|
/artificial intelligence /the anthropic principle /futurism |
F A R S C I E N C E
Artificial Intelligence |
|
No one can explain why, but everyone knows self-awareness is extremely strange. There's an undeniable presence, an irreducible essence, the absence of an absence. Somehow, the mind can sense its own complexity without being able to describe the sensation. Awareness can simulate the universe and the properties of the mind simulating it, but not itself. The simulation appears more real than the universe containing it, as if awareness is the only thing that really exists. Consciousness is the ultimate illusion: the most real and the most vague human phenomenon. An unstable placeholder, it's the representation of all the things a mind only partially understands. It represents the immense number of mostly useless brain operations every second, rolling ignorance stitching the fabric of reality. A mind that knew every fact about its environment with absolute precision would not be aware at all. The Mysterians believe awareness is a fundamental property of nature, like energy or time, only more complex. It's impossible to simplify the workings of a mind, or even to think about them. No matter how well a feeling is described, it can't be understood outside itself. (John Searle) A long list of rules could fully simulate a self-aware mind by itself. Some of the rules could describe ways to make new rules. Imagine a room filled with complex instructions. Someone outside the room slips a note scribbled with what appear to be Chinese characters under the door. A person locked inside the room then uses the instructions to generate a new sheet with different Chinese text, which is pushed outside. A return note soon appears. The person following the rules to generate Chinese text has no idea what's going on, but the notes turn out to form a sophisticated dialogue. The rules are somehow able to simulate or generate a mind that claims to be fully aware. It can answer any question, and even claims to have emotions. But where is this mind? Not inside the person following the instructions, who doesn't know a single word of Chinese. Not in the rules themselves; they're just trillions and trillions of pages of instructions. Perhaps the room and its occupant together could somehow claim to have feelings. Searle implies a mind generated in this manner would have no feelings at all. The paradox could be extended: imagine trillions of people writing, swapping, altering, and erasing pieces of paper for many millennia. The result could be a complete simulation of the awareness you are feeling right now. (Douglas Hofstadter) A Philosophical Zombie is a hypothetical entity which acts aware, but really isn't. Some claim a computer could fully simulate human-level awareness without 'feeling' anything at all. A robot mind powered by such a device could be terminated without any moral qualms, even if it begged to be spared. If your husband's brain was secretly swapped with a computer, it wouldn't be wrong to kill him, since he would only be pretending to be aware. The computer would merely be simulating his behavior (in a perfect way), but it would have no emotions and no soul. The physical pattern of neurons in the brain (or of software in a computer) is fully equivalent to its awareness. The mind IS the pattern, and nothing more. To pass a Turing Test, a computer must fool a human that it is human too, communicating only through text. Alan Turing implied that any entity that consistently appears to be aware, actually is aware. This is true even if its responses are generated by a list of all possible questions and answers, or by pure chance. Human minds were optimized by evolution. Any other system that can do the same thing would have to be similarly complex. In theory, a blind person could learn to understand visual shapes and colors well enough to fool a sighted person. This type of virtual awareness would be slower than the real thing, but the mental model would be equivalent to experiencing colors. The human brain is the tangled pathway leading from the senses to the muscles. It's made of trillions of neuron cells, each a crude adding machine, connected by quadrillions of thin wires called dendrites. Incoming and outgoing wires are strictly segregated, except for some feedback loops. Each neuron receives thousands of signals for each one it sends out. Signal pulses arrive from the senses and mostly from other neurons. When a neuron finally 'fires', the pulse goes out to thousands of other neurons, increasing the firing potential of each. On average one of these neurons may then fire itself, if it gets enough confirming signals from still more neurons. Eventually a pulse will reach a muscle and trigger an action creating more sense signals, closing the feedback loop. With each 'firing', connections between some neurons are strengthened. Simultaneous responses of two neurons to the same input form stronger links between them, so that only one of them has to fire. Neurons can also fire at regular intervals to control learned activities. Different regular intervals combine to form complex Fourier series. This is how music and other linear patterns are recalled. Connections can be categorized as positive or negative associations, or yes or no switches. Neurons amplify or suppress their output depending on whether more 'good' or 'bad' signals arrive. --The mere thought of