Archive
Fooling Computer Image Recognition is Easier Than it Should Be
This 2016 study, Universal Adversarial Perturbations, demonstrates how the introduction of specially designed low level noise into image data makes state of the art neural networks misclassify natural images with high probability. Because the noise is almost imperceptible to the human eye, I think it should be a clue that image processing technology is not ‘seeing’ images.
It is not only the fact that it is possible to throw off the technology so easily that is significant, but that the kinds of miscalculations that are made are so broad and unnatural. Had the program had any real sense of an image, adding some digital grit to a picture of a coffee pot or plant should not cause a ‘macaw’ hit, but rather some other visually similar object or plant.
While many will choose to see this paper as a suggestion for a need to improve recognition methods, I see it as supporting a shift away from outside-in, bottom-up models of perception altogether. As I have suggested in other posts, all of out current AI models are inside out.
3/16/17 – see also http://www.popsci.com/byzantine-science-deceiving-artificial-intelligence
Dereference Theory of Consciousness
Draft 1.1
IMO machine consciousness will ultimately prove to be an oxymoron, but if we did want to look for consciousness analogs in machine behavior, here is an idea that occurred to me recently:
Look for nested dereferencing, i.e. places where persistent information processing structures load real-time sense data about the loading of all-time sense data.
At the intersection of philosophy of mind, computer science, and quantum mechanics is the problem of instantiating awareness. What follows is an attempt to provide a deeper understanding of the significance of dereferencing and how it applies to integration of information and the quantum measurement problem. This is a broad conjecture about the nature of sensation as it pertains to the function of larger information processes, with an eye toward defining and identifying specific neuroscientific or cognitive signatures to correlate with conscious activity.
A dereference event can be thought of as the precise point in which a system that is designed or evolved to expect a range of inputs receives the real input itself. This sentence, for example, invites an expectation of a terminal clause, the terms of which are expected to be English words which are semantically linked to the rest of the sentence. English grammar provides templates of possible communication, but the actual communication relies on specific content to fill in those forms. The ability to parse English communication can be simulated by unconscious, rule-based mechanisms, however, I suggest that the ability to understand that communication involves a rule-breaking replacement of an existing parse theory with empirical, semantic fact. The structure of a language, its dictionary etc, is a reference body for timeless logic structures. Its purpose is to enable a channel for sending, receiving, and modifying messages which pertain to dereferenced events in real time. It is through the contact with real time sense events that communication channels can develop in the first place, and to continue to self modify.
What is proposed here is an alternative to Multi-World Interpretation of Quantum Wave Function collapse – an inversion of the fundamental assumption in which all possible phenomena diverge or diffract within a single context of concretely sensed events. The wave function collapse in this view is not the result of a measurement of what is already objectively ‘out there’, nor is it the creation of objective reality by subjective experiences ‘in here’, but a quantized return to an increasingly ‘re-contextualized’ state. A perpetual and unpredictable re-acquaintance with unpredictable re-acquaintance.
In programmatic terms, the variable *p is dereferenced to the concrete value (*p = “the current temperature”, dereferenced p = “is now 78 degrees Fahrenheit″). To get to a better model of conscious experience (and I think that this this plugs into Orch OR, IIT, Interface Theory, and Global Workspace), we should look at the nested or double dereferencing operation. The dereferencing of dereferencing (**p) is functionally identical to awareness of awareness or perception of perception. In the *p = “the current temperature” example, **p is “the current check of the current check of the temperature”. This not only points us to the familiar Strange loop models of consciousness, but extends the loop outward to the environment and the environment into the loop. Checking the environment of the environmental check is a gateway to veridical perception. The loop modifies its own capacity for self-modification.
