Tuesday, April 04, 2023

Scalable is not intelligence nor computation consciousness


 We really are not much impressed by Large Language Models (LLMs) -popularly ChatBots, the latest version in fast accruing scene: ChatGPT4. Although I do identify with language learning strategies through approximation, and yes GPT4 is far better than me when it comes to grammar skills and sentence construction, but at the end of the day it is just scalable approximation working in the context to get meaning. It is a stochastic parrot. It is quite easy to contextualize information to create knowledge. So, it is possible to achieve self-knowledge, determination of self in context. Consciousness is an entirely different matter. It’s an array of triggers and possibilities that come together in a matter of increased chance, which can only be traced in retrospect. Significant aspect of being a human is the ability to reflect, reflect on what is understood. Rather than being in constant loop of gathering for probabilistic knowing reflection is a solitude matter, choosing thoughts, connecting ideas, sensing reality in its unique diversity (and here body too plays an important role that AI lacks), valuing through experience, creating convictions, sophisticating through time. To focus is to choose, to choose what you care and value. An algorithm working on scalable pattern is incapable to understand what to value. It will move with the herd. It will solve problems to satisfy herd and keep them rivetted. It will replace template of mass entertainment and consumption. It is incapable to fathom uniqueness of individual. AI chatbots therefore are threat as it can manipulate, weaponize disinformation, assist in plagiarizing so on. The deep learning is misaligned through biases, prejudices and power dynamics. AI is set to exaggerate crass and awe us to fake. The concern raised by prescient minds at the cutting edge of AI ‘to pause AI experiments for six months’ is valid. Laws have not kept up with significant ethical challenges and possibility of social strife threatening to seriously undermine democratic institutions and individual liberty. Repeated dehumanizing degradations that is classified as jobs are being exposed, so is crude crass mindless entertainment driven consumption pattern. I see AI being a serious threat to herd churning systems like market and religion. The threat of herd getting excited through fake and disinformation is real and will keep humanity at its wits for a long time to come. Lacking faculty to discern they will fall for the manipulation. Lacking compassion and critical thinking sensibilities they will get violent. Education seems to be the only savior as it has the potential to retrieve humans from herd. Meditation, self-reflection, as also associating with thoughts that elevates us to human condition will help to understand context of life hence sharpen sensibilities and faculties to discern context. Any sensible person must have watched rewatched thought rethought Tarkovsky’s Solaris. Tarkovsky notes about the protagonist "For me it was important that he would be just like that. He should be a man of a rather limited spiritual range, average- just in order to be able to experience this spiritual battle, fear, not like an animal which is in pain and does not comprehend what is happening to it. What was important to me was precisely that human being unconsciously forces himself to be human, unconsciously and as far as his spiritual abilities would allow, he opposes the brutality, he opposes all that is inhuman while he remains human. And it turns out that despite him being -so it would seem- a thoroughly average guy, he stands at a high level spiritually. It is as if he convicted himself, he went right inside this problem and he saw himself in a mirror".

  AI will force us to reevaluate what is to be a human. Is it that what is keeping us occupied making us more human or is it reducing our humanity, human faculties and sensibilities? It is a much needed mirror that humanity is forced to look into, a pause to evaluate human civilization. It is quite clear that AI lacks significantly to become an AGI (Artificial General Intelligence), even a larger model of AI will fall short. It needs a technological breakthrough of sorts to reach human level AI i.e. AGI, while creating consciousness is much further from human grasp. Hence the concerns raised on “misaligned superintelligent AGI” cannot really be taken that seriously. Indeed, LLMs or low-level utilitarian chatbots are not what AI is all about, it may amuse some, but the astounding possibilities are in discreet setting -limited frame of reference. AI is bringing out aspect of reality that human beings are not able to even perceive let alone conceive whether discovering antibiotic Halicin -in the process glimpsing properties of molecules that we are not aware that AI found scalable, or AlphaZero showing possibilities that no human could think in last many centuries of playing chess or AlfaFold solving protein folding problem. These are example of AI’s true potential that is exploring and discovering patterns that is beyond human reason. It is creating a black box which human rationality have no access to. Scary though it is AI in discreet setting hold immense potential for the future but that shouldn’t be in any way confused with AGI. AGI, that is human like AI, is quite impossible with present technology and hardware mechanism of binary computers -that is deterministic with limited flexibility and adapting skills. The concern regarding ‘misaligned superintelligent AGI’ doesn’t really pass basic scrutiny. AGI needs a breakthrough technology, a conscious AI seems quite remote at this stage (some are even making post-singularity predictions!!). There is a exponential increase in computation capability and complexity in recent times that is ofcourse getting us closer to reality but consciousness is not a physical process which can be described by computation. Computation is based on discreet elements, specific actions, contexts, rules, and is confined despite ever increasing computation possibilities. While consciousness is part of a continuum. It is symphony at the edge of chaos. The murmuration of starlings and school of fish evading predator are vibrant conscious continuum. There is an understanding that connects, an ambiguous perception that resolves as conscious then exhibits, blends and vanish, in eternal iterative loops as it were.

