Computer Science work is becoming more experimental Computer science has taken a turn more similar of the social sciences after the advent of the Language Agents. Less and less rigorous mathematical works relative to the broad line of empirical, yet still useful, work that is being published out there in the wild. Some would say that this is related to the basic impossibility to have strong guarantees on models that are just so intelligent that can do things that are deeply human in all of the history of humanity. Bounding on Human-ness? How could one put numerical bounds onto such human thing? Yet some things are clearly analyzable numerically, such as classical economical output (macro and microeconomics values), preferences (reccomendation systems), morality (see Alger & Weibull 2013 but it's still unorthodox), love (dating platform algorithms). In some sense, we are invading a deeply human sphere with the technique, and with measurability. We are pushing what is measurable to its limits, and pushing the rigor of the physics inside the analysis of the human and of the human things. On the one hand, this is a good endeavour: finally we can try to understand clearly what are things that have been only explored through misticism, literature, or religion, like feelings of belonging, love itself, and trust. On the other hand, can we really understand numerically deeply human things, or are we in some sense distorting them with this representation? Measures on Human-Things are not measures of Physical Things. Physics is clear, we can agree on a shared unit of measure, then the abstract number will have the same physical representation for everybody. But can we do that for the domain of the res humanae? This was the topic of long debate in the space of philosophy and I have not studied this topic enough. What I can see from a layman on the topic, is that from first principles, it's really difficult to compare the experience of a single person with that of another. Are the qualia deeply personal or transferrable? The computer science analogy is getting the specific ISA level instructions from one architecture and pretending the same set will work on another. This just doesn't work. The same instructions that are valid for one, are just not intelligible to another. In the same way, I don't expect these thoughts to be intelligible to everybody is reading this, yet it's my job as the writer of this blog post to make it as intelligible as possible. However, if correct interpretive work is done, it might be possible to end up with something very similar to each other. The reader can understand the point of the writer in a semantically close manner. Continuing the computer analogy, we need some translation layer that takes one ISA level, and transforms it at best effort to another instruction, keeping some invariance that we care about, for example mathematical computation equivalence. Some Equivalence Functions are Difficult to Compute (some are just not computable!) Yet, if we talk about the qualia experience, how would you even define equivalence from one's architecture to one another? This is a question that future computational-scientific approach to understand the human can answer, yet with the current research, as far as I know, is just impossible to answer. The computer-sciencey answer for program-equivalence, would just be that it's impossible in the general case. But that is not a satisfying answer.

What is Really Impossible? We, as humans, don't like impossible things Figure 1. Most of the things that modernity considers natural, could be once considered magic. Flying metal birds, remote communication thousands of meters, thermal cameras, and many more of modern commodities. Even theorems of impossibility are seen just as difficult stones to design around to solve one's problem. One classical result is the Halting Problem. We then just try to prove that for simpler families, or completely do something else that gives similar results, such as better error reporting, monitorability, better testing. Functionally in the specific use of the program, these approaches reach the same function, but structurally they were shaped by the theoretical results.

Modernity is Ignorant of its Limits It seems to me that the modern human has no idea of its limits outside of what can be proven mathematically. If something is physically possible, that should be tried. Historically, this thinking has given us much of the progress of modernity, but at the same time, we are not aware of the risks that it can bring. We are trying, as a society to push more and more. Some are trying to solve death as a physical problem, much of people in my field are scared of the risks that an AGI can bring (Bostrom 2014) and much of the progress is still on that direction. Wonders of technology before only present in the minds, now, bending through loopholes of reality's laws, becoming reality. This is progress. And per sé is not bad. We all can benefit from the good of these technologies, and life is objectively more fruitful in terms of objects, of services, of what was magic in the past and technology in modernity, compared to the past. We can indeed have enough richness in the global economy to provide to the totality of the people on planet earth a dignity in the living conditions. And this is good.

References

[1] Alger & Weibull “Homo Moralis—Preference Evolution Under Incomplete Information and Assortative Matching” Econometrica Vol. 81(6), pp. 2269–2302 2013

[2] Bostrom “Superintelligence: Paths, Dangers, Strategies” Oxford University Press 2014