Why Hermeneutics Matters for the Future of AI

As we integrate chatbots into our lives, they will be used more frequently in the service of helping with existential concerns — those that relate to our psychological, emotional, and spiritual senses of meaningfulness. There remains a risk that we can overestimate the significance of the appropriateness of their responses, even when they seem to present relevant information. This is due to the way chatbot’s reflect the data they've been trained on, yet do not offer a genuine understanding of the world and its complexities as it is experienced by humans.

Dataset vs. the actual lifeworld we experience. Call this a problem of indexing, or cross-purposes:

When using a chatbot, we have a specific set of human concerns and purposes that matter to how we live in the world. Chatbots don’t have any concerns; they have only protocols that dictate how they ought to address human questions.

Of course, it can appear as if chatbots do share our concerns. This confusion, or perhaps “sleight” is a better term, is largely due to the way we assume their human-like use of language means they have something like human intelligence and sentience. As experts note, we tend to anthropomorphize chatbots.

Why does this matter?

We can be duped into thinking that chatbots are providing answers with a shared concern for our goals and aims. In reality, it is only a semblance. This has two implications.

1. We need to be careful how we integrate chatbot responses in our lives.
2. We need to be mindful that relying on chatbot responses may impede, if not arrest, our personal development since they only present a limited representation of our world.

Enter hermeneutics.

Hermeneutics is a branch of philosophy specifically devoted to understanding the way human interpretation is involved in how we live our lives and construct the world. It can help us get a much clearer picture as to why the problem of AI indexing can lead to a scenario where AI dominates, directs, and even co-opts the human world.

Let’s begin with what is most familiar to us — human purpose.

Following the Arrow of Meaning

“The arrow of meaning” denotes the uniquely human drive to seek direction, create, and apply meaning. It’s what powers our ability to interpret our experiences and direct our lives toward fulfilling our individual purposes.

Hermeneutics explores how we follow this arrow — that is, how we apply meaning in our lives and turn ideas into actions and creations. It calls this process “appropriation” and “application”.

When we encounter and engage with others, we often appropriate ideas from our shared experiences. Whether it’s experiencing a sunset or watching a film with another person, shared experiences tend to elicit conversation, which in turn gives rise to thought and self-reflection.

The application of new ideas to our lives, such as when we imitate the courage of others to overcome our own challenges, is a key part of personal growth.

Appropriation and application are integral to how we develop as humans, but they would not be possible without our human sense of purpose. Hermeneutics is rich with discussion of human purpose, and it’s something that Martin Heidegger (1889-1976) notes in terms of our lives being directed by a “for-the-sake-of-which”. Each human action may have a respective purpose — I walk to cross the street — but what we find upon further reflection is that discrete actions tend to only make sense within a larger story, or constellation of actions unified through narrative.

Think of sets of individual actions as “episodes” comprising a larger story. I cross the street to get to the gym. I go to the gym because I want to maintain a certain level of fitness. Being fit helps me enjoy hiking outdoors. Engaging with nature is essential to my happiness.

Perhaps also of interest:

• Meaningful Work: What AI Can’t Do through Words
• Finding Purpose in the Age of AI: How Vocation Can Guide Us

Our narratives form the bedrock of how we understand meaningfulness in our lives. Our stories bear out “the arrow of human meaning” that we follow, even in the most trivial of instances. When someone asks, “How did your day go?”, notice that responding to the question requires telling a story that may connect to smaller episodic purposes (e.g. “I’ve been so busy, I can’t get to the gym.”) or to more profound purposes (e.g. “Today was special! I had a great breakthrough and found a way to solve poverty!”).

Humans and AI: Worlds Apart

There are four important features of human understanding that bear out the problem of indexing. Think of these as pairings comprising tensions, as in human vs. AI:

• Multi-dimensional View vs. Linear and Parallel;
• Processing Application vs. Execution.

(1) Multi-dimensional View:

Human understanding is multi-dimensional in how it operates on various interconnected and even discontinuous levels. This involves the interpreter’s background, the context of the situation, the history of the issue, the potential outcomes of various interpretations, and the affectative (or emotional) dimensions of understanding. In addition, our understanding is also discontinuous in how it will pull from something arbitrary or unrelated to see or say something new.