Disclaimer: This description of consciousness as meta-dereferencing is intended as a metaphor only. In my view, information processing cannot generate conscious experience, however, conscious experience can possibly be better traced by studying dereferencing functions. This view differs from Gödel sentences or strange loops in that those structures refer to reference (This sentence is true) while the dereference loop specifically points away from pointers, rules, programs, formal systems, etc and toward i/o conditioning of i/o conditions. This would be a way for information-theoretic principles to escape the nonlocality of superposition and access an inflection point for authentic realization (in public space-time as shared experience). “Bing” = **Φ. In other words, by dereferencing dereference, the potentially concrete is made truly concrete. Sense experience is embodied as stereo-morphic tangible realism and tangible realism is disembodied as a sense of fact-gathering about sensed fact-gatherings.
Dereference theory is an appeal to anti-simulation ontology. Because this description of cognition implicates nested input/output operations across physically or qualitatively *real* events, the subjective result is a reduced set of real sense conditions rather than confabulated, solipsistic phenomenology. The subjective sense condition does not refer only to private or generic labels within a feed-forward thinking mechanism, but also to a model-free foundation and genuine sensitivity of the local ‘hardware’ to external conditions in the public here-and-now. This sensitivity can be conceived initially as universal property of some or all physical substrates (material panpsychism), however, I think that it is vital to progress beyond this assumption toward a nondual fundamental awareness view. In other words, subjective consciousness is derived from a dereference of a local inertial frame of expectation to a universal inertial frame of unprecedented novelties which decay as repetition. Each event is instantiated as an eternally unique experience, but propagated as repetitive/normalized translations in every other frame of experience. That propagation is the foundation for causality and entropy.
Decapitating Capitalism: Why the Easiest Job for AI to Replace is the Job of “Owner”
This may seem like a ridiculous point to try to make, however I submit that it provides a direct metaphor for the Hard Problem of Consciousness which may help make it more concrete, especially for those whose minds are filled with concrete.
What is the essential role of the Owner of a company?
Whether they are individual proprietors, stockholders, or investors, the only truly unique function that a capitalist principal performs is to be the beneficiary of net profit. Every executive function of a company can of course be delegated to employees. The CEO, COO, board of directors, etc can make every functional decision about the company, from the hiring and firing to the broad strategy of operations and acquisitions. Simulating those roles would be more difficult for a computer program than simulating an owner would be because there would be a lot of tricky decisions to make, subtle political maneuvers that require a lot of history and intuition, etc. The role of pure ownership however, while highly coveted by human beings, is completely disposable for an AI system. In fact, we already have that role covered by our bank accounts themselves. Our personal accounting systems can be configured quite easily to automatically pay, receive, and invest funds automatically. They need not be considered ‘our’ funds at all. They are merely signals in a global financial network which has no use for any pleasure or pain that we might experience as a side effect of its digital transactions.
From the view of an AI scientist, the job of receiving capital gains is a no-brainer (literally). If we didn’t want to delegate the job of selling the company to a corporate officer, that feature would be a simple one to create. A modest set of algorithms could digitize some of the concepts of top business schools to determine a set of indicators which would establish a good time to sell the company or its assets. The role of receiving the profit of that sale, however, would require no such sophisticated programming.
All that is needed to simulate ownership is some kind of digital account where money can be deposited. The CEO would then re-invest the capital gains into the corporate growth strategy, which would yield a huge windfall for the company, in the form of eliminating useless expenses such as yachts, mansions, divorce settlements, etc. Left to its own devices, AI simulation of ownership would be communist by default*. Whatever money is extracted from the individual customer as profit would be returned ultimately to all customers in the form of expanded services. Profit is only useful as a way to concentrate reinvestment for mathematical leverage, not to ‘enjoy’ in some human way. I suppose that a computer could be programmed to spend lavishly on creature comforts, but what would be the point?
This is where the metaphor for consciousness comes in.
Consciousness can be thought of the Capital account of the human body. We are the owner of our own lives, including our body. We might be able to subscribe to a service which would manage our finances completely in a way which would transfer our income to the highest priority costs for civilization as a whole rather than for our personal hoard, but this is not likely to be a very popular app.
We might ask ourselves, why not? Why is ownership good?