Unlike binary based AI which is essentially accumulative growth, indeed pathbreaking, the shift from binary computers to quantum technology is a paradigm shift. This transition is ‘extraordinary science’ (Thomas Kuhn), revolutionary like Copernican, Newtonian, Einsteinian etc. Breakthrough ideas and technology creates revolutionary transition, that necessitates abandoning of existing paradigms. This is how nature too works, more than Darwinian natural selections it is mutations that have played significant role in evolution. It's syncing of life with emergent, of survival understanding as chance in chaos. Even at geological scale Cuvier’s theory of cataclysmic events -world altering catastrophes, as cause of mass extinction plays more importance than Lyell’s gradual uniformitarian vanishing of species. Homo sapiens, indeed mammals, got a chance at life through asteroid strike wiping out dominant reptiles -dinosaurs, sixty-six million years ago. And now humans are the cause of cataclysmic events unfolding. Sorry for the digression but the center point of argument here is ‘misalignment of AGI’ is quite impossible since the technology involved in AGI is break from past. Large generative algorithm created AI will ofcourse carry misalignment that is embedded in the society. Most countries across the world, indeed EU, are working on regulation of AI ethics.

Knowledge is about probability and contingency. It’s probable because of the available data and critical faculties of sense contingent to the situation. The reason why Karl Popper is a compelling and relevant philosopher. Current understandings are evolving and is in eternal iterating loop with reality each individual faces. This is the journey of each individual for his or her truth, and in the sense contribution to collective truth of humanity. It is impossible to know reality perfectly. It is always an approximation that keeps updating with more inputs and insights. Generative algorithm, machine learning based on deep neural network, is also working on increasing probability. Bayes theorem is a mathematical equation to probe the probability and increase the chance. It’s the backbone of machine learning. It’s an iterative learning to keep increasing the probability and getting approximation closer to reality as the prior is updated with inputs from each event. Now this can only work if we are open minded since certainty makes learning impossible. Every moment in life is an experience that contributes to diversity of experience hence our approximation of our reality that is unique and cherished as expression of life. Life is probing into deeper conscious and possibilities. Even though quantum breakthrough will help access reality at molecular level hence push scientific breakthroughs but conscious AI is still way ahead since quantum too is working on computing. Even basic quantum is based on Schrodinger’s equation which means it is still computable i.e. deterministic. It’s the breakdown of quantum is where the possibility of conscious, the collapse of wave function (measurement as consciousness), reducing of quantum state. This is the building block of proto conscious which will provide the building block of conscious according to Roger Penrose. He argues on conscious as non computable hence non algorithmic and speaks about Orchestrated Objective Reduction -a philosophy of mind wherein consciousness originates at quantum level inside neuron rather than product of connection between the neurons or synapses. So, negating the idea of conscious sourced from computation at synapses hence brain output -neurological data, as input from consciousness.     

Computation is based on mathematics and logic herein. But much of math is constructed on unprovable true. It is based on axioms -undeniable statement assumed to be true. Theorems and proofs are constructed from axioms using rules of inference and logical deduction. There are therefore true statements that are impossible to prove. Hence logical systems will have unprovable trues. This is self-referential paradox. This is the famous Godel’s Incompleteness Theorem. “The revelation of the causal power of meaning in a rule bound but meaning free universe” (Godel, Escher, Bach: An Eternal Golden Braid by Douglas Hofstadter) “…Twisted loop of selfhood trapped in inanimate bulb called brain”. Infinite in finite. Hofstadter, unlike Penrose -biology based consciousness, argues that focus from material component to abstract patterns allows the quasi magical leap from inanimate to animate, from nonsemantic to semantic, from meaningless to meaningful, to take place. This abstract pattern is 'strange loop' -that Godel brings out through his theorem, Escher through his drawing and Bach through music. Whether it is biological based (quantum tangled neurons) or non biological (strange loops -complex set of stimuli drastically reduced into a concept) one is not sure nor competent to comment. But what is clear is all logical construction is based on reliability of proof procedure. This reliability is based on not provable by the rules that is given, the meaning impervious. Rules have to be trusted for the logical outcome. This understanding of why the rules gives you truth enables you to transcend the truth. This understanding is outside the rule or computable framework, hence is indication of conscious. It is where the system stands back and think about its own thought process. What is the reason to think in such a way? What is that what I am thinking? This self-referential paradox is self-reflection, hence awareness and understanding, hence consciousness.   

Post Script: Ray Kurzweil proposes AI consciousness by 2029 through LLM. Ofcourse his idea of consciousness is philosophical! Singularity by 2045. Also, recent research into role of glial cells adds to new dimensions.