Artificial People

I used to think that “virtualizing” people would be one way for humanity to reach immortality. This has been explored in Sci-Fi, for example in Neuromancer where this device called a construct can hold a person’s personality, with whom you can then later interact past their death.

It is also already possible to interact with chatbots, and in certain circumstances it becomes difficult to tell if you are interacting with a machine or a human, which makes you wonder if Turing’s test (which is supposed to determine if a machine has become “intelligent”), is not already soon going to be made obsolete.

Machine learning algorithms such as GPT-3 could be trained on someone’s entire digital corpus (e-mails, texts, blogs, social media activity, etc.) and later be used to generate text which would be difficult to differentiate from the original author’s own production. This has already been explored in the Be right back episode of the popular show Black Mirror. And if you think this is mere fiction, think twice! Apparently, Microsoft is working on something identical.

Those approaches are somewhat limited though, because it’s hard to argue that a person can be replicated based only their digital production.

Another venue to virtualize people would be to scan their brains down to the smallest details, and then somehow have them run in a virtual environment. Kurzgesagt explores this topic in Can you upload your mind & Live Forever. Even if we are making progress towards being able to map individual neurons and their connections, nothing says that this is enough to capture precisely someone’s essence. But let’s say that it does. It still seems like there are something like 100 billion neurons in a human brain, and together they form something like a million billion connections. Truly mind boggling. Will we ever be able to make a copy of this run in a computer? Not sure at all.

OK, so maybe we will not be able to clone existing people and have them live forever in a simulated reality within a computer. But what about creating brand new “people” directly in a computer? Trying to simulate real biological brains for that purpose would be a self-imposed limitation. After all, deep learning started as an attempt to replicate biological neural networks but most progress was made when it diverged to become a mathematical abstraction increasingly disconnected from its biological counterpart.

Some neural networks architectures such as GANs are designed so that they will learn on examples of something, and then proceed to create synthetic instances which are increasingly more and more difficult to differentiate from the authentic ones. You can use GANs to produce real looking faces of people, and it is not such a stretch to imagine some similar technology could be used to create Artificial Intelligences which would be increasingly difficult to distinguish from real people. An AI to create AIs reasoning like humans, so to speak.

It is not very clear what technology the Unreal game engine uses behind the scene, but their MetaHuman Creator tool is now capable of “creating a bespoke photorealistic digital human, fully rigged and complete with hair and clothing, in a matter of minutes”. Just to be clear, we are only talking about physical appearance here.

So, in a few years, we could have virtual humans who would look like humans, and maybe think like humans. If you were placed with them in a virtual reality setting, you would have a hard time telling who is an AI and who is another genuine human just like you. And why should we limit ourselves to virtual reality? What about having that AI be running in a robotic body as well? Sure, you would not confuse Boston Dynamics’ Atlas robot for a human, but you can argue that it does parkour better than most humans. Give it a few more years!

Maybe those virtual humans would start off as virtual kids and grow up and learn from interacting with real humans, hence developping human like behaviors and reasonning. Adopted artificial humans. They could look similar to their biological parents. They would learn from them. Biological adoptive parents would impart some of their elements to their adopted artificial kids, educate them, transmit values to them, just like you do to actual children. If parents go on living through their children, maybe it will be the same with those artificial children?

While capturing someone’s essence and making it immortal might very well be out of our reach, it seems slightly more likely that a transition over many generations from biological carbon-based humans to artificial silicon-based humans could be humanity’s way to immortality.