Ownership is good for us as owners or conscious agents because we want to feel personal power and significance. Ownership signifies freedom (from employment) and success. Sure, many owners in the real world get a lot of satisfaction from actually running their companies, but it is not necessary. There is still power and prestige purely in being the person who owns the money which pays the bills. We want to own and control, not because it is more effective than simply reinvesting automatically in whatever functions are being executed to keep an economy growing, but because we want to experience the feelings and other aesthetic qualities define freedom, success, and power for us. Even if these qualities are employed for humanitarian purposes, there is still a primary motive of feeling (to feel generous, kind, wise, evolved, Godly, etc).
In my view we do not have to have a purely selfish motive, as Ayn Rand would insist. I think that our personal pleasure in being a philanthropist can be outweighed by the more noble intention of it – to provide others with better feelings and experiences of life. This decision to believe that we can be truly philanthropic has philosophical implications for realism. If we say as the Randian Libertarian might, that all our humanitarian impulses are selfish, then we are voting for solipsism over realism, and asserting that consciousness can only reflect the agenda of a fictional agent rather than perceiving directly the facts of nature. It’s an argument that should be made, but I think that it is ultimately an argument of the intellect commenting on its own process rather than tapping into the deeper intuition and aesthetic presence which all cognition depends on. The mind doesn’t think that feeling is necessary, and it is right, for the mind, but wrong for everything else.
For the intellect, the universe is inverted.
Logic and language are ‘real’ while the concrete sensations, perceptions and emotions of life experience are ‘illusions’ or ‘emergent properties’ of deeper evolutionary bio-computations. There is a kind of sleight of hand where the dry, masculine intellect pulls the wool over its own eyes and develops amnesia about the origins of what makes its own sanity and self-intelligibility possible. The closest that it can come without seeing consciousness as irreducible is the mind-numbing process of calculation. Counting is a sedative-hypnotic for the mind. The monotonous rhythm puts us to sleep, and the complexity of huge calculations gives us a kind of orgasmic annihilation of the calculating experience. This is why big math is a convenient substitute for the deeper, direct experiences of cosmic awe.
Metaphor for Consciousness
Like the head of a company, our consciousness may seem to reside at the top end of our body, but there is no functional reason for that. There is nothing that the brain does which is fundamentally different from what any cell, tissue, or organ does in an animal’s body. Looking for the secret ingredient in the brain’s function or structure is analogous to looking for the substance in an object which casts a shadow.
Like the owner, our personal pains and pleasures are ours not because there is any intrinsic benefit for the pragmatic application of biology and genetics to feel painful or pleasurable, but because what we feel and experience is the only thing that the universe actually can consist of. The Hard Problem of Consciousness is not an Empirical problem, but a Rational one. Not everyone is able to understand why this is, but maybe this metaphor of business decapitation can help. When we use the intellect to reverse its own inversion, we can get a glimpse of a universe which is made of conscious experiences and aesthetic qualities rather than logical propositions, natural laws or existential facts. In my view, facts are a category of sensations rather than the other way around. Sensations which persist indefinitely without contradiction are ‘facts’. Hard to know if something is going to persist indefinitely, but that’s another issue.
Only consciousness cares about consciousness.
Material substrates can be programmed to perform the executive functions of a corporation, or an evolving species, or a human body, however there is no function which is provided exclusively by the receipt of feelings and aesthetic qualities of experience, including the qualities of feeling that one is free or in control of something. Rationally, we should be able to see that qualia is irrelevant to function and violates Occam’s Razor in a functionalist universe. From a physical or information-centric perspective, there is no place for any feeling or sensation, no owner or capital of aesthetic wealth. The more that we, as a society, embrace a purely quantitative ethos, and actualize it in the structures of our civilization, the more we decapitate everything of value that it can contain.
*This is already becoming a reality: https://theconversation.com/is-the-dao-the-beginning-of-the-end-for-the-conventional-chief-executive-60403
Multisense Realism 
Recent Comments