant to the Oracle project. Homer lived in preliterate Greece, and so he was a storyteller in the oral tradition. Accordingly, his stories i.e., The Iliad and The Odyssey, were passed on to the later classical Greeks by word of mouth, only then to be preserved in written form. The advent of the alphabetic writing system in Greece in many ways was a revolution as profound in its consequences to everyday life as today’s technological revolution. And like today’s technology, writing was not favourably received by everybody. One of writing’s most renowned opponents, ironically, was the philosopher Socrates2. His argument against writing principally hinges on the claim that it lacks vitality. That is, he believed that writing petriﬁes the thoughts expressed in the speech it records. While writing can record or preserve the ﬂeeting ideas which we both utter and think, it lacks the immediacy of speech which is the quintessential medium of thought3. Of course this belief betrays the primacy of speech which Socrates presupposed, in that he saw writing merely as an instrument for recording speech, rather than a medium in which vital communication can itself occur. This belief in the primacy of speech surely derives from the fact that we master language through speech – writing is an addendum.
Writing is essentially mechanical, and it is this fact which explains how it has become such an integral part of our technologies, in particular in how we interact with computers. Speech, by comparison, is unwieldy. To an extent we have compensated for this lack of immediacy in our writing based technologies by exploiting the exponential speed of electronic media. And this compensatory processing speed has led us to try to give to computers the power of speech which they so conspicuously lack, or more precisely in this case the power to recognize speech. It is these attempts which are the background to the Oracle project. From a certain viewpoint one can read Horton, through Oracle, to be mocking such attempts: The text produced from the tapes is almost wholly unintelligible. For example, we get “implicit to dissuade unsealed slide to go down to laugh in disguise…” and so on. The program used by Oracle i.e., IBM’s ViaVoice, can be ‘tuned’ to produce a more accurate interpretation of the taped reading. But instead, by leaving the interpretation loose, Horton could be said to be highlighting the gap between the mediums of speech and writing. That is to say, he points to the essential difference between them which we attempt to overcome by sheer computer processing power. He remarks that such speech recognition programs represent an impulse to turn books-on-tape back into books which, he notes, is ridiculous from the technology industries’ perspective. Of course, dictation software in general, and ViaVoice in particular, is not intended for this particular purpose. But, that might be partly why he chooses to perform this task with it – to show the futility of this whole project. After all, it can be argued that the computer does not so much recognize speech, but rather converts it into a mechanical medium, namely writing which, as a machine, it can process. In other words, dictation software is really designed to mechanize speech.
As mentioned, the choice of Homer’s Odyssey is central to the Oracle project. For centuries, before it was eventually preserved in its written form, this story existed in virtue of being memorized by generations of storytellers i.e., bards. Moreover, the Odyssey is often referred to as an epic poem because it was recounted in rhyme in order to aid its tellers in their memorizing of it. Compared to ourselves, members of oral societies possess astonishing capacities to remember the spoken word, and rhyme is one of their primary aides-mémoire. Perhaps only with children’s rhymes recited in schoolyards does this oral tradition survive in our society. Notably, the word recognition program ViaVoice used by Oracle also exploits rhyme, or more precisely sound patterns in speech, to help it interpret the spoken words. Thus, its interpretation is never entirely arbitrary, that is to say, it does not simply interpret each word individually – sound patterns at least give the program a foothold in interpreting each word in some sort of context. So, for instance, the words “there’s a distant sound” might be interpreted by Oracle as “the airs a decent hound”. In this sense Oracle’s approach is eerily human, producing variations of the Odyssey in similar manner to how the ancient bards also produced variations of the story through the generations, echoing this oral tradition. Horton observes that this approach produces a reversal of the intended process of writing generally, which is to harden the ideas expressed in speech. What we in fact get is a loosening of what is spoken, namely the Odyssey according to its preserved written form as read by Alex Jennings. The machine becomes poet – an interpretative creator of texts.
