Thursday, November 28, 2013

KNOWLEDGE DRIVEN TECHNOLOGY & MANAGEMENT



 KNOWLEDGE DRIVEN TECHNOLOGY & MANAGEMENT

THE PROBLEM:

Technology is the application of knowledge for practical purposes. Hence it should be guided by theory. But the technological advancements in various sectors has led to data-driven discoveries in the belief that if enough data is gathered, one can achieve a “God’s eye view”. Data is not synonymous with knowledge. By combining lots of data, we generate something big and different, but unless we have knowledge about the mixing procedure to generate the desired effect, it may create the Frankenstein’s monster - a tale of unintended consequences. Already physics is struggling with misguided concepts like extra-dimensions that are yet to be discovered even after a century. Weirdness of the concepts of  superposition and entanglement are increasingly being questioned with macro examples. The LHC experiment has finally ruled out super-symmetry. Demand for downgrading the status of the Heisenberg’s uncertainty postulate is gaining momentum. Yet, fantasies like dark energy or vacuum energy, where theory and observation differ by a factor of 1057 to 10120, get Nobel Prize! Theoreticians are vanishing. Technologists are being called scientists. Increase of trial and error based technology that lack the benefit of foresight, are leading to more nonlinearly non-green technology necessitating Minamata Mercury Convention (for reducing Mercury poisoning) type conferences to prescribe do’s and don’ts for some industries. Technology has become the biggest polluter.

With increasing broadband access, wireless connectivity and content, dependence on gadgets like smart phones, tablets, etc, is growing. Apart from its impact on vegetation (browning), birds, and the ecosystem in general, the impact this human–machine bond will have on our lives is yet to be fully assessed. The current trend is to create a product out of an idea (not necessity), for which technology is invented later. The necessary recommendation algorithms are compartmentalized in different branches of science. For example, to find the accelerating expansion of the universe and define the nature of dark energy, researchers used baryon acoustic oscillations as the yardstick. It was created from sound waves that rippled through the universe when it was young and hot and became imprinted in the distribution of galaxies as it cooled. In sync with the idea, Google+ and Apple’s Siri came up with learning algorithms that respond to one’s voice. Apple’s new iPhone fingerprint sensor is directed at the machine knowing our bodies. Such devices start by recognizing one’s thumb or voice; then other’s voices, the way they move, etc. If such devices put such information together with information about one’s location and his engagement calendar, it will be an integral part of our life. Social media is changing the kinship diagram through emotionless physical relationships. Network Administrators and algorithms regulate ‘date’ vetting. Human beings are increasingly submitting themselves to machines and becoming mechanized.

As the available resources get depleted and demand for more intelligent solutions and services using nano-technology increases, there is pressure for more re-generative and ‘intelligent’ – GREEN and SMART– technologies emphasizing the need for knowledge collaboration in engineering. Green technology encompasses a continuously evolving group of methods and materials, from techniques for generating non-exhausting energy sources like solar or wind or tidal power to non-toxic clean products (based on their production process or supply chain) that are environmental-friendly and biodegradable. It involves energy efficiency, recycling, safety and health concerns, renewable resources, etc. Yet, it has to fight the ever increasing greed for easy money. For example; as the world gold prices surges, small-scale ‘artisanal’ gold mining has become the world’s leading source of mercury pollution. Miners use mercury to separate flecks of gold from rocks, sediment and slurry and then dump or burn the excess. It exposes ground water and air to mercury poisoning. But to motivate the miners to adopt green alternatives is nearly impossible. Recycling without the knowledge of its adverse side-effects is causing more pollution world wide. But the greed for higher Return on Investment is eulogized as prosperity and advancement.

“SMART” stands for “Self-Monitoring, Analysis and Reporting Technology”. It gets input from somewhere, applies some ‘intelligence’ or ‘brainpower’ to it and the result is innovative. For example, regular glasses used in spectacles are shaped in such a way as to bend light for correct vision - to make the world appear sharper and clearer. Photo-chromatic lenses contain molecules that react to certain kinds of light and change tint in sunshine. Though it seems intelligent, these are just physical reactions. By adding a camera and a computer to a pair of glasses, many innovations can be made. A video camera at the corners of the spectacles that feed into a tiny pocket computer that light up parts of an LED array in the lenses can enable the wearer to see objects in greater detail. It could include optical character recognition for reading newspaper headlines. The glasses use cameras and some software to interpret the data and put zoomed in images on a screen in front of the wearer’s eyes. This is only exemplary.

