Avatar Intelligence: The Next Stop in the Web3 World
2023/02/13

Thanks to technological advancement, AI can do more and more incredible things. As early as 2011, IBM's supercomputer Watson defeated two of the best human players, Brad Rutter and Ken Jennings, in a trivia show called "Jeopardy!" In this competition, Watson scored higher than the scores of the two players combined.
Watson's knowledge was not coded into its memory. Instead, it had educated itself by reading Wikipedia and several other encyclopedia websites (all natural language source materials). However, Watson didn't need to read through all web pages as a human would. After reading one page, it might have already reached itsconclusion that, for instance, the probability of Argentina winning the World Cup is 36%. You might also read the same page and conclude the probability is 80% because you are better at reading comprehension. Watson, on the other hand, compensated for its disadvantage in reading by absorbing way more source materials. It has an excellent Bayesian inference system that could integrate all the indexed information to reach a more reasonable conclusion, which takes little time.
Bayesian inference systems have played a crucial role in the development of AI since. Whether it’s AlphaGo or ChatGPT, we can see a glimpse of Bayesian inference in them. These AI technologies have expanded our horizons, enabling virtual worlds such as Web3 and the Metaverse.
Today, when we talk about the Metaverse and Web3, one fact we cannot ignore is we are all carbon-based organisms and cannot live forever in a virtual world completely detached from reality. We need a way to exist in the virtual world for longer than our physical existence, or even indefinitely. There needs to be a solution that turns the virtual world depicted in the Matrix films into a reality, which brings us to AvI (Avatar Intelligence).

AvI stands for Avatar Intelligence. While traditional AI focuses on training algorithms to achieve abilities like those of humans, Avatar Intelligence aims to recreate the Matrix films by uploading our consciousness to the network or building a program to replicate the thought patterns of users. This means that each person will have a unique digital avatar in the Metaverse or Web3, which can perform various tasks independently under users’ authorization. This will be the ultimate use scenario for Web3.
Before Matrix proposed the concept of Avatar Intelligence, another idea that is quite popular is Mind Uploading.

Mind uploading uses brain scanning and conjecturing to simulate the brain’s state in computers. The computer will simulate how the brain processes information and how the entire neuronal system works to generate a level of awareness like the human brain.
Mind uploading sounds great, but there are two problems:
The first step is to recreate the interconnected state of all the neurons of the brain. To do this, the brain has to be sliced into pieces a million times thinner than a slice of orange. Information from the brain is stored in every nook and cranny of the physical connection among neurons. Their size, shape, and the number and location of the connections are all relevant to the information we are trying to retrieve from the brain. We have to obtain all these data to create a computer simulation of the whole brain's neuronal state.
Cutting the brain carries high risk, and even if there are users who agree to have their brains sliced into pieces, it would be difficult to perfectly reassemble them back to their original state. Thus, it is difficult to imagine this technology being widely adopted.
If we want to create a computer that operates like the human brain, it must be able to access all stored data in a short amount of time, which means the data have to be stored in RAM rather than on hard drives.

In our experiments, if we try to store data collected from lab mice in RAM, it would require a storage space 12.5 times that of the largest single-memory computer (a device built for storage rather than processing) in history. Not to mention that the human brain contains roughly 100 billion neurons, which is as many as the number of stars observable in our galaxy and a million times the number of neurons in a lab mouse’s brain. This difference is even wider when it comes to the number of connections, which is equivalent to how manygrains of sand there are in a one-kilometre long and two-metre deep sand beach.
To further complicate matters, we don't even know how much information the human brain can store, not to mention what it would take to transfer such information to a computer. We need to transcode the information first and keep it in some kind of storage for computers to access. Any errors in this process would be fatal. If we don't know how much information needs to be stored before the transfer process begins, we may use up all storage space before the transfer is finished so that the information sequence may become damaged and unusable. Furthermore, since this is the equivalence of uploading someone’s life, we must have at least two (if not three) backups. Otherwise, the consequences of losing such data could be catastrophic.
All this is to say that in the short term if we truly want our digital avatars created in a swift and convenient way to explore all the wonders of Web3 and the Metaverse, we must enlist the help of AI, and thus Avatar Intelligence.
Last modified 1mo ago