Lovelace’s Notes on Babbage’s Analytical Engine not only included a fuller description of the way the Engine worked (compared to Menabrea’s French-language account). They also analysed the Engine’s potential and limitations, among which the passage quoted in (Turing, 1950):
The Analytical Engine has no pretensions whatever to originate anything. It can do whatever we know how to order it to perform… (note G). (Lovelace, 1843)
Turing’s translating Lovelace’s insight—that Engine’s (in)ability to originate something—in terms of an (in)ability to surprise him (or humans generally) is intriguing. Originality and surprise are distinct concepts which overlap only occasionally. I may deploy novel methods to generate outputs that challenge accepted norms but fail to surprise anyone (as is the case of many unrecognised artists), and conversely: my outputs may have been produced by following to the letter a set of instructions, and yet they may surprise many (the set of instructions may have been designed to maximise the chances of triggering ‘surprise’ in my audience).
Peculiar as it is, Turing’s ‘originality as capacity to surprise’ translation is significant: it reflects methodological assumptions that still shape much of today’s cognitive sciences, including debates on artificial agency. This section highlights the contrast between these assumptions (unpacked in §2.1.) and the types of questions that arise from first- and second-person accounts of originality (§2.2. and §2.3.).
Third-Person Accounts of Originality
The year immediately following the publication of his 1950 ‘Computing Machinery and Intelligence’, Turing gives a broadcast which focuses almost exclusively on Lovelace’s originality insight (Turing, 2004). This BBC broadcast helpfully spells out what I shall call the ‘downstream epistemic’ understanding of originality:
If we give the machine a programme which results in its doing something interesting which we had not anticipated, I should be inclined to say that the machine had originated something, rather than to claim that its behaviour was implicit in the programme, and therefore that the originality lies entirely with us. (Turing, 2004, p. 485)
Such an account is ‘downstream’ in that the originality assessment is wholly dependent upon the recipient’s characterisation. The criteria at play are also overwhelmingly epistemic: rather than dwell on when something might be deemed ‘interesting’, Turing focuses on the anticipation criterion, or what is sometimes called the ‘epistemic limitation’ in the contemporary literature: ‘Satisfaction of the epistemic–limitation condition by a machine amounts to performance by the machine in ways unforeseen by its creator (or someone with access to the resources its creator did)’ (Abramson, 2008, p. 160).Footnote 3 On this account, originality would be conditional upon the impossibility of anticipating the agent’s behaviour, given what we know about the machine’s (or human’s) ‘initial state, and input’. Since such a definition departs from our everyday understanding of originality, one may seek to query the reasons behind it (if an attentive mother finds it impossible to anticipate the everyday behaviour of her toddler, the latter is not necessarily deemed ‘original’).
The rationale behind this ‘epistemic’ delineation of originality may be reconstructed along the following lines: without presupposing what needs to be demonstrated—the possibility of machine agency—some sort of ‘thinking’ autonomy could only be inferred from epistemic limitation conditions. The problem with this line of reasoning is that it allows methodological constraints to overly narrow down the definition of what needs to be demonstrated: autonomous agency. The latter comes in different kinds and degrees. While a pupil who has mastered long divisions may be said to have ‘operational’ autonomy in that respect (just like a computer that can play chess), creative autonomy presupposes the ability to question and change given rules or conventions (Boden, 1998). The latter is also different from what may be referred to as ‘fantastic’ autonomy: it is much more difficult to create something new from within a background of accepted norms and conventions (i.e. creative autonomy) than to do so in the absence of any prior constraint or interpretive convention. Such ‘ex-nihilo’ creativity may be called ‘fantastic’ in that it is not the way we humans generate novelty. Yet, the third-person, epistemic perspectives on originality sometimes seem to presuppose this ‘radically unprecedented’ sort of autonomy.
These third-person perspectives are also unable to capture the diachronic dimension inherent in the term ‘originate’. While some inadvertent artistic output may be deemed original, for the artist to be deemed to have originated something, that output needs to have some future: unless it prompts the emergence of some intellectual, artistic or socio-cultural practice, that output’s originality only captures its fleeting novelty in relation to existing practices. Whether an output’s saliency perdures beyond that fleeting assessment—in which case the output’s creator has ‘originated’ something—depends on both the qualities of that output (some posit the latter as conditions of ‘objective’Footnote 4 or ‘inter-subjective’Footnote 5 creativity’Footnote 6) and its audience’s receptivity.
First-Person Accounts of Originality
Rather than focusing solely on what counts as ‘original’, a first-person perspective will seek to understand what drives originality as an effort. What needs or desires does a quest for originality seek to fulfil? What, if anything, might compromise a capacity to originate something? Might Lovelace’s ‘cannot originate anything’ verdict ever apply to human agents? These questions fit within a wider, long-standing philosophical endeavour to understand human agency.