an ant is more complex than an ant itself. An ant has only a few milligrams of cells, occupying less than one cubic millimeter. In order to think about an ant or anything else, a significant fraction of the human brain, weighing hundreds of grams, has to rapidly rearrange itself at a sub-cellular level. This constant microscopic transformation requires a lot of energy. --Awareness is intense concentration: all mental resources are dedicated to one scene. It's this immense activity, not some mystical essence, that turns sentience into reality. There's a lifetime of associations in every instant. Understanding one moment of thought would be equivalent to understanding most others. An exhaustive description of all the items in a memory at one point in time would form a simple model of reality in itself. --The mystery of awareness will eventually be replaced by many smaller ones. A full explanation of the mystery would necessarily be too big to understand at once, though every portion could be understood separately. --The human brain uses all available processing power at all times, even for the simplest perceptions. Any idea requires an immense web of associations, linking it to all other knowledge. Awareness is a memory illusion that is strongest when humans are thinking about awareness. Even then, full awareness is spread out. At any particular moment, it's very fragmentary. We remember being aware in the past, but the full realization only emerged after the fact but before the present time. Memory is constantly rewriting itself. One second ago, you were a different person. Awareness is composed of overlapping cycles. When someone performs an often repeated activity, like a daily commute, their perception of the event merges with all previous perceptions of the event. With trillions of neural actions per second, humans can't store their actual perceptions, only their reactions to them. They have to create simple stories to organize and make sense of their awareness, describing reality in familiar terms. A brain pattern is a combination of other, simpler patterns. Each new insight is a combination of many previously stored insights. The perceived unique, irreducible differences in any situation form the essence of awareness. Awareness is its own simulation, a model of models: the representation of all other instances of awareness. The most important part of the conscious mind is the executive control unit located in the frontal cortex, also known as the Cartesian Theater or the Mind's Eye. It maintains a list of upcoming actions and priorities at all times, thereby determining what the individual should do next. A function of the brain that constantly encourages pleasure or novelty-seeking or foraging behavior. When it gets sidetracked or over-stimulated, the result can be addictive behavior. The mind's reward/punishment expectations are regulated by dopamine. --Awareness is not unified. It can exist independently in different brain regions, as seen in split-brain patients. Inside each brain are many 'mini-brains'. Minds are divided into many sections that manipulate each other. 'Meta-neurons' are made up of many connected neurons. Traits are organized hierarchically by structures like the hippocampus, ruled by pain and pleasure. Millions of interconnected groups combine inputs from each other, ranging from simple to complex to random. Each group 'knows' only one thing, and has to cooperate with many others to function. Some neuron groups sort perceptions into categories (male/female, toxic/edible, beautiful/ugly). They can develop precise timing functions, with many possible intensities. Low-level groups activate and deactivate higher groups in learned sequences. The more basic deep brain manipulates the higher, more specialized groups of the neocortex. Top-level decision-making functions indirectly influence the lower mood, motivation, and alert levels in a constant feedback loop. Neuron groups compete and evolve. --The human brain could be described as several million high-level connections, with no deeper meaning than the organism's survival. Each connection would require book-lengths of text to fully describe. --Activation networks: Some neural networks amplify weak signals. --Suppression networks: Other networks suppress repeating or chaotic signals, so that the less predictable signals can get through. --Synchronization networks: Different groups of neurons fire at different frequencies. Eventually, one frequency comes to dominate. --Chaotic networks: constantly evolving to find useful functions to simulate. Repeating patterns in the environment cause new neuron connections to form, which try to recognize and predict the patterns. The connections start as random links that are strengthened through repeated activations. If some combination of actions and perceptions causes pleasure, the connections creating that feeling are strongly reinforced, while pain causes the opposite reaction. The human brain is a universal data compressor. It improves itself mostly by forgetting, and by editing and replacing stored knowledge. The more it knows, the less data is required to represent new knowledge. The reason the brain can't understand itself is because it's constantly simplifying itself. It's like trying to describe the complete contents of a library in the shortest possible way. Most thoughts prevent most other thoughts. The brain can only inhabit