Artificial Intelligence (AI) is the current buzz word. AI is of two types called narrow (ANI) and general (AGI) artificial intelligence. ANI is the intelligent function at one narrow task like playing a chess game or searching the web and is increasingly ubiquitous in our world. ANI may outsmart humans only in the area in which it is specialized - hence not a big transformative concern. But AGI, which is potentially intelligent across a broad range of domains, is cause for concern. We mix the different sensory inputs by our intelligence and apply our freewill to determine net response, but an AGI would probably think or mix differently in unexpected ways. If we command a super-intelligent robot to make us happy, it might cram electrodes into the pleasure centers of our brains. If we command it to win at chess, it may calculate all possible moves endlessly. This absurd logic holds because AI lacks our instincts and the notions of absurdity and justification of mixing inputs. It does what we program it to do, but without freewill. Once the embryo starts breathing, it breathes perpetually till death, but the child also has limited free-will and uses his instincts. After being switched on, computers obey commands, but have no free-will or instincts. Since these cannot be preprogrammed, AI can never be conscious.

Knowledge is not data, but the ‘awareness’ of exposure/result of measurement associated with any object, energy or interaction stored in memory as an invariant concept that can be retrieved even in the absence of fresh inputs or impulses. It describes through a language the defining characteristics of some previously known thing – physical properties and chemical interactions - by giving it a name that remain the same as a concept at all times – thus immune to spatiotemporal variations - till it is modified by fresh inputs. The variations of the object, energy or interaction under different specific circumstances and the predetermined result thereof form part of knowledge. In a mathematical format, it depicts the right hand side of each equation or inequality representing determinism. Once the parameters represented by the left hand side are chosen and the special conditions represented by the equality sign are met, the right hand side becomes deterministic. In ancient times, it was technically covered under the term Aanwikshiki, which literally means describable facts about the invariant nature of everything.

Engineering and Management which deal with the efficient use of objects or persons; are related to left hand side of an equation – free-will; which presupposes knowledge of the deterministic behavior of objects or humans that can be chosen or effectively directed to create something or function in a desired manner in a maximally economic and regenerative way. This was called Trayi – literally the three aspects of behavior of mass, energy and radiation in their three states of solid, fluid and plasma in all combinations – physical and chemical properties (protestation, loyalty and expectation for humans). The responsive mechanism was called Danda Neeti – principles of inducement through reward and punishment (essentially material addition or reduction). The regenerative mechanism was called Vaartaa – problem solving. These four basic tenets, equally valid for both technology and management, are also immutable - invariant in time, space and culture leading to deterministic consequences. Lack of knowledge of the deterministic behavior to guide choice of the freewill components has led engineering and management astray. The fast changing technology and management principles point to their inherent deficiencies that need immediate remedy. Knowledge guidance is the only way out.

There is a pressing need for knowledge to take the lead for greener technology keeping in view sustainability, cradle-to-cradle design, source reduction, viability, innovation, etc. Hence it is necessary that pure science guide technology in the right direction in ALL sectors. Till date all efforts in this regard have been sector specific such as energy, chemicals, medical, real estate, hardware, etc. As a result, green and smart technology has been reduced to transferring problems in a discrete manner – they solve problem in one area (for example by recycling something) ignoring the effect of the new process or its by-products on other areas. It is high time to discuss a global strategy to meet the new challenges.

