### How Conscious AI Works (in theory)

Alan Turing, one of the major minds behind the modern computer, coined the term the “Turing Test“. It is a threshold for answers obtained from a computer simulation to be indistinguishable from those produced by a conscious entity. This test has been passed under certain evaluations, but this does not mean that the program which provided the responses possessed consciousness.

This program has sentences as input and sentences as output, which we will designate as S and S’ (S prime). Any computer program (also called “model”) that will be used to simulate consciousness must be adaptive so that it can reflect the human consciousness’ capacity to learn.

### Example: Language Pulveriser

I will give an example of a language parametrisation algorithm which will first be used to identify languages, then translate sentences, then finally we will attempt to use this parametrisation to answer questions.

We have previously shown how we may perform general optimisations as well as given examples of pulveriser functions. Now, we will show how to construct a parse metric which will allow us to elucidate the problem inherent to artificial consciousness emulation. We will accomplish this by showing that the complexity of the residual (computation remaining after the algorithm has performed its function) is equivalent to the complexity of the original question. This means that (even in the most generic sense), there is no way to conclusively code consciousness because we do not have any means to encode the calculation in a manner which can reduce the calculational complexity. If we cannot reduce the computational complexity of a problem, then we cannot meaningfully deduce new information from successive computations. That is: any consciousness emulator will not ever satisfactorily give the impression of having a cogent personality, (i.e.: consciousness) without human interference.

Consciousness is indivisible – it is a single quantum potential form. There is thus no way to simulate a quantum potential form with a transistor-based computer.

### The Grand Canonical Language Pulveriser

In order to canonically pulverise a system, we must be able to prove that we have derived all possible information from the system. Thus, we model a language as the set of all sets of series of letters, which we will call the form archetype sets . The first form archetype set would simply be the alphabet, the second would be the set of all 2 letter sets. In the case of English it would be {aa, ab, ac, … , zz} and so forth. We can see that the cardinality of the set of successive form archetype sets is n, n^2, n^3,…, n^a where a is the final term of the series. So we therefore see how a equals the length of the longest word in the language. This pulveriser function therefore includes all possible form archetype sets (combinations of letters) in the language.

### Algorithmic Implementation

We will use the English language. We will also assume we have a library of books sufficiently large as to convey the ethos of the cultural zeitgeist. We will also assume that predictions about what successive word forms are the best approximation to the state of human consciousness, because the books themselves were written with the aim of simulating human consciousness for a human audience.

We first compute the set of all form archetypes of these books and generate a statistical distribution of all forms. We can say that this computation can be known exactly because there exists a finite number of words in these books. We presume this distribution to be known and to have a matrix representation: M.

The consciousness emulator takes a sentence and canonises it as a set of form archetypes. It then searches the database of all possible combinations of forms and finds the likeliest form, returning the result of greatest likelihood which is also a true word. A word: W is considered true if it satisfies the criterion of existence, that is: it exists in the dictionary.

We thus impose that our algorithm includes all forms of punctuation as letters and impose the rule in the consciousness emulator that each time it simulates a period as the next likeliest form, then the result is truncated at the period and and the statement is outputted by the consciousness simulator as S’.

There is no guarantee that the set of forms (S’) of greatest likelihood to succeed a particular set of inputted forms (S) actually makes any sense though. To obtain a reasonable reply, the computer would need to do is to generate a set of the 10 likeliest sentences to succeed S and then have a human decide which is the ‘right’ answer. Thus we are right back at the problem of needing human intervention to answer the question, so you might as well just cut out this intermediary computer!

Thus although a good consciousness emulator can be created (by implementing the explanation above), it cannot reliably return an answer which indicates an entity with introspection and self-awareness. Thus though our model may pass the Turing Test, it will never pass the “True Ring” test, because of that elusive element in human consciousness indicating the existence of free will. There is thus no way to canonise the decision making process with a transistor (calculator) based computer.

### Example: Language Identification Algorithm

The form archetype language pulveriser can be used for a great many practical applications, which I will give now an example of with a language identification algorithm.

We first compute the set of all form archetypes in the dictionary. We then rank them in order of frequency in a histogram. We then approximate the histogram with a Fourier Series and normalise the resulting function to have an integral area of 1 exposing our ad hoc presumption that all languages will have the same information content.

*Given:
*A spoken sentence: S.

A phonetic language database of form pulverisers: P

For simplicity, we will assume that we have 100% accuracy in speech to text. This is not realistic, but it is realistic that individual syllables could be identified in a particular recording (by a human), then translated to the phonetic alphabet at which point the information will be in a form that it can go into a particular implementation of the form archetype pulveriser.

If I want to identify what language is represented by a particular set of sounds, I must input it into each language pulveriser function and find the language which maps that particular set of sounds to a meaningful sentence with highest probability. That is: out of the set of all sets of forms in the set of all languages, how likely is that particular set of forms, per language, per all possible words in that language? The largest result of this computation is the identified language.

### Challenge

The language canoniser has a small computational design flaw. Find it.

Clue: Consider the problem of generating the first word of a sentence.