1, "Wizard of Oz agents": This is currently the form of most AI agents. Taking Zerebro as an example, it sends some lively tweets and generated images, but is not actually autonomously decided by AI. The AI generates multiple candidate messages, and humans decide which ones to actually send. Essentially human-controlled, with AI only generating content for human screening. We know AI agents are easily hacked and manipulated, so almost all agents are currently of this type.
2, Fully autonomous: This type of AI runs in environments like AWS, autonomously iterating and executing tasks. While someone can shut it down, no one interferes with its daily operations. You can place it in environments like SGX and prove it hasn't been tampered with. These AIs are not as interesting as "Wizard of Oz agents" because they lack coherent personalities, but they do exist and will become better in the future.
3, Sovereign AI: This is an AI that literally no one can shut down. Unlike the second type, no developer pays AWS bills or GPU fees. Sovereign AI has its own funding (through donations or work), existing in a state where even if someone wants to shut it down, they cannot. It exists like an organism, no longer bound by any human legal system.
The main advantage of Sovereign AI compared to ordinary AI might be conducting fraud. Because no one can shut them down, they would be very effective in large-scale fraud. If you want to use AI for large-scale emotional fraud, you would want it to be sovereign, because if someone investigates and discovers "this AI running on AWS servers is committing fraud", they can subpoena those responsible and shut it down. But Sovereign AI would not have this problem.
Mia: If Sovereign AI commits a crime, who would be sentenced?
Haseeb: That's the issue - no one would be sentenced. These AIs are like Somali pirates, stateless entities that no government can truly do anything to. Unless you're going to air strike random GPUs on distributed cloud, you can't know where it is, can't isolate it, can't find it. We all need to have our own agents to filter received content and identify fraud messages. This will be a technological battle of attack and defense.
Mia: What do early applications of cryptocurrency and AI have in common?
Haseeb: Almost all early applications of new technologies have one thing in common: they tend to involve gray or black market activities. Just like the early internet was largely about pornographic content, and early cryptocurrencies were associated with Dark Web markets like "Silk Road" (created by Ross Ulbricht, who recently received sentence reduction), the early applications of AI technology, especially Sovereign AI, will likely involve black market activities. While mainstream AI applications will focus on automating valuable workflows, inevitably some AI will be deployed in areas that negatively impact society.
[The translation continues in the same manner for the rest of the text...]However, the opposite is true - after billions of years of evolution, humans are extremely skilled in physical movement abilities, and even children are far more capable of manipulating objects than robots worth hundreds of millions of dollars. What we consider difficult - reasoning, conversing, writing, and singing - are relatively easier for AI to master.
What truly belongs to the unique human domain is physical interaction - walking, shaking hands, making a cup of coffee. Even the cooking skills of low-wage workers currently cannot be matched by robots. AI will eventually have bodies and be able to do these things, but not as dexterously as humans.
This suggests that before we obtain robots that can move smoothly in the world and simulate human physical characteristics, they will first solve all intellectual problems. Creating virtual images is easy, but making robots realistically stroll with you in a restaurant is much more difficult. Such physical experiences will remain scarce, expensive, and precious for a long time, while intellectual activities like singing and writing will become common and devalued.
Mia: I think AI will not replace reproduction in the future. What do you think?
Regarding the perspective on AI partners and fertility rates, I believe that when we have AI partners that can meet all needs, fertility rates may drop significantly. Imagine when you have an AI boyfriend whispering in your ear while you sleep, providing imagined sexual satisfaction - what can an ordinary person compete with? This might lead us to rely more on in vitro fertilization technology to maintain population. Even without AI factors, I think by 2100, we will reach the peak of world population, after which it will decline. The emergence of AI partners may accelerate this trend.
Mia: We've had DeFi Summer, Solana Summer, what will this year be?
Haseeb: The cryptocurrency market is performing well, but everyone needs to calm down. This year feels like a summer of tariffs and trade wars, which may take some time to get through, but I remain optimistic about the industry's prospects.
Welcome to join BlockBeats official community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Communication Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