THE PARADIGM SHIFT:

Earlier, some individual scientists with their over-towering genius developed a postulate and took the lead in Universities or Research Institutions to develop suitable experimental setups to test those. These days, individual scientists have to network and collaborate across State and National boundaries to take advantage of State and International funding. They generate incredibly massive data without any postulate. Communication technology has made the efforts of individual researchers coalesce into a seamless whole merging identities of who contributed what. The 2013 Nobel Prize in physics was the result of many ideas that were floated around in early 1960’s by at least six scientists. During that time lots of new particles were being discovered and it was a fair bet that some particle would be discovered in the vacant 124-126 GeV range. Hence it was proposed as a gamble. The model tested at the LHC was not that of Higgs and Englert, who got the prize, but one for which Weinberg and Salam had already won a Nobel! The general mechanism was first postulated by Philip Anderson a couple years before Higgs and Englert. Already there are protests against the decision.

As individual efforts became obscured and team efforts took-over, more and more data are accumulated making their storage and analysis a big problem. When subatomic particles are smashed together at the LHC, they create showers of both known and unknown new particles whose signatures are recorded by four detectors. The LHC captures 5 trillion bits of data (more information than all of the world’s libraries combined) every second. After the application of filtering algorithms, more than 99 percent of those data are discarded, but still the four detectors produce 25 petabytes (25×1015 bytes) of data per year that must be stored and analyzed. These are processed on a vast computing grid of 160 data centers around the world, a distributed network that is capable of transferring as much as 10 gigabytes per second at peak performance. Are these data really necessary? Can we be sure that useful data are not being discarded while filtering, particularly when do not know what we are searching for or are searching selectively? Is there no other way to formulate theory?  Is the outcome cost-effective?

The unstructured streams of digital potpourri are no longer stored in a single computer - it is distributed across multiple computers in large data centers or even in the “cloud”. It demands developing rigorous scientific methodologies and different data-processing requirements - not only flexible databases, massive computing power and sophisticated algorithms, but also a holistic (not reductionist) approach to get any meaningful information. One possible solution to this dilemma is to embrace a new paradigm. In addition to distributed storage, why not analyze the data in a distributed manner as well! Each unit (or node) in a network of computers perform a small piece of the computation! Each partial solution is then integrated to find the full result. For example, at LHC, one complete copy of the raw data (after filtering) is stored at the CERN in Switzerland. A second copy is divided into batches that are then distributed to data centers around the world. Each center analyzes a chunk of data and transmits the results to regional computers before moving on to the next batch. But this lacks the holistic approach. The reports of the six blind men about the body parts of the elephant are individually correct. But unless someone has seen an elephant, he cannot make any sense out of it.

THE BIG-DATA CHALLENGE:

The demand for ever-faster processors, while important, is not the primary focus anymore. Processing speed has become completely irrelevant now. The challenge is not how to solve problems with a single, ultra-fast processor, but how to solve them with a large number of slower processors. Yet, many problems in big-data cannot be adequately addressed by adding more parallel processing. These problems are more sequential, where each step depends on the outcome of the preceding step. Sometimes the work can be split-up among a bunch of processors, but that is not easy always. Time taken to complete one task is not always inversely proportional to the number of persons. Often the software is not written to take full advantage of the extra processors. Failure of Just-in-time and super-efficiency in management has led to world-wide economic crisis. We may be approaching a similar crisis in the scientific and technological field. The Y2K problem was a precursor to what could happen.

Addressing the storage-capacity challenges of big-data involves building more memory and managing fast movement of data. Identifying correlated dimensions is exponentially more difficult than looking for a needle in a haystack. When one does not know the correlations one is looking for, one must compare each of the ‘n’ pieces of data with every other piece, which takes n-squared operations. The amount of data is roughly doubling every year (Moore’s Law). If in our algorithm, for each doubling of data, we have to do two-squared times computing, then in the following year, we have to do 16 times (four squared) as much computing. By next year, our computers will only be twice as fast and in two years our computers will only be four times as fast. Thus, we are exponentially falling behind in our ability to store and analyze the collected data. There are non-technical problems also. The analytical tools of the future require not only the right mix of physics, chemistry, biology, mathematics, statistics, computer science, etc., but also the team leader to take a holistic approach – free of reductionism. In the big-data scenario, mathematicians and statisticians should normally become intellectual leaders. But mathematics is more focused on abstract work and do not encourage people to develop leadership skills – it tends to rank people linearly to determine an individual pecking order introducing bias. Engineers are used to working on teams focused on solving problems, but they cannot visualize new theories.

