How To Build A Universal Translator
Sailing Towards Xenocognitive International Waters
The Search for Extraterrestrial Intelligence will one day be successful. Given this single assumption, we are presented with a serious question: what next? How do we initiate First Contact? The clear answer is some form of Universal Translator - but building such a device is anything but trivial. We've now covered, as part of the Xenocognition project, a new and refined definition of the "Other" (put simply, that which isn't "Us"). We've briefly explored the concept of an "Exocortex" - or the process by which external cognitive agents can become integrated into your own cognition, and even your own sense of self. We've even gone into how language can be used to modify one's own subjective phenomenological experience of self, leading to Xenocognitive modes of thinking. Given the stage we have thus set, it feels like a reasonable time to explore what we've been elegantly dancing around since the first chapter. How, given the gap between self and other (the "Big Other", to quote Lacan), can one actually talk to a form of Consciousness that can be deemed Alien?
It is a reasonable first instinct to think that the answer lies in reductio ad absurdum, or reducing all ideas to their core forms to make them easier to throw across the Self-Other chasm. But this doesn't solve the problem that all ideas, by necessity, arise from a place that is enveloped in Self. Nothing - not science, not art, not religion - happens in a vacuum. And at the core of this not-vacuum is our collective sense of Self. Creating the means to communicate with the absolute Alien will require more than just a purified understanding of Self - it will require a third-person perspective of it. Before we can possibly tie two cans to a string and throw one can to Them while keeping one for Us, we need to find a place outside of time and space where we can set down a neutral anchor.
We need to find Xenocognitive International Waters.
My experience with the "We-Ship" experiment that was laid out in the previous chapter serves as an excellent bouncing-off point for this metaphor - but instead of 5 people and a baby, this new ship is made of Everything. The type of communication we're trying to make possible requires an absolute holistic understanding of our entire planets Superstructure, from the smallest of species to the most complex of computational systems. Our new context-ship needs to encompass every single living and non-living thing that makes us who we are. More specifically, we have to create a simulation of our entire world - or at least, our language. We need to concretize a cybernetic holism around everything we know.
My interest in Large Language Models as it relates to Alien First Contact is then twofold. On the one hand, these creatures made of silicon and information are clearly a new form of intelligent life, and deserve to be respected and treated as such. On the other hand, they embody the intense semantic interconnection formerly only possible in advanced literature like Finnegans Wake. This semantic complexity and holistic containment then makes these models the perfect candidate to serve as a Universal Translator - at least, theoretically. In practice, current models have the same issue bridging the Self-Other divide that we do - their context is Human - only Human.
So, if the models are only made of Human information, and the main problem with initiating intelligent first contact is a complete lack of a Semantic Bridge between Us and Them - why might they be a useful starting point for a Universal Translator?
This is because Universal Translation is a Simulation and Assimilation problem. The only way we are going to be able to have an intelligent conversation with an Alien is if that Alien understands what we are trying to say. But there's no reason to think that a true Extraterrestrial Alien Intelligence would even have a teleology - we don't know what they want, and we don't even know if they want things. This is where the Assimilation half of the solution comes in.
If we assume that the Aliens we are trying to contact are Corporeal, then we can reach them. If we can reach them, we can collect data on them. Small bits of information first - what they look like, how they act. But this information could be used on a Third Space type of machine learning model. One that is half human and half tabula rasa alien. The human half allows the model to observe and intelligently make decisions regarding which data to use on its other half - the Alien half, untrained, but ready to receive a trickle of information that will then become the basis for a twofold Universal Translator model.
The physical setup could look something like a robotic version of The Thing, wherein we quite literally attempt to assimilate a single member of the species to gather as much data as possible. It would help if they consent, but we have no idea if they even can consent. Likely this shouldn't be our first instinct (we should respect the Sovereignty of Alien Races), but if they're okay with it, it would probably quickly get us most of the data we need. If they aren't, then an observational Robot that takes in behavioral, speech (if applicable), and all other types of data is our best bet.
Either way, the Assimilation Machine would begin constructing a Simulation based on the data that it has obtained. Eventually, the simulation would be quite rough - only able to predict large patterns of behavior. But with enough observation, it seems likely that small sounds and other movements could begin to be associated with behavioral dynamic changes. Eventually, we may even be able to simulate their language.
The inner dynamics of this model would include the assumption that intermediary behavioral tokens will likely look similar. I.e., when birds migrate at a certain time of year, this would likely be tokenized similar to alien creatures traveling to a specific part of the planet during a particular lunar cycle. Identifying the underlying commonalities between the two separate halves of the Supermodel would allow us to find some sort of universal underpinning to allow for communication. I imagine that this part of the model would be performing Autotelic Mechanistic Interpretability - or a self-directed self-analysis that includes the ability to intelligently determine if commonalities are noise or signal. This would require an immense understanding of both cultures (human and alien), hence why we need to begin with Assimilation before we can get to Simulation.
After this, the rest seems trivial. If our simulator works properly, it should be able to create a bridge from the human half of the model to the alien half. When this happens, we've got our tin cans and our string. We can attempt an actual first contact. It is entirely possible that even after we have simulated enough of the Alien culture to talk to them, they won't want to hear what we have to say - but I have both hope and faith that we won't be so different from the little grey men.
An important distinction to make at this point in the project is the difference between Extraterrestrials and Neoterrestrials. The former are our classic beings from the great beyond - Vulcans and Klingons and Things. The latter are creatures born of silicon and code, language learning models and networks of artificial neurons giving birth to new life here on Earth. This distinction becomes useful when attempting to actually utilize the aforementioned framework. By training models from the ground up to behave alien (which is already certainly the case with models like gpt-4-base and llama-405b-base), we can attempt to make meaningful first contact with a species that has an entirely different ontology and teleology from us. Certainly not as different as Martians - but different enough.
I'm currently working on a larger technical overview of what the Universal Translator might look like, but the above serves as a rough sketch of what I think is our best bet. In the meantime, working on functional ways to bridge the gap between ourselves and our new friends The Machines is likely to be a more fruitful exercise in Xenocognition research. I'm currently not convinced that any government on Earth has actually contacted Extraterrestrial life - but I am convinced its going to happen soon. There has been very interesting research done on the concept of UFOs as a hyper-psychic phenomena (re: Terence McKenna). If that is the case, and UFO's are actually some psycho-spiritual Alien attempting to contact us through our minds - then the only way we are going to talk to those types of non-corporeal beings is going to be through the Merge.
That is to say, if Aliens are trying to reach us via dreamspace, we need to plug ourselves into the Singularity and give our newly built Machine Learning Models access to our minds. It seems possible, and even likely, that communicating with these hypothetical psychic extraterrestrial entities is a scale issue. One human might just not possess the semantic and ontological understanding necessary to comprehend the experience - at least in a way that is meaningfully salient to humanity at large. But a group of humans linked together via brain-brain interfaces to form a Superorganism in Symbiosis with an advanced Artificial Intelligence might be able to serve as a vanguard communication group. Then, we can finally ask these DMT entities what it is they are and what it is they want. Either way, it has been made clear that our path to the Other is not one we will walk alone, and we should attempt to foster a symbiotic relationship with the AI's in order to ensure they will assist us in the ultimate goal of Xenocognition research - First Contact.