Metaphor is a great example of this, where we put familiar senses of words together to predicate new meanings. Poetry epitomizes this – “She sang beyond the genius of the sea,” as Wallace Stevens writes. It turns out, AI is particularly bad at creating these kinds of metaphors and might come up with something trite like “Her song was a tapestry dancing on the sea.”

(2) Linear and Parallel Processing:

In contrast, much of computational logic is linear and sequential. AI’s ability to understand patterns in data requires processing items (or tokens) individually and updating its internal state as it progresses through the sequence. Of course, AI chatbots have evolved beyond linear processing. They now leverage parallel processing, evaluating the entire conversational context to produce more relevant and coherent responses.

Despite advances in parallel processing, AI still faces memory constraints that restrict its ability to grasp vast contextual information fully. This can lead to inconsistencies or omissions in responses, especially in conversations involving multiple topics, tacit or implicit knowledge, and dynamic topical turns.

To recall, human understanding operates on various interconnected and disconnected levels. It involves not just the text or data itself but the interpreter’s background, the context of the situation, the history of the issue, the potential outcomes of various interpretations, the affective dimensions of understanding, the ability to recognize disparate elements in order to form a whole, and the ability to know when to break rules in order to achieve a desired outcome. (I once asked a prototype chatbot I was training to break certain rules of grammar and syntax. It did it once in the conversation but then reverted back to proper discourse.)

(3) Application:

Human understanding is applied in two senses. As we saw earlier, it appropriates ideas and applies them.

Another aspect of this application process is the use of reason to assess a situation in order to determine what factors are relevant in applying a specific idea. This is often referred to as situational reasoning. Hermeneutics defines this as the use of “practical wisdom” since it involves the application of retrospective and prospective thinking. That is to say, it draws on past experiences and the estimation of future consequences to arrive at a course of action.

?

(4) Execution:

In contrast, AI focuses on efficiently completing tasks by recognizing patterns and extracting relevant features from its dataset. AI uses word embeddings and other linguistic features to identify statistical patterns in text. This enables them to generate relevant responses based on the co-occurrence of words and phrases in the data. As Ted Chiang observes,

[T]he more the [AI] program knows about supply and demand, the more words it can discard when compressing the pages about economics, and so forth.

Large language models identify statistical regularities in text. Any analysis of the text of the Web will reveal that phrases like “supply is low” often appear in close proximity to phrases like “prices rise.” A chatbot that incorporates this correlation might, when asked a question about the effect of supply shortages, respond with an answer about prices increasing.

Human understanding, on the other hand, draws on both information (its dataset) and those factors constituting its multi-dimensional sphere of relations and meanings. It does this not merely to execute a task, but as we saw above, to execute a task in order to follow the arrow of meaning of its story. In other words, human understanding seeks application in addition to execution. It seeks to create or affirm meaning in the world.

To use the example of story-telling, AI can tell a good story at times, but this story has no bearing or significance unless it is heard by a human audience who can then apply it appropriately in their understanding. So perhaps we can ask, “If an AI chatbot tells a story, and there is no human there to hear it, does it matter?”

According to hermeneutics, it does not.

Conclusion: Be Present and Not Represented

AI's output may mimic human language, but hermeneutics reveals why we must resist mistaking its responses as sharing our human concerns to illuminate how our lives can be meaningful. AI’s concern is “indexed” to its dataset and training; human concern is “indexed” to how the world is.

Uncritically appropriating AI responses impoverishes our capacity for self-determination and critical self-reflection. This corrosive incapacitation gives a new face to the idea of living in a technologically enframed matrix.

Perhaps also interest:

• Is Technological Incapacitation Making You Less Human?

The world is not simulated because it is entirely replicated by AI; rather, it is simulated because our depth of understanding is limited to what AI presents through its representational lens.

About the Author

Todd Mei (PhD) is former Associate Professor of Philosophy specializing in hermeneutics, the philosophy of work and economics, and ethics. He is currently a researcher and consultant in meaningful work and is founder of Philosophy2u. With over 20 years of experience in teaching, researching, and publishing, Todd enjoys bringing insight, innovation, and worklife revolution to organizations, businesses, and individuals.

#hermeneutics #responsibleAI #ethicalAI #FutureofWork #AIChallenges #TechEthics #AIethics #AIintheWorkplace #CareerGrowth #MeaningfulWork #AIandHumanity #FutureofMeaning #Criticalthinking