Although smaller studies via distributed processing provide depth and detail at a local level, they are also limited to a specific set of queries and reflect the particular methodology of the investigator, which makes the results more difficult to reproduce or reconcile with broader models. The big impacts on the ecosystem including effects of global warming cannot be studied with short-term, smaller studies. But in the big-data age of distributed computing; the most important decision to be taken is: how to conduct distributed science across a network of researchers - not merely “interdisciplinary research”, but a state of “trans-disciplinary research” - free from the reductionist approach? Machines are not going to organize data-science research. Researchers have to turn petabytes of data into scientific knowledge. But who is leading data-science right now? There is a leadership crisis! There is a conceptual crisis!

Today’s big data is noisy, unstructured, and dynamic rather than static. It may also be corrupted or incomplete. Many important data are not shared till its theoretical, economical or intellectual property right aspects are fully exploited. Sometimes data is fudged. Data should comprise of vectors – a string of numbers and coordinates. But now researchers need new mathematical tools, such as text recognition, or data compression by selecting key words and their synonyms, etc., in order to glean useful information from and intelligently curate the data-sets. For this either we need a more sophisticated way to translate it into vectors, or we need to come up with a more generalized way of analyzing it. Several promising mathematical tools are being developed to handle this new world of big, multimodal data.

THE NEW APPROACH:

One solution suggested is based on consensus algorithms. It’s a mathematical optimization system. Algorithms with past data are useful for creating an effective SPAM filter on a single computer, with all the data in one place. But when the problem becomes too large for a single computer, a consensus optimization approach works better. In this process, the data-set is chopped into bits and distributed across several “agents” each of which analyze their bit and produce a model based on the data they have processed - something similar in concept to the Amazon’s Mechanical Turk crowd-sourcing methodology. The program learns from the feedback, aggregating the individual responses into its working model to make better predictions in the future. In this system, the process is iterative, creating a feedback loop. Although each agent’s model can be different, all the models must agree in the end - hence “consensus algorithms”. The initial consensus is shared with all agents, which update their models and reach a second consensus, and so on. The process repeats until all the agents agree.

Another prospect is quantum computing, which is fundamentally different from parallel processing. A classical computer stores information as bits that can be either 0s or 1s. A quantum computer could exploit a weird property called superposition of states. If we flip a regular coin, it will land on heads or tails. There is zero probability that it will be both heads and tails. But if it is a quantum coin, it is said to exist in an indeterminate state of both heads and tails until we look to see the outcome. Thereafter, it collapses - assumes a fixed value. This is a wrong description of reality. The result of measurement is always related to a time t, and is frozen for use at later times t1, t2, etc, when the object has evolved further and the result of measurement does not depict its true state. Thus, we can only know the value that existed at the moment of observation or measurement. Scientists impose their ignorance of the true state of the system at any moment on the object or the system and describe the combined unknown states together as superposition of all possible states. It is physically unachievable.

The quantum computers, if built, will be best suited to simulate quantum mechanical systems or to factor large numbers to break codes in classical cryptography. Quantum computing might be able to assist big-data by searching very large, unsorted data-sets in a fraction of the time for parallel processors. However, to really make it work, we would need a quantum memory that can be accessed while in a quantum superposition, but the very act of accessing the memory would collapse or destroy the superposition. Some claim to have developed a conceptual prototype of quantum RAM (Q-RAM), along with an accompanying program called Q-App (pronounced “quapp”) targeted to machine learning. The system could find patterns within data without actually looking at any individual records, thereby preserving the quantum superposition (questionable idea). One is supposed to access the common features of billions of items in his database at the same time, without individually accessing them. With the cost of sequencing human genomes (where a single genome is equivalent to 6 billion bits) dropping, and commercial genotyping services rising, there is a great push to create such a database. But knowing about malaria without knowing who is having it, is useless for treatment purposes.

