Google

Winc

YouTube

Spotify

Scientific Sense Podcast

Sunday, March 31, 2019

The hype of AI

A recent article (1) further reinforces what autonomous vehicle industry has been doing. Neural net systems with feedforward and feedback control architectures trained by historical data on specific surfaces and conditions. Remnants of 1960s technologies, ably assisted by zero cost computing, have been percolating across the autonomous landscape. This trajectory is problematic for many reasons.

First, a brain trained on historical data selected by a biased human is a disaster waiting to happen. The situation is no better with hand-coded heuristics as demonstrated by recent aircraft failures. What computer and data scientists have to understand first is that their own brains still remain to be vastly superior to code they write running even on a super-computer. Hence, blind attempts at removing the human from complex decision-making processes are likely to fail.

Second, hype and ignorance have propelled AI to the stratosphere without significant practical use cases. AI is a tool and it is not a panacea. AI still fails when it encounters the unexpected. This is important as it indicates conventional computing and Silicon based architectures, albeit great engineering innovations, have nothing to do with "intelligence." We have not advanced AI much from the 80s, when the "oldies," used to call it expert systems. Granted, simulated voices, believable human faces, and incredible jumping robots are great inventions, but unfortunately, these have nothing to do with AI.

And finally, high human resource intensity in model building often leads to costly failures. For practical AI, two important things need to come together - rapid and flexible prototyping with automation and considering AI to be augmenting the human, not replacing her.

(1) http://robotics.sciencemag.org/content/4/28/eaaw1975

Saturday, March 30, 2019

Alzheimer’s - we fail again!

Recent news (1) that high profile experiments, targeting a solution for the famous disease, Alzheimer's, has failed again is sobering. It clearly indicates that life sciences companies are on the wrong track to improve a condition, most dread. It is also a constant reminder that systemic problems cannot be solved by treating symptoms or tactical observations. The engineering view of medicine has run its course and it has been very successful in fighting opponents that can be clearly identified. But now, a system failing because of overuse, cannot be mended by such crude methodologies.

Immortality is prohibited by contemporary Physics. So, the optimization problem narrows to maximizing utility within an afforded time horizon. Humans have been naturally optimistic, an evolutionary advantage. They have been attempting to extend life rather than optimizing within constrains. Therein lies the paradox for healthcare. As artificial intelligence progresses, it is conceivable that a human could have reasonable estimates of life span and disease incidences, at birth. For the first time in history, we may be in a position to focus on optimization of utility rather than extending a highly uncertain horizon.

It is clear that the human hardware deteriorates in predictable ways. Most of it appears to be related to plumbing, an inability to remove waste at an optimum rate. From the brain to the kidneys, waste removal efficiency seems to decline over time, just as in any physical system. We may need to accept this as an unbreakable law and find ways to slow down the deterioration. In this context, research in the direction of cure for auto-immune diseases may be misplaced. What could be more important is predicting the likelihood of disease early and slowing its progression.

Artificial Intelligence could have a significant impact on human utility and happiness. If one can get over the hype and confusion, it will become clear that AI could provide useful guidance for humans to best utilize their limited time in the universe.

(1) https://www.sciencemag.org/news/2019/03/another-major-drug-candidate-targeting-brain-plaques-alzheimer-s-disease-has-failed

Thursday, March 28, 2019

F-Theory, Occam's razor and ignorance

Physicists are funny people - they like complexity and large numbers. They recently found F-theory (not sure what "F" stands for except for the obvious possible guesses), that demonstrates a "quadrillion ways, string theory could make the universe." That is impressive and could provide rich fodder for PhD theses and "peer reviewed," publications for a century to come. But does it advance knowledge? It is less clear.

Complexity is a problematic concept. As the true geniuses of the yesteryear correctly emphasized, "God does not play dice," and unless she has a quantum computer in the basement, she will not embark on a journey that has "quadrillion," ways to make the universe. Large numbers could be ego boosting behind the ivy walls but it has no practical use. It is probably time for those attempting to formulate the next exotic "theory of everything," to get out of their windowless offices and smell the roses.

There is no problem that has been solved by increasing complexity. Those who advanced thinking always preferred simplicity. Knowledge is clearly inversely correlated with complexity. Even the money men, who typically do not know much, seem to have gravitated to this universal law. So, why are we here at this point in time? One possible answer is that engineering heavy education has churned out engineers who want to measure "reverberations," not much larger than the size of a proton or mathematicians, enamored by "large numbers." Even the business guys seem to have learned bad habits as most want to use, "big data."