Yet, crucially Oracle’s creativity lacks any semantic dimension. That is, while the computer converts human speech into writing, its dictation software is not at all concerned with the content of this writing. One might say that Oracle is blind to linguistic content in general. That is why its interpretation of the tape is often unintelligible – it interprets each spoken word without any regard for its meaning. For us writing down what is said involves being able both to mechanically write out the signs which represent the spoken words and to comprehend those words – Oracle can do the former, but not the latter. Given that poetry is essentially semantical in nature, that is, its vitality derives from its meaning, we see that Oracle’s output merely resembles poetry.
So far I have suggested that Oracle’s activities are a charade. That is, at first blush Oracle creatively interprets human speech, whereas in fact, its activities merely seem creative – underneath there is only an empty mechanical process. Yet, this reading of Oracle seems too harsh. Indeed, were one to watch Oracle’s monitor actively generating text one would likely think that there is genuine thought occurring – one would feels that something is going on inside that computer, however primitive and mechanical its processes might be in comparison to humans. Further, Oracle possesses an autonomy, albeit limited, which is surely the deﬁning feature of a minded being. This autonomy is evidenced by the fact that its textual output is never the same – each time it returns to the same passage it offers up a new interpretation.
The suggestion that machines can be minded, or even are so, is much debated. The English mathematician Alan Turing4 contended that if a machine can behave in a manner indistinguishable from intelligent behaviour, then in effect that machine has a mind in this regard. In reply to this contention the American philosopher John Searle5 presents an argument to show that such outward resemblance to minded behaviour is not evidence for genuine mindedness. Brieﬂy, Searle imagines a man locked in a room. In the room the man has a book which lists a set of rules indicating how to reply to any string of Chinese characters using Chinese characters written out on individual cards. The man, or operator, speaks no Chinese and so relies entirely on the rule-book to reply to Chinese sentences passed through an opening into the room. Thus, the man offers intelligible replies to such sentences using the cards, which he posts back through the opening a while later. One might conclude that to the outsider the operator in the room understands Chinese. This claim is parallel to Turing’s concerning intelligent behaviour. But Searle observes that this operator does not understand Chinese, and therefore the inference that he does from the evidence of his replies is invalid. Likewise, even if a machine like Oracle were not only able to accurately interpret speech, but could also manipulate words in a syntactically correct manner, call it super-Oracle, still this would not be evidence that it is in any sense conscious. Super-Oracle would merely seem minded.
However, Searle’s Chinese room is not comparable to the kind of machine with which we are concerned i.e., a super-Oracle. The programs of a super-Oracle would perform the same function as the Chinese room’s rule-book, but notably without an operator who can or cannot speak Chinese. Accordingly, for such a computer its apparent ability to understand Chinese cannot be refuted by pointing out that its operator cannot in fact understand a word of Chinese, since no operator is involved. Further, more generally it can be argued that the fact that Searle’s operator cannot speak Chinese is wholly irrelevant, it is the ‘room’ which ostensibly understands Chinese. The fact that the operator cannot understand Chinese does not refute the claim that the room can. Still, we can sympathize with Searle’s overall claim that understanding does not reduce to an ability to manipulate signs i.e., a mastery of syntax – even a super-Oracle cannot surely be said to understand the language it operates on.
There seems to be one crucial difference between Oracle and the Chinese room: there is no rigid connection between Oracle’s input and output – as noted, each time Oracle ‘hears’ the same passage it produces a new ‘interpretation’. In the case of the Chinese room, even though it is operated by an autonomous human, its output is determined by the rule-book, and accordingly it does not produce a new response each time to the same input. Perhaps, however, this rule-book could be written in such a way that the human operator does have a choice of words, and so the Chinese room’s output would vary for the same input. Nonetheless, still its operator has no relevant criterion by which to make a choice between words – he operates blindly. In Oracle’s case, its interpretation i.e., output, is governed by how well it has been trained – the more thoroughly trained Oracle is the higher the probability that it always produces the same interpretation each time a passage is repeated to it. No training is involved in the production of the rule-book for the Chinese room; its rules are given, so to speak. And in this respect Oracle is more human. Oracle’s training mimics how we learn to speak, and later write; namely by a constant process of trial and error. In this sense, Oracle is not purely a machine. It is important to understand what is meant by ‘machine’ in this respect. Here, by machine I mean any physical object which is designed to perform a speciﬁc task. In order to perform its task all of a machine’s operations are predetermined. In this respect the Chinese room is essentially a machine because all of its input to output relations are predetermined, even if its rule-book is written so that its operator has a choice of words – still this choice is strictly predetermined. Oracle, by contrast, is less of a machine because its input to output relations are not entirely predetermined.
