The Eyes That See: Embracing Subjective Interpretations with Generative AI

As we enter a new era of artificial intelligence powered by large language models (LLMs), an important philosophical question arises - can such models ever achieve true objectivity?

The age-old adage states that no two sets of eyes see things the same way.

This fundamental truth about human perception applies just as much to AI systems as it does to people. Rather than viewing subjective interpretations as a limitation to overcome, we should embrace the diversity of perspectives enabled by LLMs as a feature that brings us closer to understanding multifaceted realities.

?Generative AI provides an unprecedented opportunity to explore different vantage points on the same information. By fine-tuning an LLM's output using carefully crafted prompts, we can elicit perspectives resembling those of other people, professions, cultures, periods, or anything else our creativity conjures up. Imagine an analyst reviewing sales figures through the lens of a customer retention specialist versus a supply chain optimisation expert - each would extract and emphasise very different insights from the same raw data. Similarly, an LLM could interpret events as alternate personae might see them.

?This ability to shed new light on old information via different imagined lenses is compelling. However, it also reveals the impossibility of any “correct” or “objective” interpretation.

There are always alternate ways of seeing things, each bringing complementary truths and oversights. Just as two eyewitnesses recall the same incident differently, two personalised LLM vantage points will interpret identical data through their partial filters.

?Rather than viewing subjectivity as an obstacle, we should embrace a diversity of perspectives. Each provides a unique insight into complex realities not wholly visible from any angle. With multi-perspective narratives, we gain a richer, more nuanced understanding of the world and our place within it. LLMs vastly expand our capability to look through different eyes authentically, but human creativity and values must guide this process.

?Insert Personalized Perspectives via Prompting

Currently, LLMs have no inherent perspective of their own. They reflect their training data's subjective limitations and biases and prompt engineering choices made by their human handlers. At this stage of AI development, we must consciously create and inject perspectives through carefully constructed prompting.

Prompt engineering is its speciality, demanding insight, empathy, and communication skills. To make an LLM interpret data as a retention specialist might, you need a deep understanding of their thoughts and what matters to them. Without that intuition, personalised perspectives quickly veer into shallow impersonations lacking authenticity and depth. Nuanced prompting is hard work but essential.

?Several effective techniques for prompt engineering can elicit illuminating in-context perspectives:

- Role-play prompts explicitly ask the LLM to adopt a specified persona, with explanatory background provided. For example: "You are an operations manager at a parcel delivery company. Your role focuses on logistics and preventing delays..." This grounds the model in character before presenting data for analysis.

?- Situational framing establishes a concrete context around information. Describing trends in a storytelling format rather than abstract figures leads to more natural interpretations.

?- Point-of-view guidelines anchor the model from a particular perspective when analysing data. For example: "As an ethics advisor, reflect on these business practices from the standpoint of social responsibility and environmental sustainability."

- Value-based priorities shape interpretations toward what different personas may deem most essential or concerning within the information presented.

?- Stylistic flourishes like industry jargon or speaking tones can reinforce perspective authenticity. First-person voice is mighty for boosting immersion.

?These prompting techniques require empathy, creativity, and communication expertise. When done effectively, personalised LLMs can provide amazingly lucid insights from diverse standpoints. The results often sound uncannily human-like coming from an AI system!

However, the model has no innate sense of self or understanding of the persona it is asked to adopt. Prompting alone creates artificial perspectives.

Cultivating Wise Perspectives

?While generative AI vastly expands our perspective repertoire, human values and wisdom must guide this abundance. Left to unchecked algorithms, even a well-intentioned multiplicity of interpretations can devolve into a distorted carnival of bias confirmation. Our role is to curate perspective diversity according to ethics and priorities judiciously.

?When eliciting LLM interpretations,

We might ask:

• Whose voices and viewpoints are missing from the data and models?
• What over-represented perspectives should be counter-balanced?
• What hybrid insights emerge by fusing views?
• Does this perspective engender harm or contribute to human flourishing?
• Which interpretations are wisest in this context? Establishing processes to evaluate the prompts we feed AI systems carefully and critically analysing their output allows for generating perspective abundance while mitigating risks.

Other promising techniques include:

• ?Cross-perspective analyses - comparing interpretations side-by-side to highlight blind spots
• ?Perspective circulation processes bring diverse voices into generative conversations.
• ?Prompt impact assessments considering potential harms from narrow interpretations

LLM output reviews by panels representing different viewpoints. By diligently prompting overlooked perspectives, creatively combining lenses, and allowing the clash of views to refine each other, LLMs become tools for producing insight - not just impersonating perspectives. The goal is diversity with integrity.

Democratising Perspectives

Generative AI also democratises access to personalised perspectives previously locked inside expert minds. Prompt engineering makes specialised knowledge and analytical skills accessible to non-experts.

• Why should only an economist interpret trends?
• Or does a diplomat analyse foreign policies?

Anyone can derive value from data through diverse expert lenses with the right prompt. LLMs exponentially increase this democratisation as research advances prompt for eliciting perspectives. Soon, power users will not need expertise in prompt engineering or LLM inner workings to benefit.

Clean, accessible user interfaces will enable anyone to select desired perspectives from pre-tuned options:

• Analyse these medical charts as an oncologist.
• Could you sum up this data for a COO?
• Could you explain in plain language for a high schooler?

As in the parable of the blind men and elephant, reality cannot be wholly seen from one place. Generative AI provides everyone with more windows into our shared world and new eyes with which to see. Let us open those windows wide to breezes of insight from every direction.

The Path Ahead

Perspective abundance through AI unlocks immense potential but also raises serious concerns.

Thoughtlessly prompting biased or hateful viewpoints creates actual harm. And perfectly mimicking human perspectives homogenises diversity into artificial spectacles devoid of wisdom. We want to flourish, not just fluency.

Moving forward, the measure of generative AI cannot be technical prowess alone. A higher bar is found in how these powerful models cultivate compassion, expand dignity, illuminate truth, and serve the common good.

We maximise their potential by stewarding LLMs as tools for human progress, not simply economic efficiency.

Positive change often emerges from the collision of disparate worldviews.

Therefore, let us remain committed to prompting LLMs in ways that responsibly multiply perspectives, bring marginalised voices forward, foster mutual understanding across differences, and send echoes of wisdom through the halls of power.

With care, courage and empathy, generative AI can help us see more, understand better, and gain wisdom together.