Agent Sources

A new AI front alters my outlook on media's future

This post is about “agents,” a hot new area of AI innovation. It covers the potential behind this fast-developing R&D space, the means to shape a new media ecosystem, and how agents blend with traditional means of content creation (in my case, augmenting a book—coming soon!).

Six weeks ago, I joined demo night at Betaworks, a New York product studio that’s spawned category pioneers like GIPHY, Instapaper, and Hugging Face. This event focused on “agents,” an emerging sub-category of artificial intelligence. These software agents are developed with a specific task in mind, which they autonomously pursue on a user’s behalf.

Agents are expected to become powerful and pervasive. As they do, they may displace trusted sources of information we rely on today.

Potential use cases are wide-ranging. Imagine having an AI teammate as your personal meeting planner and summarizer. Or envision a development partner that actively scouts and integrates relevant code into your projects. Even more, think of having a constant AI companion sharing information about questions, venues, or people you encounter (see Humane’s AI pin as a new form factor).

Agents point to a near-future state where intelligence is more integrated into our lives. Their potential extends beyond productivity tools and companion apps.

How might agents alter fields delivering news, media commentary, or domain expertise? This question of trusted sources— and what they might be in the future — led me to the scenario below. The point worth stressing is the last take in the thread.

This scenario isn’t science fiction or a futurist projection. Engineers are developing components for a new information ecosystem oriented around trust as we speak.

Scenario posted on Perspective Agents X, September 20, 2023

GPTs for Everyone (and Everything)

One notable addition to this new scenario came a few weeks ago at OpenAI’s developer day. Beyond introducing LLM updates, Sam Altman turned heads by announcing a “GPT Builder.” Now, any of us can create our own custom ChatGPTs. These personal agents don’t require coding knowledge. They only need written instructions, making agent-building accessible to non-technical users.

The current list of custom agents on ChatGPT's front page spans visualization engines (DALL-E), a games coach (GameTime), a negotiation advisor (The Negotiator), a coloring book creator tool (Coloring Book Hero), and a laundry advisor (Laundry Buddy).

Home page of ChatGPT, post GPT Builder introduction.

OpenAI plans to coalesce indie agents into a new GPT Store, allowing people to monetize their expertise in a new way (allegedly coming soon).

GPTs as Trust Agents

Now consider the potential of “trust markets,” characterized by GPTs carrying specialized, expert knowledge. GroupLang is an emerging start-up established by experts in machine learning from MIT, Facebook, and Netflix. They recently introduced a groundbreaking idea of integrating computer agents into a trusted marketplace.

It goes like this. Imagine a virtual bazaar of GPTs that function as digital twins. It comprises sophisticated AI replicas embodying human expertise that, in aggregate, revolutionize how information is shared and consumed.

At the heart of this marketplace are emulations, virtual alter egos meticulously crafted to simulate the perspective and presence of trusted experts.

Consider someone like Paul Graham, the founder of Y Combinator. Apart from his status as an entrepreneurial force, he’s also renowned for his extensive collection of essays on life and start-up success.

Now imagine a Paul Graham digital twin, trained on a corpus of his work, research archives, and writings. His digital self — which he owns — competes for status and attention in a competitive, dynamic marketplace. His value fluctuates based on the quality and relevance of his insights. In Graham’s case, leaders from all walks of life, from aspiring entrepreneurs to seasoned executives, may flock to his GPT to seek guidance, answers, and inspiration. Graham’s GPT allows his ideas to scale.

Let’s take it a step further. The emulator marketplace resembles a stock or commodities exchange. Each GPT’s worth rises or falls based on the demand for their expertise. The most insightful become a highly sought-after asset. Avatar experts monetize their perspective through fees and ascending market value and reputation.

In this world, creating brand equity takes on a whole new meaning. The perspective given to users is valued through sustained engagement in high-quality information, decision-making aids, and/or on-demand advice.

Lastly, envision such markets as better alternatives to the incumbent media-and-expertise-industrial-complex. The supply and demand for insights aren’t based on performative acts, grifting, or polarizing rhetoric. In this market, clickbait and disinformation don’t pay.

Informational value is delivered by matching needed perspectives with quality expertise. High-value agents become new custodians of knowledge. Their insights shape the future in ways unimaginable in a world dominated by human biases, flawed media incentives, or technical limitations.

Becoming an Agent Builder

So, what does an entry point into this emerging ecosystem look like? In my case, the proliferation of computing agents dovetails with my book, Perspective Agents. It’s an opportune place to see how a GPT agent can augment content created for a book.

While developing the book, I sensed writing it was a race against a machine. I wondered if the appeal of a book would be less relevant as GPTs became mainstream. I had a gnawing sense writing it could be a fool’s errand. Turns out, these worries were misguided.

Here are a few reasons:

1. Depth of Intelligence: Building an agent requires expertise, research, and data. Consolidating vast amounts of research that inform a book — academic studies, interviews, and notes is a means to train an agent model.

2. Flexibility: In my book, I explore societal themes that are dynamic and ever-evolving. While chapters delve into timeless human elements such as meaning, work, and identity, machine intelligence influences how we seek meaning, earn our livelihood, and shape our identities. This is happening in real time. Fusing a printed book with a live agent allows thinking to stay current, even leading.

3. Personalization: The disorientation accompanying such fundamental change is deeply personal. Having a corpus of content that cuts across business, leadership, and well-being allows readers to choose their own path. Want a sense of how people dealt with technology-induced change in history? Care to investigate how polarization and conflict are shaped by the technologies we use? Want to envision how AI and virtual reality will reshape the world we will soon inhabit? Interacting with a bot can help users go as deep and wide as they’d like.

The point is that a living agent can help readers grasp what’s happening, tuned to their interests. It can be a shortcut for the time-strapped or a way to recall important take-aways. So, it makes sense my entry point is building an accompanying agent for the book.

Perspective Agents GPT as a prototype agent

Perspective Agents GPT can help users expand their personal perspective to better adapt to unfamiliar changes we face. Once fully trained, the agent will give users deep, nuanced insights into the causes of tech-driven change, drawing across various fields, including technology, history, philosophy, sociology, and literature.

This GPT is not limited to a single field or viewpoint. Instead, it synthesizes ideas from various domains, such as the sociological impact of technological advancements or the human implications of AI development.

I hope that, by accompanying stories in the book with a live agent replica, Perspective Agents can propel greater learning of the relationship between AI and human culture.

You can check out an alpha version of Perspective Agent GPT here. I’ll continually update the learning model and integrate feedback as a test project.

This exploration into AI agents, particularly the concept of digital twins and expertise markets, points to a transformative shift in how we access and interact with information.

Individual GPTs, tailored to specific domains, are not just a theoretical concept but a rapidly unfolding reality. If their promise is realized, personal GPTs will offer customized, insightful perspectives that starkly contrast traditional media and information sources' biased or limited viewpoints.

With imaginative leaps, these new tools present an alternative—and hopeful take— on the future of personalized and trusted media.

For more on OpenAI’s GPT Builder and market vision, I recommend you check out Sam Altman’s keynote below.