That aspect of Oracle which is not machine-like i.e., mechanical, is instead organic. Thus, Oracle is in part an organism. Of course, ordinarily an organism is deﬁned as a living entity. However, as a physical object, I suggest that an organism differs from a machine in terms of how it is designed. As noted, a machine’s operations are predetermined; by contrast, an organism’s operations are determined in relation to its environment. Thus, for example, we say that a tree has an organic form, because its form has been determined by its environment – the prevalent direction of the wind, whether it stands on ﬂat of inclined land etc. Likewise, an airplane has a mechanical or inorganic form because it is designed with a speciﬁc function in mind. Accordingly, we see now why speech is essentially organic while writing is not. Speech arose from our interaction with our environment, while writing was designed by us to perform a particular function, namely to record what is said. The fact that speech is constantly evolving is evidence of its organic nature – its design alters with changes in the environment, namely those new aspects of the environment we ﬁnd need to communicate to others.
Thus, in order for a computer, such as the one used by Oracle, to master speech it must learn to recognize speech in an organic way, namely by allowing its operations to be partly determined by its interaction with the environment; which in this case is speciﬁcally the degree of correctness the computer achieves in its interpretation of the spoken word. By not training the computer to accurately or correctly interpret speech, Horton effectively emphasizes the organic nature of Oracle’s design in this regard. It is this organic design, I suggest, which encourages us to ascribe creativity to Oracle’s operations. In this sense, talk of whether machines such as Oracle are minded or not does not help us to understand the crucial distinction between machines and organisms. Moreover, not all organisms are minded, and indeed among the plethora of organisms presently existing, we surely ascribe to them varying degrees of mindedness. It is Oracle as an organism, rather than as a putative minded being, which is relevant in explaining its creativity. Can we therefore admit a degree of poetic creativity to Oracle? I stated that poetry per se is essentially semantical, and Oracle’s programming is blind to the meaning of the speech it interprets. Nevertheless, while we might not be able to describe Oracle’s output as poetry, inasmuch as rhyme is an aspect of poetry Oracle is indeed creative. And this creativity derives from Oracle’s organic nature.
© Hugh Alcock
- The Odyssey read by Alex Jennings (audio tapes), London: Penguin Books.
- Socrates’ views on writing are clearly laid out in Plato’s dialogue Phaedrus (see especially 274-279).
- Roughly, according to Socrates, speech, as the medium of thought, is the vehicle for the acquisition of knowledge i.e., the truth. Moreover, dialectical debate i.e., a reasoned inquiry through the discussion of specific concepts, can by itself provide knowledge. A famous example of such an inquiry features a discussion between Socrates and a boy slave of a young aristocrat named Meno. Without teaching the slave the concepts of geometry – which the slave has never learned – Socrates seems to succeed in demonstrating that still the slave can recognize for himself certain truths of geometry in virtue of his discussion with Socrates. (see Plato’s dialogue Meno 81b-8
- Alan Matheson Turing is most famous for his achievement of working out the Enigma code machine used extensively by the Germans to send secret messages during world war II. His views on consciousness are found in his article ‘Computing machinery and Intelligence’, Mind, 1950.
- See Searle’s article ‘Minds, Brains, and Programs’, The Nature of Mind, David Rosenthal ed., pp. 509-519, Oxford: Oxford Univ. Press, 1991. The Chinese room