Another approach is integrating across very different data sets. No matter how much we speed up the computers or computers together, the real issues are at the data level. For example, a raw data-set could include thousands of different tables scattered around the Web, each one listing similar data, but each using different terminology and column headers, known as “schema”. The problem can be overcome with a header to describe the state. We must understand the relationship between the schemas before the data in all those tables can be integrated. That, in turn, requires breakthroughs in techniques to analyze the semantics of natural language. What if our algorithm needs to understand only enough of the surrounding text to determine whether, for example, a table includes specific data so that it can then integrate the table with other, similar tables into one common data set? It is one of the toughest problems in AI. But Panini has already done it with the Pratyaahaara style of the 14 Maheshwari Sootras.

One widely used approach is the topological data analysis (TDA), which is an outgrowth of machine learning - a way of getting structured data out of unstructured data so that machine-learning algorithms can act directly on it. It is a mathematical version of Occam’s razor: While there may be millions of possible reconstructions for a fuzzy, ill-defined image, the sparsest (simplest) version is probably the best fit. Compressed sensing was born out of this serendipitous discovery. With compressed sensing, one can determine which bits are significant without first having to collect and store them all. This allows us to acquire medical images faster, make better radar systems, or even take pictures with single pixel cameras. The idea was there since Euler, who puzzled over a conundrum: is it possible to walk across seven bridges connecting four geographical regions, crossing each bridge just once, and yet end up at one’s original starting point?  The relevant issue was the number of bridges and how they were connected. Euler reduced the four land regions to nodes connected by the bridges represented by lines. To cross all the bridges only once, each land region would need an even number of bridges. Since that was not the case, such a journey was impossible. A similar story is told in B-Schools. If 32 teams play a knock-out tournament, how many games will be played totally? One reasoned that in every game, one team will be defeated. Only one team will remain undefeated till the end. Thus, the total number of games is 31. This is the essence of compressed sensing. Using compressed sensing algorithms, it is possible to sample only 100 out of 1000 pixels in an image, and still be able to reconstruct it in full resolution - provided the key elements of sparsity (which usually denotes an image’s complexity or lack thereof) and grouping (or holistic measurements) are present.

Taking these ideas, mathematicians are representing big data-sets as a network of nodes and edges, creating an intuitive map of data based solely on the similarity of data points. This uses distance as an input that translates into a topological shape or network. The more similar the data points are, the closer they will be to each other on the resulting map. The more different they are, the further apart they will be on the map. This is the essence of TDA. Many of the methods in machine learning are most effective when working with data matrices, like an Excel spreadsheet, but what if our data set does not fit that framework? 

TDA is all about the connections. In a social network, relationships between people can be mapped: with clusters of names as nodes and connections as edges illustrating how they are connected. There will be clusters relating to family, friends, colleagues, etc. But it is not always discernible. From friendship to love is not a linear relationship. It is possible to extend the TDA approach to other kinds of data-sets, such as genomic sequences. One can lay the sequences out next to each other and count the number of places where they differ. That number becomes a measure of how similar or dissimilar they are and one can encode that as a distance function. This is supposed to reveal the underlying shape of the data. A shape is a collection of points and distances between those points in a fixed order. But such a map will not accurately represent the defining features. If we represent a circle by a hexagon with six nodes and six edges, it may be recognizable as a circular shape, but we have to sacrifice roundness. A child grows with age, but the rate of growth is not uniform in every part of the body. Some features develop only after certain stage. If the set at lower representations has topological features in it, that is not a sure indication that there are features in the original data also. The visual representation of the flat surface of Earth does not belie its curvature. Topological methods are also a lot like casting a two-dimensional shadow of a three-dimensional object on the wall: they enable us to visualize a large, high-dimensional data set by projecting it down into a lower dimension. The danger is that, as with the illusions created by shadow puppets, one might be seeing patterns and images that are not really there. There is a joke that topologists can not tell the difference between their rear end and a coffee cup because the two are topologically equivalent.

