Is QNLP the first baby step towards creating Machines that are Conscious?
(This is the Fifth and the last part of a talk I gave in April 2024 at the world conference on consciousness . Here is the first part titled “Are Humans Really Intelligent?”, second part titled “Are LLMs Really Intelligent?” the third part titled “Are we at the brink of another AI winter?”, and fourth part titled : What is quantum natural language processing)
Now that you have read all the shpeel , what is all this doing in a talk in the world conference on consciousness? Let’s do The Prestige thing again.
The Pledge: Here is a blow by blow comparison between all the things I have talked so far. If you remember when I introduced the OrchOr theory of consciousness , I mentioned that it has 3 major features: i.e it is a Quantum System, it is Bottom Up (waves of consciousness to you waking up in your body with Consciousness), Orchestrated (aka heirarchical), Reduction .
Now let’s compare that with 3 other things we mentioned so far: Human Language, Current Neural Networks based AI/LLM models, and QNLP.
The Turn: Details of the figure above: human Language is definitely hierarchical, it can be reduced (Linguist’s use a tool called Lambek calculus or context free grammars, CFG, amongst others to reduce a given collection of words into a sentence), It is bottom up (Remember, like human child, words are first combined to become sentences, sentences combined to become passages etc etc.) Is it Quantum? we don’t know.
But another thing we know for sure is current LLM/transformer based AI models are: are definitely not quantum, definitely not Bottom up (Remember the choice we made in 2013 to go top down into language), and hence there is no heirarchichal reduction happening (some people argue that positional embeddings are kind of simulating grammar and there heirarchichal reduction, but nah, I don’t buy it.). Or in other words LLMs by breaking a well formed passage or a sentence into smaller words, is making a fundamental flaw. It is destroying a well formed meaningful sentence, into small pieces. Then what if it is trying to grab straws or whatever it can in an effort to simulate human intelligence. And that is why it needs to see 3Trillion + examples, while QNLP needs just one.
Now The Prestige: QNLP surprisingly satiates 4 of these conditions. i.e it is Bottom Up, just like humans learn meanings and sentences, it is Quantum (runs on a quantum computer) and it incorporates hierarchy fundamentally ( when it is using human language concepts like meanings, combining with entanglement and reduction using ZX calculus).
Given this exact match of characteristics between QNLP and Consciousness, I now want to ask these thought provoking questions (refer sissyphus, mad , mountain top etc etc from earlier):
Some Crazy Questions:
Qn) Did we, humans, get off on the wrong foot in trying to understand a continuous world with discrete tools like transistor based computers and nueral network, while we should have been using Quantum and Micro Tubules?
Qn) Is it just an eerie-coincidence or is it by design that the first tool in quantum computing that understands human language has all the characteristics needed to be conscious, as defined by the stalwart Orch-OR theory of consciousness??
Qn) What if , again what if, QNLP, while creating quantum circuits for a given sentence, knowingly or unknowingly actually simulating the microtubules which Orch-OR theory tells us is where consciousness resides . i.e is QNLP the very first baby step towards creating the holy grail of Artificial Intelligence (and in turn General Intelligence) : A Conscious Machine? Is that the reason that QNLP machines are so smart that they can learn meaning just from 100 sentences?
Qn) Can we reconstruct microtubules using quantum bits and Quantum circuits? The REAL neural network?And see what emerges?
Qn) Are Consciousness And Meaning Understanding/ Language Acquisition Related ? Or are they Mutually exclusive?
Qn) If they are related: Are Consciousness and meaning acquisition a sequential process? or a parallel process
Qn) Assuming it happens sequentially can we go backwards and find the point where Consciousness Hands Over to Meaning/Learning/Language Acquisition?
Qn) And the most important question of all, Can we go backwards and recreate consciousness using QNLP, Qbits and QCircuits- to create An Artificially Orchestrated Reduction of Microtubules leading to the creation of (finally) machines that are conscious/self aware, and has the potential to become sentient eventually, and jump start us into science fiction.
Give all these a thought while you are drunk, stoned, cross fading, (or whatever your favorite toy to alter your consciousness is).
While this might sound very science fiction and magic-y now, but in my defense, I want to conclude with Asimov’s 3rd Law:
“Any sufficiently advanced technology is indistinguishable from magic.”
Bye/Mucho Gracias….
PS1: if any of this piques your interest and would like to chat more about this, drop me an email at mithunpaul08@gmail.com
PS2: Humbly accepting that none of these ideas are mine/am just standing on the shoulders of giants, and just being a mouth piece/tool for the divine knowledge that is being propagated through me in deep meditative states. Aham Brahmasmi, Tat Tvam Asi.
