Tranter, Kieran --- "And Then 'friends'" [2021] LawTechHum 14; (2021) 3(2) Law, Technology and Humans 1

---

Introduction

And then Friends

Kieran Tranter

Queensland University of Technology, Australia

This brief editorial focuses on the contribution in this volume titled ‘Machines Will Never Replace Humans!’ compiled by GPT-3. The brief text is provocative. It is provocative in demonstrating the potential efficiencies and complexities of machine-produced natural language text for ‘writing’ professions like law and the academy. It is further provocative as it reflects back the image and representation of the human within the digital. There is a denotive suggestion that humans are valuable and significant as lawyers because they possess intuition. There is a further suggestion that humans, or more precisely the imprint of humans in the digital, are televisual consumers of dated sitcoms, revealing the disconnect between existent digital archives and the totality of humanity.

Text by Robot

This issue of Law, Technology and Humans contains a commentary produced by the artificial intelligence (AI) system GPT-3. A project undertaken by Benjamin Alarie and Arthur Cockfield, it is, as far as we can identify, the first law review contribution drafted by an AI.

GPT-3 is developed by OpenAI and is the most familiar of the current generation of natural language processing AIs.^[1] The technicalities of GPT-3 are summarised by Alarie and Cockfield in their exegesis. The significance of GPT-3 is that it produces high-quality natural language text in response to a text seed. For the contribution in this issue, the seed text related to why AI will never be able to replace human lawyers. This means that the first AI law review contribution presents the argument that humans have ineffable qualities—empathy, institution—that machines will never possess.

Considering the text published and that, significantly, unlike some of GPT-3’s published outputs, Alarie, Cockfield and our copyediting team made no changes to the primary text, there are two significant notes I would like to reflect on.

Hybridity of Authorship

The first note concerns the hybridity of authorship. Read the contribution carefully; it is actually rubbish, resembling a marginal freshman’s essay in a foundation to law or issues in law course. However, as a rough first draft that an enterprising student or law professor could then work with to develop a more polished and coherent text, it does raise concerns. As Richard Susskind has documented for 30 years, computer-aided and, more recently, AI-informed predictive text tools have assisted students, lawyers and professors to proof and check their writing.^[2] GPT-3’s capacity to generate, from a seed prima facie, coherent, comprehensible text suggests a more significant automation of writing. Although the computer-aided tools that checked the spelling, tense and grammatical construction provided a level of automation of copyediting, GPT-3’s text generation capacity suggests the future of writing could be a more hybrid experience of managing and supervising text construction. The romantic ideal of the author alone drafting prose, already an unstable myth when considering the intertextual and intratextual traces and influences that animate the ‘author’, is further deconstructed. This clearly has implications for text-generating professions, such as the academy and lawyering, where knowing and identifying the author is fundamental for determining authority and accountability.^[3]

Humans in the Archive

However, it is the second note, what GPT-3 says about us humans, that I wish to dwell upon. GPT-3 is an entity birthed in the digital archive. It has been trained on public ‘knowledge’ repositories, Wikipedia being the most obvious. What it outputs is the product of machine-learnt processing of the records, posts and comments that humans have deposited in the digital. What it found, when asked to build text in response to the seed that machines could not replace human lawyers, was a 20-year-old sitcom about 20–30-somethings in New York. It found Friends.

There is a considerable literature on the cultural representations and meanings of Friends. It is a show that seemingly has had a renaissance in the streaming age, finding a new, younger audience rocking the ’90s and early ’00s fashion. That GPT-3’s text went rather quickly from human lawyers to Friends seemed an unfathomable leap by the black box. The show does not portray lawyers or, indeed, project many of the obvious signifiers of law, lawyers or legality. It is a show focused on the mundane domesticity of shared apartment living, economic precariousness and relationship churning during the ‘long’ adolescence in the Global North. The witty, slightly cynical ensemble banter does anticipate digitally enabled group chats, possibly explaining some of its attraction to contemporary audiences.^[4] Or maybe it is just all about Jennifer Aniston’s hair.^[5] The gap between the serious issue, especially for law professors involved in an industry funded through the production of the next generation of human lawyers, of whether the digital could render human lawyers obsolete and the superficialities of Friends seems bizarre—a profound error proving that GPT-3 is not the natural language processor breakthrough that it is claimed to be.