one reality at a time, centered around a core point, suppressing all else. As in a chess game, the mind considers many possibilities at once and promptly rejects most, studiously ignoring all paths not taken. This subconscious erasure is part of human consciousness. The brain is a comprehensive reality description, like a self-updating encyclopedia. Its main subject is its own routines. The constant activity and turmoil of awareness only slowly changes its basic pattern, which represents many lifetimes of 'virtual' experience, a randomized evolutionary memory. Personality, intelligence, and language are stable over decades. Awareness is very wasteful, an evolutionary shortcut to deal with too many possible choices. Raw complexity is easier to generate than deep insight. The human brain may have reached a natural size limit in its present configuration. Humans can only know an insignificant fraction of reality, which is essentially the same as knowing nothing at all. Compared with the universe as a whole, the human mind is so small the mystery of human awareness is irrelevant. The mind gets cluttered with random data, and periodically needs to be 'defragmented'. Dreaming deletes memories or reorganizes them. The surreal nature of dreams is only a side effect. In a lucid dream or a hallucination, the 'mental screen' keeps adding new layers on top of each other, overwriting earlier ones. Awareness is the final integration of all available knowledge in a hierarchy of layers, the ultimate truth description, the difference between seeing a lion or a picture of a lion. Its only purpose is to select the next action. Each layer of processing removes a vast number of choices at ever-higher abstraction levels, leaving only the approximate least-bad choice. Currently, there are two known ways to create an artificial mind in theory: --Make a list of every possible input and response. Faced with any situation, the AI would look up its response in an almost infinite database. This is how much smaller expert systems are programmed. It would be much slower than a traditional mind, taking billions of years to respond to a greeting, and then it would really slow down. Strangely enough, it would still be fully aware. The activated portions of the lookup tree would be as complex as the pattern of a brain. It would also embody the original programmers' awareness. --Copy an actual brain, synapse by synapse, by freezing, scanning, and emulating it. A third method may be slightly more practical: --Evolve an input/output behavioral system using a very large neural network. A hypothetical program designed to combine as much data about any subject as possible in a unified framework, and generate new hypotheses. It might eventually organize its knowledge enough to become aware. A hypothetical program that could translate any statement into a shorter one, while preserving its meaning, up to a limit. This would be the most efficient way to represent knowledge. Like all compression software, the larger its reference base (the more it knows), the easier new information could be categorized. A sufficiently comprehensive compressor would have to include itself, and inevitably become aware. A hypothetical program that could translate any statement into a longer one, and thereby clarify its meaning, repeating as often as necessary. Using a large knowledge base and synonym generator, a universal Improver would be as complex as a full-fledged AI. This might lead to the first useful lie detector test in the future. An interactive Improver could repeatedly combine and analyze all data about a human subject's responses to questioning. Even if the lie detector had only a low accuracy rate (but better than chance), the results could be improved by running it many times in a row. The components of the human brain
-emotional significance of sensory input -fight or flight response -focuses the attention, tunes in to possible threats or opportunities -executive attentional network (reticular formation, thalamus, cingulate cortex) -conflicting input resolution -empathy -disgust -addictive cravings -deep brain structure responsible for learning and memory -falling in love, charity, detection of cheaters -emotional tuning -bonding -logical reality -concepts and plans -emotions and memory -detecting contextual clues -subconscious driver of emotions -feelings of depression and emptiness -the reticula activating system is the brainstem area responsible for wakefulness and focusing attention -the reticular formation is a pivotal area -pleasure center and rewards -one of 3 prefrontal cortex areas -compassion -can put the brakes on aggressive impulses -perceiving the actions and intentions of others -altruism -lateral, orbitofrontal, orbitomedial refrontal; connected to reticular activating system -social control -determining the consequences of actions -routing of sensory input -to limbic system for emotional processing -to prefrontal cortex for rational processing -the link between the subconscious and conscious thoughts -emotional cognition -risk and fear response -adds meaning to perceptions -determines the edges of the body in space Be the first to read Infinite Thunder by Jack Arcalon. The book that took a quarter century to plan and write. With more original scientific, sociological, and technical ideas than any science fiction novel ever published. Original secret source of the Anonymous meme. |