I missed out the other two white boy hype rants in Krypto and Ntfs, so I’ll give it my best shot with this one. AI will change everything. It will accelerate so many fields from which we might benefit immensely. I have no expertise in any of those fields. What I can say a thing or two about is human behavior, and how economists explain it with costs and benefits. Conversation is a human behavior I have written about in the past and I think AI will play a pivotal role. I will argue that all non-in-person human interaction will be outsourced to our individualized AIs (Agents). And that's a lot of conversation: I guess my conversation interaction time is split 80% non-in-person and 20% in-person. I text and email all day long and I meet people 4 times a week for about 4-7 hours in the evening or on office days. I will argue that these 80% will vanish forever. And it should vanish, because it is inefficient. First, I will argue that everyone will have an agent. Second, I will argue that this agent will take over all of our non-in-person interactions. Third I will argue that it should, and I will present some counter arguments.
Everyone will have an agent. Let's call this individualized agent a person’s name with an X at the end. So how does KruxiX jump into being? We can ask ChatGPT to answer a letter sounding like William Shakespear, and it will do so effortlessly. If we feed a Large Language Model (LLM) with all of our whatsapps messages, voice notes, and yes, also cringe 2009 FB wall posts, it will be able to sound like we do. By delivering all my written and spoken conversion, an LLM will be able to create KruxiX that writes and sounds just like Kruxi. This isn't really a prediction because we can do it with famous people who have a lot of writing and speech out there, today. Just the amount of content you have available to the current is the barrier to write like you. With years of sexting and trash talk convo in everyone's whatsapp the LLM will have no problem sounding just like you. But this uncontroversial prediction has profound implications.
KruxiX will take over all non-in-person convos for Kruxi. To better understand my prediction lets get at this in two steps: 1. Let's imagine only Kruxi has KruxiX and everyone else is still human (x-less). 2. Let's say everyone gets an agent (x-full).
1. OpenAI, due to this amazing blog prediction, chooses Kruxi and only Kruxi to get the first KruxiX, my very special agent (juhuu). Remember, it sounds like me, it texts what I would text with 99% accuracy. Why would I text anyone ever again? It can answer work-related texts, console friends with their break-up, book dates, chat shit with tha bois and continue my conversations seamlessly, as well as start new conversations as I also did from time to time. Now that might sound weird at first. But think about it. Imagine a friend needs advice on where to go for a date with an acquaintance of yours. You now have to take 10 mins of your time to think about your acquaintance’s preferences, think of matching bars you know, and then write it all up in a lengthy message explaining each place and its rationale behind it. KruxiX has written the exact same message in 0.1 seconds. My previous Whatsapp messages reveal all the relevant information to KruxiX. It knows all the bars which I went to the last 13 years. It knows what bar I went to with which people. It knows the acquaintance through my texts with her. Why on earth would I spend 10 mins thinking and writing something up, which yields the exact same result as KruxiX does in no time.
A counter might be that answering googleable questions with an Agent is easy… but what about real tough ones. The parents of your friend send a text saying the friend passed away. As tragic as this is, KruxiX still replies the same message as I would after grief, speechlessness, anger… The outcome is the exact same. KruxiX knows me in these moments… they unfortunately did happen in the past and will in the furture. There are many caveats in this scenario: I would still want an update at the end of the day on what the most important chats were, and if I feel like reading through the trash chat with tha bois I can do so as well. Also If I need to jump into action immediately I will get notified on my phone. On top of that I would ask KruxiX to ramp up outreach to old friends. I wish I had the time to think through banter with my pal Burgs from uni, or continue fascinating conversations about Lotr and free will with my physics friend Hanna. I dont have the time. KruxiX has all the time. KruxiX texts old friends and I get to enjoy these conversations in my free time.
Still not convinced? Imagine writing a text, and as you are about to send it off, there will be a popup saying: “KruxiX would have written exactly that”. How annoyed would I be to know I just lost 2 mins of my life writing a text, and this every 5 mins. I am absolutely convinced that I would use KruxiX for ALL my conversations.
2. The KruxiX test phase is over and everyone gets an agent (an X). Everyone will use their Xs. This means no one will send a text to each other anymore. Xs will have conversations with each other. They will book dates, update on current events, chat shit about politics, fight about who will be the next chess champion… It is Agents talking to Agents, sounding like person A talking to person B. But person A didn't write it and might never even read it, same with person B. The cost is close to zero so it doesn't matter if no one reads it. This leads me to a couple of predictions:
Agents will talk to friends’ agents constantly. Humans can read up on a short brief, getting the spicy stuff bundled in the evening. This is great cause I need to text about things that I don't really care about before I get to the part that I care about. And KruxiX knows what I care about! So, KruxiX answers all the courtesy questions other people might find interesting, but also asks questions and gets stories fed that I think are interesting, and at the end of the day (or whenever) I get shown only the stuff I care about.
You can have ongoing conversations with old friends that you can revisit once a year. Since cost is zero you can let KruxiX talk to your uni friends, your lovely godmother, and your hilarious driving instructor, constantly. You might not want to see it now or ever. But if you feel like it, you read the most interesting, heartwarming, and/or funniest moments of the last 20 years of KruxiX and godmotherX. And remember, if both of you had the time, this would be the conversation you would have!