Our interactions with the world around us are structured by beliefs, norms or desires that are shaped by our socio-historical circumstances. Today, a large proportion of philosophers will readily acknowledge this contingency,Footnote 7 even if they will differ in their articulation of its epistemic and normative implications. Putting aside nihilistic, ‘anything goes’ accounts (whether in epistemology or ethics), one step forward is to de-dramatise this acknowledged contingency:
Because we are not unencumbered intelligences selecting in principle among all possible outlooks, we can accept that this outlook is ours just because of the history that has made it ours; or, more precisely, has both made us, and made the outlook as something that is ours. (Williams, 2000, p. 491)
To be genuinely relaxed about the latter conclusion is not without its challenges: from an epistemological perspective, it requires abandoning any striving to know the world as it is in itself (independently of our representations of it). From an ethical perspective, one must be able to account for the possibility of change: that some contingent history has made us so does not mean we cannot set in motion a chain of events that will change its trajectory. It is less easy to show how we do that, if all we have to trigger the movement of critical scrutiny are our culturally conditioned habits of evaluation, rather than some Archimedean point safely removed from the contingent mess of human affairs.
Whether in the domain of aesthetics, epistemology or ethics, it is only once one takes on board the ‘encumbered’ nature of our intelligence that the full significance of originality as an imperative—rather than nicety—comes into light. If that is so, some may wonder why originality should be of any relevance as an indicator of artificial intelligence? If what drives originality as an effort (and perhaps a sign of intelligence) is the need to be able to challenge and enrich the web of socio-cultural expectations that shapes us from birth, then there must be room for sophisticated learning machines that have no such need for originality. Unlike humans, surely such machines could be in a position to control what they let themselves be ‘shaped’ by?
In its pragmatism, the above rejoinder however confuses needs and imperatives, and in doing so misses the heuristic value of Lovelace’s insight, which has only grown since Turing first sought to capture it. For better or worse, many efforts to build learning agents are premised upon the validity (and helpfulness) of a (young) human–machine analogy: ‘We believe that to truly build intelligent agents, they must do as children do’ (Kosoy et al., 2020, p. 4). So far this analogy has been interpreted narrowly. Efforts to understand children’s learning processes often proceed from the assumption that these learning processes are mostly aimed at building a model of the world that minimises prediction errors. A good fit for many problems involving perception and motor control, the latter goal certainly makes for smoother, more efficient interactions with one’s environment. Yet, it arguably ‘leaves out much that really matters for adaptive success’.Footnote 8
To understand the value of other goals like play—or originality—an evolutionary narrative (of the type associated with the prediction error minimisation frameworkFootnote 9) will only go so far if it is not associated with a study of socio-cultural imperatives. At an individual level, originality as a goal is likely entangled with identity formation (as a process that tends to take a more dramatic turn in teenage yearsFootnote 10). At a collective level, endeavours to originate something will be tied to the mechanisms that underlie socio-moral change. This collective aspect calls for a second-person account of originality.
Second-Person Accounts of Originality
Few twentieth century philosophers can claim to have been more preoccupied by the factors that may compromise our capacity for ‘new beginnings’Footnote 11 than Hannah Arendt. Animated by a concern to see us retain our capacity for political action, which ‘marks the start of something, begins something new’, Arendt feared the consequences of its atrophy under the weight of everyday ‘thoughtlessness’ (Arendt, 2007, p. 113).Footnote 12 At stake is not just the possibility of ethical or political change (‘viewed objectively and from outside, the odds in favour of tomorrow unfolding just like today are always overwhelming’) (Arendt, 2007, p. 112). The loss of our capacity for originality would also undermine what is distinctive, and perhaps dignified, about the way we are always in the process of constructing human selfhood. Never settled in advance, ‘we are not anything definite until we reveal our “who” [to others] in speech or deed’ (Hinchman & Hinchman, 1984, p. 202).
Because the beliefs and norms that preside over the receptivity of those ‘others’ are themselves shaped by contingent circumstances, each endeavour to originate something presupposes the judicious combination of both ‘reproductive’ and ‘productive’Footnote 13 imagination (Ricoeur, 1975):
Any transformative fiction – any utopia, any scientific model, any poem – must have elements of reproductive imagination, must draw from existing reality sufficiently so that its productive distance is not too great. (Taylor, 2006, p. 98)
It is this emphasis on originality’s dependence upon the interpretive articulation of what Turner calls the ‘unoriginal structures that inform originality’Footnote 14 that matters for our purposes. The difference between ‘mere’ output novelty and efforts to originate something cannot be grasped unless one acknowledges this second-person aspect of originality. As a hermeneutic effort, originality is not dependent upon success: what matters is that there be an endeavour to judiciously combine the productive and reproductive imagination(s) referred to earlier. Without this second-person dimension, originality could be solely a matter of productive imagination. While the latter could be argued to be within the reach of today’s machines,Footnote 15 the interpretive capabilities required for reproductive imagination are still peculiarly human.Footnote 16 Arendt is far from the only philosopher to emphasise the importance of these interpretive capabilities,Footnote 17 yet she articulates them in a way that brings home what is strikingly absent in both Turing’s and contemporary emphases on unforeseeability as a (downstream) criterion for originality.
In contrast to such unforeseeability criterion, Arendt’s account highlights precisely the opposite:
The very originality of the artist (or the very novelty of the actor) depends on his making himself understood by those who are not artist (or actors). (Arendt, 1989)
Arendt’s emphasis on the second-person perspective is crucial: to be capable of originating something requires interpretingFootnote 18 a shared web of socio-cultural expectations in such a way as to remain intelligibleFootnote 19 while at the same time challenging those expectations. This challenge may result in an experience of novelty and/or surprise. The next section articulates the relationship between originality endeavours and surprise, which only occasionally overlap (and which third-person perspectives unhelpfully blur).