What Happens When AI Simulates Its Own World
The Story Behind "Smallville"
Ladies and Gents,
Welcome back to the Random Thought Of The Week.
We’ve all heard of Chat GPT, the revolutionary artificial intelligence chatbot by OpenAI that has already changed our world (and will maybe take over our world, but I digress). But what happens when we take Chat GPT and create an open-world style simulation that is reminiscent of “The Sims”? Well, turns out, it’s shocking. So my friends, let’s book our ticket, pack our bags, and take a direct flight into the glorious simulated city known as “Smallville”.
Stanford and Google researchers published an experiment last month in which they developed an open-world style simulation that used Chat GPT to create 25 “agents”. These agents then were let loose in a sandbox (pictured above) and allowed to interact and communicate with each other. While the agents never actually moved, the sandbox was an imaginary visualization of the simulation and the interactions.
Agents were programmed by plugging in defining paragraphs such as these:
“John Lin is a pharmacy shopkeeper at the Willow Market and Pharmacy who loves to help people. He is always looking for ways to make the proces of getting medication easier for his customers; John Lin is living with his wife, Mei Lin, who is a college professor, and son, Eddy Lin, who is a student studying music theory; John Lin loves his family very much; John Lin has known the old couple next-door, Sam Moore and Jennifer Moore, for a few years; John Lin thinks Sam Moore is a kind and nice man; John Lin knows his neighbor, Yuriko Yamamoto, well; John Lin knows of his neighbors, Tamara Taylor and Carmen Ortiz, but has not met them before; John Lin and Tom Moreno are colleagues at The Willows Market and Pharmacy; John Lin and Tom Moreno are friends and like to discuss local politics together; John Lin knows the Moreno family somewhat well — the husband Tom Moreno and the wife Jane Moreno.”
TLDR - Each person in the simulation was programmed with a different attitude, personality, and life. Each agent acted accordingly to their description, making their actions unique and characteristic.
Agents “woke up” every morning and their AI interface made decisions about what to do at each step of their day. While they were influenced by their character description, that didn’t limit them from achieving simple actions like eating breakfast and brushing their teeth (for a full 2 minutes…no cavities in Smallville).
They also organized special events like parties. Agent Isabella Rodriguez threw a Valentine's Day party at the imaginary Hobb’s Cafe. 12 agents ended up hearing about the party, but only five of them actually showed up (Sorry Isabella, must’ve been a lame party). Agents talked about the party and spread the word similar to how actual humans would. For example, Agent Maria invited her secret crush Agent Klaus to the party.
Even though this party simulation may seem simple, it’s quite complex, so let's break it down. There was only one initial order asking Isabella to throw the party. The rest was up to Isabella to plan, decorate, and spread the word. After she spread the word, each AI had to remember that the event was happening. Not only that, but the AI agents also had to plan their day around the party. The agents then would decide to spread the word themselves (like Maria).
The AI agent Maria recognized that it was a Valentine's Day party, a possible romantic setting. She then put together that the romantic setting might be a good opportunity for her to spend time with Klaus (Well played Maria…well played).
I’m so invested in the thick romantic subplots of Smallville. I’m waiting for part two of this simulation to resolve the subplots.
How?
The simulation used a nuanced approach to mimic human emotions and behaviors, as well as a nuanced approach to interpretation and reaction to those emotions and behaviors. It does this by breaking down each situation into a series of simple decisions. Each decision the agent makes throughout its day is determined using AI. The agent’s memory, personality, and past events are all input, and then the AI simulates what might be the best course of action. For example, one agent was asked the question “If you had to get something [Ayesha Khan] likes for her birthday, what would you get her?”
The agent responded with “I want to get her something that demonstrates my appreciation for our relationship. Maybe I could get her something that reflects our shared interests, like a book.”
Guys, AI is better at giving gifts than I am. I don’t know if that speaks more to AI complexity or my incompetence. Maybe a little of both?
The Political Scene Heated Up
There was even a prominent political scene in Smallville. Look at this Smallville conversation discussing the political options for the local mayor:
Isabella: I’m still weighing my options, but I’ve been discussing the election with Sam Moore. What are your thoughts on him?
Tom: To be honest, I don’t like Sam Moore. I think he’s out of touch with the community and doesn’t have our best interests at heart.
This is fantastic. Despite the conversation being so formal, it’s still the most amazing thing I’ve read in…well, I don’t know how long. I’m still waiting for the Smallville political coup that overthrows the local government and tries to establish a de facto autocratic regime.
This conversation also has me thinking…maybe I’ll use AI next time I go vote to really spice things up. I wonder who it would vote for.
What This Means For Us…
Now, while we shouldn’t be donning our aluminum foil hats just yet, this study does pose ethical concerns. The way the AI agents learned and adapted to human emotions and behavior was, from a technological aspect, unprecedented. This complex behavior poses issues about relationships or friendships with AI. The number of middle-aged men out there that would talk to an AI agent more than their own wives would be astronomical.
While AI is far from sentient (many think that AI will never be sentient), this study shows that AI has the complex ability to make decisions about what might be best for a certain person at specific points in their day.
It begs the question: How long until AI can make decisions about what is best for itself? Asking for a friend.
Random Shower Thought Of The Week
Maybe plants are really farming us humans, giving us oxygen until we eventually expire and turn into mulch which they can then consume.
Morbid, I know.
My Friends,
I hope you enjoyed the Random Thought Of The Week. Thank you to my cous for keeping the spread of AI misinformation at bay. If you enjoyed, please share it with someone interested in AI. As always, I would like to leave you all with a quote from two great philosophers. One is named Bill and the other is named Ted. “Be excellent to each other, and party on dudes!”
Until next week,
H.R. Berry