Some researchers emphasize the need to develop a broad spectrum of flexible tools that can deal with many different kinds of data. For example, many users are shifting from traditional highly structured relational databases, broadly known as SQL, which represent data in a conventional tabular format, to a more flexible format dubbed NoSQL. It can be as structured or unstructured as we need it to be, depending on the application. Another method favored by many is the maximal information coefficient (MIC), which is a measure of two-variable dependence designed specifically for rapid exploration of many-dimensional data-sets. It was claimed that MIC possesses a desirable mathematical property called equitability that mutual information lacks. It has been disputed that MIC does not address the issues of equitability, but rather focuses on the statistical power. MIC is said to be less powerful than a recently developed statistic called distance correlation (dCor) and a different statistic, HHG, both of which have their own problems and are not satisfactory either.

            In all these, we are missing the woods for the trees. We do not need massive data – we need theories out of the data. Higgs boson is said to validate Standard Model, which does not include gravity – hence incomplete. The graviton, also predicted by SM but described differently in string theory, is yet to be discovered. That the same experiment disproves Super symmetry (SUSY), which united gravity with SM, questions it and points to science beyond SM.

A report published in July, 2013 in the Proceedings of the National Academy of Sciences USA, shows that to make healthy sperm, mice must have genes that enable the sense of taste. Sperm have been shown to host bitter-taste receptors and smell receptors, which most likely sense chemicals released by the egg. But the idea that such proteins might function in sperm development is new. Elsewhere researchers have found taste and smell receptors in the body that help to sense toxins, pick up messages from gut bacteria or foil pathogens. This opens up a whole world of alternative uses of these genes. When we assign functions to genes, it is a very narrow view of biology. Probably for every molecule that we assign a specific function to, it is doing other things in other contexts. If anyone bothered to read the ancient system of Medicine Ayurveda or properly interpret Mundaka Upanishadic dictums “annaat praano”, they will be surprised to rediscover the science that is indicated through the latest data. We are not discussing it here due to space constraint. There are many such examples. Instead of looking outward to data, let us look inward and study the objects and develop the theories (many of which have become obsolete) afresh based on currently available data. The mind-less data-chase must stop.

THE WAY AHEAD:

            Theory without technology is lame. Technology without theory is blind. Both need each other. But theory must guide technology and not the opposite. Nature provides everything for our sustenance. We should try to understand Nature and harmonize our actions to natural laws. While going for green technology, we must focus on the product that we use, than on the packaging that we discard. As per a recent study, in London people waste 60% of the food they buy to eat while others go hungry. Necessity and not idea should lead to creation of a product. Minimizing waste is also green. Only products that are really not essential for our living need advertisement. The concept of every business is show business must change. Product liability laws should be strengthened; specifically in FMCG sector. But what is the way out when economic and military considerations drive research? We propose an approach as follows:

·                    The cult of incomprehensibility and reductionism that rules science must end and trans-disciplinary research values inculcated. Theory must get primacy over technology. There should be more seminars to discuss theory with feedback from technology. Most of the data collected at enormous cost are neither necessary nor cost effective. This methodology must change.
·                    The superstitious belief in ‘established theories’ must end and truth should replace fantasy. We have given alternative explanations of ten-dimensions, time dilation, wave-particle duality, superposition, entanglement, dark-energy, dark matter, inflation, etc, before international scientific forums with macro-examples without cumbersome mathematics, while pointing out the deficiencies in many ‘established theories’. Those views have not yet been contradicted.
·                    To overcome economic and military pressure, International Conventions on important areas like the Minamata Mercury Convention should be held regularly for other problem areas under the aegis of UNESCO or similar International bodies.
·                    There is no need to go high-tech in all fields. We should think out of box. Traditional knowledge is a very good source of information (most herbal product companies use it successfully). If we analyze these scientifically without any bias, we can get lot of useful inputs. The chain of “Amma Canteens” in Tamil Nadu, India, is an excellent example of green technology. It supplies fresh food at cheap rates with minimum infrastructure, storage, transportation, pollution, wastages and maximum employment. The focus is on the locally available product and not the package. There can be many more such examples and innovations without big-data.
·                    General educational syllabus must seek to address day-to-day problems of the common man. Higher education should briefly integrate other related branches while focusing on specialization.
·                    Technologist is a honorable term. But stop calling them scientists.

N.B.: Here we have used ancient concepts with modern data.