The buzz around artificial intelligence (AI) and conversational agents like ChatGPT has reached fever pitch. Like many, we are trying to figure out how we feel about these rapid additions and announcements. Criticality can sometimes feel like a habitual posture. Whether this is an extension of an authentic inner life, or of the state of the world, remains an open question. While it's tempting to dismiss the current furor around advances in language models as mere hype, the capabilities being demonstrated are, well, awe-inspiring, and deserving of scrutiny. We remain at a point of ambivalence but find ourselves increasingly drawn to the potential of natural language tools to free us from the hegemony of institutional logics and expectations. Refusal or an inability to internalize institutional logics may no longer be a barrier to engaging with these systems in positive ways.
We often draw inspiration from Austrian philosopher Ivan Illich's critique of institutionalized Christianity in our work. One of his better-known arguments is that by turning the work of care over to institutions we sacrifice our own potential for grace. We justify not helping others by arguing that it is the job of the health system. We do not work out our differences through dialog, but by calling on law enforcement. Illich believed that by choosing to turn toward others, we realize the gifts of grace. By turning that ability over to institutions, we forfeit that gift. Although not a particularly religious person, I find this argument compelling and much of my work has been directed at modelling approaches that de-centre institutions in their implementation.
GPT-4 and ChatGPT reveal capabilities that are monumental in their implications. One notable innovation is that language models are increasingly able to mediate among various systems and agents to accomplish goals. In a recent OpenAI demo, a hand-drawn annotated image of an imagined website was photographed and shared with a ChatGPT bot built on GPT-4. The developer then requested that the site be produced with a step-by-step outline of what needed to be done and the code necessary to deliver on this. A working website emerged 30 seconds later.
While Apple drove the birth of a marketplace for applications, it seems as if large language models will birth an explosion of APIs that allow for a range of systems to communicate with each other and with human beings through something that approximates everyday speech.
An API, or Application Programming Interface, is a set of protocols, routines, and tools for building software applications. It specifies how software components should interact and allows different applications to communicate with each other. APIs can be used to retrieve data, submit data, or perform actions on a remote server, making it easier for developers to integrate various software components into their applications. In simple terms, an API acts as an intermediary among different software applications and systems, enabling them to work together seamlessly.
APIs are being built on large language models that allow users to engage with downstream systems using natural voice commands. Rather than working within the interface of a particular system, I can say what I want and the agent will communicate these desires in ways legible to those systems. The same will be true of human systems that rely on ossified routines and logics. We ask an AI agent the legal implications of some course of action and it queries mountains of legalese to brings us back an answer we can understand.
At first, I was concerned that this would limit access to those that speak a particular form of unaccented English, but I now believe that this assessment was short-sighted. While these systems may struggle with particular dialects and accents, they open up legibility to and within thousands of previously impenetrable systems. Much of our life is spent making ourselves understood to a range of institutions and technical systems – institutional, technical, academic, artistic, and otherwise. At tax season, I cannot submit my financial information in a format that works best for me.
Later in the same OpenAI demo linked above, 15,000 words of tax code were dumped into ChatGPT and questions were asked about basic deductibles for a particular period (2018). The answer was quick and accurate. Accuracy is not guaranteed, particularly in these early days, but the goal is for users to be liberated from the necessity of accommodating the epistemologies of institutionalized knowledge. While I still believe that unjustified authority needs to be dismantled, the ability to bypass its norms and expectations through intermediary agents feels like a step in the right direction.
We are about to see a Cambrian explosion of intermediaries. The assumption is that these agents will overwhelmingly act in service to goals of comfort and efficiency, wealth creation and control. This is likely true. However, they will also free us from the need to internalize the assumptions and patterns of institutions: academic, medical, legal, or otherwise. I also believe that these agents offer a new canvas for aesthetic encounters. What happens when agents are tasked to deliver on ideas of a beautiful life shared with them in natural speech?
Recent claims by OpenAI’s Sam Altman suggest that they intend to provide users with considerable control over the outputs of these models and over the values that inform their agency in the world. I am encouraged by this claim, as it suggests that other uses will become possible. Some of these uses will be small-minded, selfish, and cruel. Others will be expansive, convivial, and kind. This will be true as long as we have selfish people. And, thankfully, kind people.
The Promise of AI and Personal Relationships
There is, of course, another side to this. Just as Illich criticized the bureaucratization of Christianity for undermining authentic values like love, community, and humility, similar concerns have been raised about AI. As AI systems become more pervasive, there's a risk of prioritizing efficiency, productivity, and profit over human connections and empathy. Artists and others can interrupt these priorities and suggest other ways that agents can be deployed. By prioritizing the aesthetics of AI experiences, we can work to improve AI’s ability to enhance, rather than corrupt, human relationships and values. Recent steps suggest that the tools will allow and encourage this.
Conclusion
A key aspect of embracing AI agents is enhancing their ability to translate and intermediate, adapting to various linguistic styles, technical environments, and cultural contexts. Recent trends suggest that this will increasingly be the case and I am excited to develop applications to complicate and expand the experiences available to us. Adaptability could allow AI systems to create different models of the world, fostering an openness to exploring diverse perspectives and possibilities.
While there are legitimate concerns about centralization, capital, and the underlying cultural beliefs driving AI development, the potential for AI agents to act based on our intentions and draw from our experiences suggests the possibility for a more connected world.
The explicit reference to values by OpenAI and others in AI development raises the intriguing prospect of innumerable AI agents representing a wide range of value systems. By integrating diverse cosmologies and ways of knowing, we might create AI agents that engage in dialogues with one another and with us, negotiating how we live, occupy, and interact with the world around us.
Recent innovations in large language models have me optimistic about the potential of AI technologies to foster autonomy, conviviality, and novel aesthetic experiences. If we are able to address concerns about depersonalization and centralization, and embrace the potential for diverse AI agents, I can imagine a world where AI serves as a powerful tool to enhance our relationships, empower our creativity, and enrich our understanding of the world around us.
The current potential of AI lies in creating agents that understand and act on our intentions, based on the values we have explicated and then refined over time. By allowing AI agents to understand us through our natural, vernacular speech, we open up the possibility of a more beautiful world where AI systems act in support of ideas of relationality, conviviality, and connection.