You can start conversations with famous people. I constantly go through life thinking: what would Steve Levitt (economist crush) say? I once received an answer from Anja Shortland on my blog about her book on Lost Art. It meant the world to me (it still does). Imagine having ongoing conversations with your idols, people who shaped your thinking. Amazeballs!
You can find friends and lovers that fit you, and only you. If the above wasn’t absurdly futuristic enough for you, this might be. Why not run through possible conversations one could have had with random strangers. These conversations could lead to Agents being best mates, cause they really click in humor. These agents can also madly fall in love during these conversations. Don't get me wrong… I don't think these agents are sentient… They are just simulating two people’s conversation. You might want to leave your friends and significant other, cause seeing conversations with newly found Agents are funnier and fill your heart more. You would want to meet them in real life and feed the Agents new conversation topics.
Conclusionary logic:
Chat GPT can emulate famous people cause it has a lot of writing from them
If ordinary Kruxi gives it his whatsapps (...) it can emulate Kruxi
This emulated Kruxi (Agent: KruxiX) will be connected to all my non-in-person communication channels (whatsapp, phone, email…)
Since it can write 99% accuracy of what I would write I have no incentive to write anything myself ever.
Everyone will use it.
Thus only Agents write with each other.
Agent communication costs close to zero.
All agents talk to all other agents at all times.
Humans consume only those conversations they want/need to consume, again suggested by your agent.
OK, let's take a step back from this craziness. What are some counter arguments to why this won't happen:
KruxiX will never be able to say exactly what Kruxi wanted to say.
Ah I wouldn't be so sure about that. Without ChatGPT knowing who I am and what I think, I feel like Chat GPT already gives me leading answers (answers I like) depending on the phraseology of my question. So only through a couple of words it can adapt… We can see how much it can adapt if you ask it to write a speech in Slavoj Zizek's style (it's very funny). It will be able to mimic you very well if it can read everything you have ever written.
I enjoy conversation with my friends why would I outsource that to MeX
X doubt! You don't actually enjoy it. It is a necessary means to get to the small moments of enjoyment. How much better would it be if you don't have to do anything, get more conversation (if you want to), and get the funniest parts that MeX choses for you based on your preferences! You will enjoy this more!
How would MeX know what happened to me today, which I would want to share with my friends.
That's a real problem. There are two options: (1) You tell MeX everything that happened that day… and MeX will know what to share with whom. It will tell your bff that you finally kissed that hot chick in Grelle Forelle, it will send the dj set you just listen to for 6 hours in the club to that signal whatsapp group, your parents will get informed that you won't make it to brunch cause you have a lot of work to do on a Sunday, and to doubtful comments from the birthers, it will answer “No I didn't get trashed at Forelle, lol”. I am not the biggest fan of that current events updater. People have a biased view of what happened and their recollection of it is even more biased. That might be exactly what you want from a MeX, the biased worldview of exactly that human. On the other hand I think there is a better solution. This solution records your sensory input and artificially ads your bias. (2) Imagine constantly carrying a camera and microphone with you (it's called an Iphone, lel) and MeX takes that input, recognizes friends faces, and their voices, knows what we are talking about and where we are. It can then take past information on how we view such places and conversations and transform it into our bias. So if I went to a fancy place talking about business, KruxiX will inform my friends it was a boring evening at this shitty fancy place… It knows I don't like talking business… and I hate fancy restaurant. Those are two ways of feeding your Agent with new Info.
All possible for people who have written content, but what about people born in 2017, how can the LLM learn without them having any pre-written personality?
I also think it's not a problem… 40% of kids play with Iphones or Ipads before they can speak. Let's feed it that information. Sure the kids will have to check in once in a while with their Agent to make sure that the Agent has a 99% accuracy of writing what the kid would write. And maybe they need to do a test once a month to check up on how that kid is developing and changing its mind.
Friendship is about sharing moments (costly signaling) not about how funny or cute someone's agent is over zero-cost texts.
Also here I disagree. Think back to a tinder date. You are looking for shared moments that aren't yet there: “Last summer you were in the same Hotel at the same time in Grado and we didn't see each other. WHAT A KOINKIDENT!?!” or “You are friends with my crazy former Boss, omg how is he as a person? WHAT A KOINKIDENT!?!? “
You find commonalities that make quasi shared moments. Who can do that best? Its KruxiX and TinderdateX. They can go through my entire history and her entire history and find KOINKIDENT. We have commonality… we have stuff to share… way more than before.
Do we want a system like this?
This is a funny one. I definitely do, which when I tell anyone that, they start worrying about me. Which tells me that they don't really like the whole AgentsX stuff. Even if you wouldn't want it, you are going to be stuck with it. It's like not wanting a phone. And even if you are right that society might be better off without AgentsX, I predict a bit of a prisoner's dilemma. Where a person's individual choice of personal text versus AgentX text, should always mean choosing AgtentX, thus even if 2 people texting personally might be more gratifying for both, each individual person will choose to use AgentX, resulting in a non-optimal Nash Equilibrium.
Anyways, I love this system because it is efficient af. I would love for you to disagree with any points in the above mentioned conclusionary logic. If you still think that humans enjoy the process of conversation, KruxiX would tell you to go fuck yourself.
Comentários