However, that Friends, specifically the televisual consumption of Friends, is the metaphor for what is ineffably human according to GPT-3, could give us humans a moment for pause. GPT-3 is a complex bit of code that crunches a reasonable portion of the accumulated digital archive of humanity. It extracts patterns and relationships, eddies and deep currents within the ocean of big data that it consumes. The trick in the seed text that we provided GPT-3 was not to ask it about the future of lawyering and law work. That was puffery. What we really asked it was, ‘What is there about humans that make them unique, so unique that the role of lawyer, a role infused with fundamental power over life and time,^[6] must remain with them?’ We asked GPT-3 to tell us what wisdom resides in the digital archive about human beings.

GPT-3’s output would disappoint romantics who hold onto a nobler vision of humanity. Within its sort of well-constructed sentences and slightly disjointed paragraphs, two traits of human being are presented.

First is a superficial affirmation of ‘intuition’. This is clearly a word that is on high rotation within the digital archive, in combination with the words ‘robot’ and ‘human’. There is a fabulous asymmetry in the denoting that humans are ‘intuitive’. Intuition is a catch-all for the emotive spontaneity of decision and action: ‘I intuitively acted.’ ‘I intuitively knew that it was the right choice.’ ‘We all intuitively knew that Friends would end with Rachel and Ross getting back together.’ There is a touch of the mystical with intuition—something guided by muses, angels or fairy godmothers. Contemporary neuroscience has gone a significant way to secularise intuition as an explainable electro-chemical process.^[7] Parallel social constructivist accounts locate intuition, along with ‘common sense’ into cultural milieus and interpretative communities.^[8] However, GPT-3 did not embark on a deep dive into the complexities and contestations of human intuition. Rather, institution was presented as a posited fact. Humans are intuitive. The asymmetry arises because in affirming intuition as a human virtue, what is being presented is that humans are a black box. They make decisions and take actions that are not transparent, reasoned or accountable. There seems to be a strange double standard. In humans, ‘black boxing’ is desirable; yet in AIs, it is not. GPT-3’s text highlights how the ‘intuition’ trump is used to suggest that humans need to supervise AI or be in the loop in decision-making processes. In doing show it reveals how weak foundations to this claim are. It is an obstacle to considering, in a transparent, reasoned and accountable manner, the actual strengths and weaknesses of decision-making and doing-in-the-world entities, whether they be humans or machines.

Second, humans consume media. Twice in the text does GPT-3 circle around to humans watching screens. The first instance leads to the paragraph about Friends and pontification about the actor who played Chandler. The second is GPT-3 writing that it was sitting down with a friend to watch a movie. For a text about law and lawyering, there is a lot of streaming by humans going on. The cynic would not be surprised. Interrogate the digital archive for human doing and what gets projected back is the consumption of media. The vaulted empathy and intuition praised in GPT-3’s text are not directed towards making the world better or about achieving justice. They are directed towards media consumption, satisfying random curiosities and the neurotypical reading and responding to interpersonal emotions, all fairly lowball doing in the world. The impression is humans as an indulgent pleasure caste consuming and following their whimsical flights of fancy, while others do in the world. This is possibly the actual reality of the digital archive that GPT-3 engages with. It is the record of the chattering, middle classes of the Global North—the white privilege that critics can easily identify in Friends.^[9] It is a comfortable world of televisual consumption and care for a friend’s relationship status. It is not the frantic gig-hopping by the precariat, the long hours of smartphone assembling in a Party-approved factory, the experiences of a woman in Afghanistan or a young First Nations man being approached by police officers in the Northern Territory. It is a world removed from substantial global realities experienced by humans in their located diversity.

