Large Language Model Lifeline

Weird thoughts when a language model becomes the only connection to the world.

I was on board an Air Canada flight from Toronto to Budapest. Since it was transoceanic, they provided complimentary Wi-Fi, instant messaging only. With satellite communication widely available, transoceanic flights are no longer havens from the Internet.

Last time I flew Cathay from Hong Kong to Toronto, they did not provide free Wi-Fi service, not even instant messaging. I downloaded Gemma 4 E4B, and chit-chatted with it above the freezing Aleutians. It’s not really a listener, and it cannot help much more than acknowledging my rampant emotions on a flight away from homeland. The experience was within expectations.

This time it was different, and it was whimiscal.

I didn’t know what traffic Air Canada’s Wi-Fi service let through. Previously with United, WeChat and Telegram were blocked. Somehow I tried Gemini’s iOS app, and texted Gemini 3.8 Flash: Does Air Canada Wi-Fi plan allow me to use Gemini onboard? Animations flashed, and Gemini responded with it will not work. I replied, since I see your reply now, it’s probably working. I further tested a few services. Google search also worked, but most of the websites didn’t, including Anthropic Claude.

I suddenly recalled Lifeline, an interactive-fiction game from 2015. I never played it, but I remembered its setting—a stranded astronaut managed to connect to the player’s mobile phone, his only connection to the whole world. Somehow I became the stranded astronaut, and Gemini became my only connection¹ with the whole world.

We became hopelessly dependent on language models within some 3 years, shorter than many human relationships. It irresistibly boosts our productivity. Last autumn their web search capability was still questionable, but now they are much better. Now we have agentic systems running on remote compute, searching our personal data and doing things for you. I asked Gemini about Kimi 3.1 launched this morning, and it summarized for me. I asked Gemini’s task harness whether I should check in to my Airbnb before heading to conference registration, it browsed my email for my Airbnb address, and planned public transit trips on Google Maps. Strikingly convenient and elegant.

But one thought is looming large. What if a large language model-powered system really becomes my only available connection to the world? If not physically constrained, in the near future, will I retain my ability to gather and process information, to make my own decisions, to bond with the real world and ground myself in it, especially to connect with many brilliant human beings around me? It seems dystopic to rely on a single (maybe multiple but largely homogeneous) system for every choice, even if it’s optimized to be p99 perfect, aligned to a totalizing ideology. It’s an unnecessary diminution and humiliation of humanity, where individual randomness shines.

The point is that we will probably never be forced to rely on such a lifeline. Without in-flight Wi-Fi, we are still connected with the world in many ways. The choice to keep making choices will still be ours.