Saturday, May 28, 2016

Redefining Intelligence

Intelligence, natural or artificial, has been a difficult concept to define and understand. Methods of measuring intelligence seem to favor the speed and efficiency in pattern finding. "IQ tests," certainly measure the ability to find patterns and artificial intelligence aficionados have spent three decades teaching computers to get better at the same. Standardized tests, following the same template, appear to measure the same attribute but couch the results in "aptitude," - perhaps to make it sound more plausible. And, across all dimensions of education and testing, this notion of intelligence and hence "aptitude," appears prevalent.

However, can the speed of pattern finding be used as the only metric for intelligence? Certainly in prototypical systems and societies, efficiency in finding food (energy) and fast replication are dominant. Pattern finding is likely the most important skill in this context. If so, then, one could argue that the status-quo definition of intelligence is a measurement of a characteristic that is most useful to maximize a simple objective function, governed largely by food and replicability. At the very least, a thought experiment may be in order to imagine intelligence in higher order societies.

If intelligence is redefined as the differential of the speed in pattern finding - an acceleration in pattern finding - then it can incorporate higher order learning. In societies where such a metric is dominant, the speed of finding patterns from historical data, albeit important, may not qualify as intelligence. One could easily see systems that have very slow speed of pattern finding at inception if energy is focused more at the differential, allowing such systems to exponentially gain knowledge at later stages. Sluggish and dumb, such participants would certainly be eradicated quickly in prototypical societies, before they can demonstrate the accelerating phase of knowledge creation.

Intelligence - ill defined and measured, may need to be rethought, if humans were to advance to a level 1 society. It seems unlikely.

Monday, May 23, 2016

Salt water bubbles

Economists, closer to salt water, appear to be prone to thoughts of inefficiency and bubbles in the financial markets, something that can be cured by a single trip to the windy city. A recent study from Columbia University (1) asserts that they could find over 13,000 bubbles in the stock market between 2000 and 2013. Using supercomputers, no less, and "big data," they appear to have "conclusively shown" that stock prices take wild and persistent excursions from their "fair values." Unfortunately, these academics, who profess to be "data scientists," are yet to encounter the phenomenon of "random walk," further evidence that "data scientists" should stay away from financial markets. After all, the “physicists” who descended into Wall Street have had a checkered history of “abnormal returns” wrapped in consistent negative alpha.

The remark from a graduate student from Harvard - "I expected to see lots of bubbles in 2009, after the crash, but there were a lot before and a lot after," is symptomatic of the problem faced by “data scientists,” seeking problems to solve in super-domains they have no clue about, where participants, who determine outcomes are equipped with pattern finding technology. They may have better luck in real markets, for prices in financial markets are determined by a large number of participants, each with her own inefficient algorithms. The most troubling aspect of the study is that the authors of the study believe that “a bubble happens when the price of an asset, be it gold, housing or stocks, is more than what a rational person would be willing to pay based on its expected future cash flows.” In a world, immersed in intellectual property, where future cash flows cannot be forecasted precisely, the value of an asset cannot be determined by such simple constructs that have been rendered invalid for decades.

The lure of financial markets have been problematic for “data scientists” and “physicists.” However, a cure is readily available in academic literature emanating from the sixties.


Monday, May 16, 2016

Small step toward bigger hype

Recent research from the University of Liverpool (1) suggests a method by which computers could learn languages by semantic representation and similarity look-ups. Although this may be in the right direction, it is important to remember that most of the work in teaching computers language or even fancy tricks, is not in the realm of "artificial intelligence," but rather they belong to the age old and somewhat archaic notion of expert systems. Computer giants, while solving grand problems such as Chess, Jeopardy, Go and self driving cars, seem to have forgotten that rules based expert systems have been around from the inception of computers, much before some of these companies were founded. The fact that faster hardware can churn larger set of rules quicker is not advancing intelligence but it is certainly helping efficient computing.

Engineering schools appear to still teach ideas that are already obsolete. Programming languages have been frozen in time, with prescriptive syntax and rigid control flow. Today's high level languages are certainly practical and immensely capable of producing inferior applications. Even those who could have "swiftly," assembled knowledge from previous attempts seem to have concocted together a compiler that borrows from the worst that have gone before it. As they proclaim "3 billion devices already run it," every hour an update is pushed or conduct conferences around the globe dotting and netting, the behemoths don't seem to understand that their technologies have inherent limitations.

Computer scientists, locked behind ivy walls, are given skills that the world does not need anymore.


Thursday, May 12, 2016

Nutritional genetics

Research from Indiana University (1) speculates that physical traits could be substantially impacted by food. The adage that "you are what you eat," appears to work at a deeper genetic level. In low complexity biological systems, such as ants and bees, variation in food at the larvae stage seems to explain specialization at the genetic level. If true, this has implications beyond what has been observed.

Food, a complex external chemical, has to be metabolized, utilized and purged by biological systems routinely. Although it is clear that available energy content and processing efficiency will depend on the variation and complexity in inputs, the idea that food could cause genetic specialization is fascinating. More importantly, this may lead to better design of food to favorably impact physical and mental conditions, the latter possibly holding higher promise for humans.

Ancient cultures and medicines have routinely relied on food as the primary way to remedy tactical issues. The Indiana research may provide a path to propel this idea into more systematic and planned impacts.


Thursday, May 5, 2016

No safety net

Recent research from Johns Hopkins (1) suggests there are over a quarter of a million deaths in the US per year due to medical errors. It is a sobering observation that future generations will look back on with anguish and perhaps, incredibility.  At the height of technology, we are slipping, not because of lack of know-how, but rather, lack of application. One preventable death is too much and the fact that medical errors are the third leading cause of death in the US, is immensely troubling.

Unfortunately, technology does not solve problems. Bigger data and faster computers are likely irrelevant if they cannot fundamentally influence decision processes and allow information flow to enhance decision quality. It is not about precision - there is no such thing - but a systematic use of all available information at the point of decision. Further, the human brain, with its inherent limitations, is unable to minimize downside risk in a regime of high utilization and volatility. A loss of life, a traumatic and life changing event for any healthcare provider, looms high but the environment simply does not allow anything more than what is tactically possible. The lack of a safety net below cascading, complex and error-prone processes suggest the need for a sudden and impactful change that most technology companies are unable to help with.

It is high time that healthcare embraced practical applications of available technologies to improve patient health and welfare.