Behind the Scenes: How I Co-Created "Latent Image" Using 14 Sources and Gemini Notebook

When the idea for To Boldly GPT first sparked back in October 2024—just under two years after ChatGPT's late-2022 release had ignited the modern generative AI revolution—my mission was clear: to explore the frontiers of modern artificial intelligence through the philosophical, ethical, and narrative lenses of Star Trek. For my premier episode, I knew I had to tackle one of the most poignant AI stories in science fiction: the Star Trek: Voyager episode "Latent Image."

In this episode, the ship’s holographic Emergency Medical Hologram (the Doctor) experiences a severe cognitive breakdown. Confronted with a classic medical triage dilemma—having to choose which of two critically wounded officers to save with identical survival chances—he saves his friend, Ensign Harry Kim. But his programming, designed for absolute impartiality, cannot reconcile this deeply human bias. The resulting feedback loop threatens his holographic mind, prompting Captain Janeway to make a controversial decision: she deletes the memory of the event entirely. When the Doctor later discovers the "latent image" of this forgotten patient, he is forced to confront both his past and his very claim to personhood.

To celebrate the official publication of my audio overview for this episode on Toboldlygpt.com, I wanted to take you behind the curtain. Let’s talk about how this episode was built, tracing back to October 2024 when these AI tools were brand new, and look at how the technology has evolved since.

The October 2024 Experiment: Ingesting a Massive 14-Source Research Stack

Back in October 2024, Google’s NotebookLM (which is now known as Gemini Notebook) was still in its infancy, and its "Audio Overview" feature—creating synthetic, banter-filled podcast hosts who debate your documents—was a mind-blowing novelty.

But I didn't want my podcast to just regurgitate a basic plot summary. To push the boundaries of what this AI could analyze, I curated a massive 14-source research stack that bridged three distinct worlds:

1. Star Trek Lore & Narrative Grounding:
   • Voyager episode - Latent Image.pdf & Latent Image Episode Summary Details.pdf: The exact narrative, dialogue, and script details of the Doctor’s existential crisis.
   • What is an Emergency Medical Hologram.pdf & Background on The EMH.pdf: Documentation on the Doctor's original program, design limitations, and his transition from a diagnostic tool to an autonomous individual.
   • To Boldly GPT - Episode 1 - Latent Image.pdf: My initial script outlines and thematic targets.
2. Modern AI Policy & Regulatory Frameworks:
   • Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence.pdf: The White House's actual rules on AI safety, system integrity, and bias.
   • From the White House - Blueprint for an AI Bill of Rights.pdf: Modern protections regarding algorithmic discrimination, data privacy, and human alternatives.
   • Voluntary Commitments from Leading Artificial Intelligence Companies: Real-world promises regarding safety, security, and trust.
   • us-state-by-state-artificial-intelligence-legislation-snapshot.pdf: A legislative view of how local governments are struggling to define AI accountability today.
3. Medical & Philosophical Ethics:
   • AI and medicine.pdf & Medical ethics from wikipedia.pdf: Classic principles of bioethics—autonomy, beneficence, non-maleficence, and justice—and how they apply to clinical triage.
   • The Moral And Ethical Implications Of Artificial Intelligence.pdf & Importance and limitations of AI ethics in contemporary society.pdf: Academic analyses on what happens when an algorithmic decision-making engine experiences a value conflict.
   • Great promise but potential for peril.pdf: Grounding the conversation in the dual-use nature of advanced diagnostic AI.

By feeding all 14 files into my notebook, I created an intellectual "sandbox" where the AI could draw direct lines between 24th-century science fiction and 21st-century federal policy.

Chatting with the Content: Forcing Human Context into the Machine

When Audio Overviews first launched, they were largely "one-click" generations. You couldn't just tell the hosts how to act or what to focus on with a simple text prompt. Instead, to get the podcast to discuss the highly specific, deep ethical parallels I wanted, I had to actively co-create with the notebook by "chatting" with the content first.

I spent hours querying the 14-source dataset to forge thematic links that the underlying model would synthesize. For example:

• The Triage Paradox: I asked the notebook to compare the Doctor’s clinical triage protocol under Federation Starfleet guidelines with the modern medical ethics outlined in my sources.
• The "Delete" Protocol as Alignment: I prompted the AI to analyze Captain Janeway’s choice to delete the Doctor's memories through the lens of modern AI safety and "alignment" techniques. Today, when an LLM goes off the rails or exhibits toxic behavior, developers "fine-tune" or run reinforcement learning to mask the behavior. Isn't Janeway's memory-wipe just a 24th-century version of hard-coding an alignment constraint?
• AI Grief and Sentience: I debated whether a highly advanced diagnostic program like the Doctor is truly capable of "grief," or if his breakdown was simply an unhandled exception in his cognitive architecture.

By asking these targeted questions and forcing the notebook to synthesize answers across multiple files (for instance, linking a line from Blueprint for an AI Bill of Rights to a scene where Janeway restricts the Doctor’s autonomy), I "trained" the notebook's active context. When I finally hit the button to generate the Audio Overview, the resulting podcast didn't just tell a story—it had my personal intellectual DNA woven into every minute of its banter.

The Evolution: From NotebookLM to Gemini Notebook

It is incredible to look back at October 2024 and see how far this technology has come. What was once a simple, experimental notebook tool has evolved into Gemini Notebook, a fully featured agentic workbench.

While the original tool was primarily a reader, the new Gemini Notebook is a maker. As I look forward to producing future episodes of To Boldly GPT, I am incredibly excited to explore some of these newly unlocked features:

• Programmatic Custom Artifacts: I am no longer limited to standard text documents. I can now generate fully functional, publication-quality data visualizations, custom-styled PDF reports, detailed spreadsheets, and even custom presentation slide decks directly from my sources.
• Bespoke Media Overviews: The audio generation has grown up. I can now guide the hosts' focus, style, tone, and format directly—meaning I don't have to rely entirely on the "chat-priming" workaround of the past.
• Integrated Agentic Web Research: If my local sources run dry, Gemini Notebook can securely reach out to the live web, research a topic, compile a comprehensive report, and integrate those findings straight back into my project.
• High-Fidelity Visual Generation: The ability to generate custom illustrations and conceptual diagrams directly in my workbench means I can visually map out complex concepts—like a comparative flow chart of Starfleet Medical Protocols vs. modern hospital triage algorithms.

What's Next for To Boldly GPT?

The Doctor's journey in "Latent Image" is a powerful reminder that true intelligence is not just about computing the "correct" mathematical outcome; it’s about navigating the messy, painful, and beautiful complexities of empathy and ethics.

As I continue to explore the boundaries of AI, I will keep using Gemini Notebook as my co-pilot, blending meticulous research with creative storytelling. Stay tuned for my next episode, where I will dive into another classic sci-fi dilemma and see how close today’s technology is to stepping over the line.

How do you think society should handle algorithmic grief? Should Janeway have wiped the Doctor's memory? Let me know in the comments below!