It affirms that the digital is not yet the twin or data double of humanity. It is not the total archive of the human. This is, writ large, the bias in machine learning training data identified by critical algorithm studies. Claims to what is human emanating from this dataset will be partial and coloured, and any profound insight is more likely to be in what is not represented and what does not come ‘ready to hand’ to AI interrogators like GPT-3. It also shows that the fundamental programming principle of GIGO (garbage in, garbage out) still reigns supreme. GPT-3 confidently tells us that the actor who plays Chandler is ‘John Ross Bowie’. Of course, it wasn’t. It was Matthew Perry, and hasn’t he aged, as witnessed in the recent HPO Max Friends reunion special?^[10]

GPT-3 is not the Model T Ford for the automation of law or the academy. It is not the breakthrough product that delivers on the cultural imagery’s expectation of a much-considered emergent technology. The Model T delivered affordable, reliable mechanised transport, something promised and anticipated by earlier motor vehicles but that had yet to be fulfilled. GPT-3’s text is credible but only superficially so; the paragraphing is rudimentary and the fact-checking absent. The potential is evident, and it is possible that GPT-4 or GPT-5 might resemble the cultural imagery of a robot author, poet or scriptwriter, in that there might also be the possibility for the robot lawyer or the automated law research professor. However, such speculation is trivial. Ultimately, what the text by GPT-3 reveals is a version of the digital divide, showing what forms of being human are becoming imprinted into the digital archive and what forms, like the walk-on characters of colour in Friends, remain marginal and supporting. If the future involves the digital increasingly delineating human wellbeing and potentiality, then who or what is archived as the human becomes one of the most significant political projects in the present. This should be the takeaway from GPT-3’s data dump on Friends, intuition and law, that who will be human in the digital future is being constructed in the present.

Bibliography

Cobb, Shelley. “ ‘I’d Like Y’all to Get a Black Friend’: The Politics of Race in Friends.” Television & New Media 19, no 8 (2018): 708–723.

Dale, Robert. “GPT-3: What’s It Good For?” Natural Language Engineering 27, no 1 (2021): 113–118.

Dunn, Jennifer C. Friends: A Cultural History. London: Rowman & Littlefield, 2019.

Floridi, Luciano and Massimo Chiriatti. “GPT-3: Its Nature, Scope, Limits, and Consequences.” Minds and Machines 30, no 4 (2020): 681–694.

Isenman, Lois. Understanding Intuition: A Journey in and out of Science. Cambridge MA: Academic Press, 2018.

Lieberman, Matthew D. “Intuition: A Social Cognitive Neuroscience Approach.” Psychological Bulletin 126, no 1 (2000): 109–137.

Quaglio, Paulo. Television Dialogue: The Sitcom Friends vs. Natural Conversation. Philadelphia: John Benjamins Publishing, 2009.

Sherrow, Victoria. Encyclopedia of Hair: A Cultural History. Westport: Greenwood Publishing Group, 2006.

Susskind, Richard. The End of Lawyers? Rethinking the Nature of Legal Services. Oxford: Oxford University Press, 2008.

———. The Future of Law: Facing the Challenges of Information Technology. Oxford: Clarendon Press, 1996.

———. Tomorrow’s Lawyers: An Introduction to Your Future. Oxford: Oxford University Press, 2013.

Tranter, Kieran. Living in Technical Legality: Science Fiction and Law as Technology. Edinburgh: University of Edinburgh Press, 2018.

Winston, Ben, dir. Friends: The Reunion, HBO Max Aired 27 May, 2021.

---

^[1] Floridi, “GPT-3.”

^[2] Susskind, The End of Lawyers?; Susskind, Tomorrow’s Lawyers; Susskind, The Future of Law.

^[3] Dale, “GPT-3: What’s It Good For?”

^[4] Quaglio, Television Dialogue.

^[5] Sherrow, Encyclopedia of Hair, 38–39.

^[6] Tranter, Living in Technical Legality, 152–163.

^[7] Lieberman, “Intuition.”

^[8] Isenman, Understanding Intuition.

^[9] Cobb, “ ‘I’d Like Y’all to Get’.”

^[10] Winston, Friends: The Reunion.