Complexity is utility diminishing. Theories that push in that direction is utterly useless.

(1) https://www.scientificamerican.com/article/found-a-quadrillion-ways-for-string-theory-to-make-our-universe/

Tuesday, March 26, 2019

The era of bioelectronics

The most complex electromagnetic and chemical system known, the human body, so far has substantially avoided manipulation by electromagnetic means. This may be changing. Recent news (1) about a transistor design that enables integrated, real-time sensing and simulation of signals from living organisms, could lead to better diagnostics and treatment. Low cost Silicon has impeded innovation and applications in non-conventional substrates. There have been few biocompatible designs for the lack of appropriate materials and incentives.

Chemicals have been easier and in the presence of many low hanging fruits, researchers did not spend much time on alternatives. As they solved simpler problems, auto-immune diseases start to dominate the human architecture. The heart-breaking failure of a recent drug for Alzheimer's (2) is symptomatic of the end of the chemical era. The brain likely responds better to electromagnetic stimuli but contemporary pharmaceutical companies are ill-equipped to pursue this line of thought.

Simple diseases such as Hypertension and Type 2 Diabetes, that command over half of the healthcare costs in the log run, could be positively influenced by better monitoring and treatment mechanisms that are integrated into the body. CHF and other Cardiovascular events could be picked up earlier and intervened optimally by the same mechanisms. As the sun gets hotter and nastier, embedded devices in the skin could shield the body from harmful rays. Organ failures could be arrested, nutrition deficiencies could be remedied, better food and treatment regimens could be suggested and humanity could possibly move to a more advanced health regime.

It is exciting. Integrated bioelectronics with embedded artificial intelligence could be a game changer.

(1) http://advances.sciencemag.org/content/5/2/eaau7378
(2) https://www.cnn.com/2019/03/21/health/alzheimers-drug-trial-failure-aducanumab-bn/index.html


Sunday, March 24, 2019

The fallacy of experimental Physics

Experiments, albeit useful from a practical perspective, have never advanced theory. The Hubble constant is being pushed, pulled and tweaked by engineers and physicists as if that is going to lead somewhere. The "standard model," was dead on arrival. It is just that tenure seeking academics and measurement driven engineers, will not let it go. Now, the "cosmological crisis," (1) apparently is going to propel the theory to the next level.

It is unlikely. Theory emanates from imagination and not from "precise," measurements. Theory is about the unknown and not the incremental. Theory is about scope and not scale. Theory is about emergence and not process. Theory is about mindless excursions and not planned travel. Theory is about finding connections among the disconnected. Theory is about freedom and not programs. Theory is not about travelling to adjacent planets but conceptualizing what may be beyond, Theory is not exploration of the tangible but the unknown. Theory does not require heavy steel, just paper and pencil.

Humanity waits for the arrival of the next genius. For over a century, experimentalists roamed the planet with nothing to show. It showcases why the existence of a singular mind at a space-time coordinate defines the trajectory for knowledge.

We may be stuck. Our more and more "precise," measurements will asymptotically reach complete ignorance.



(1) https://www.scientificamerican.com/article/best-yet-measurements-deepen-cosmological-crisis/

Tuesday, March 19, 2019

AI and the weakest link


The recent debacle in aircraft design is a constant reminder that software engineers and “data scientists,” excited by the possibilities, could create havoc in many different industries. In transportation, as autopilot systems get smarter, they could take over virtually everything a vehicle does, terrestrial or otherwise. What the designers seem to have missed recently is that an aircraft is a conglomeration of data transmitting mechanical sensors and sophisticated software. Traditional engineering education would have informed the designers that the system is only as good as the weakest link, but the younger ones may have skipped those courses. Here, faulty data from an old technology may have confused the brain. There are multiple issues to consider here.

First, the design of systems needs to be holistic. This is easier said than done as a complex vehicle is designed in parts and assembled. Teams who work on these components may have different skill sets and the overall blueprint may not consider the biases in designs created by separate teams. For example, if the brain is designed with little flexibility to discard faulty data, the expectation would be that it is unlikely. However, if the data is emerging from mechanical devices, with no embedded intelligence, it is almost a certainty that faulty data will arrive at some point in the senor’s life. Two recent aircraft failures in Asia and Africa and the one much earlier over the Atlantic seem to have been caused by bad sensors sending bad data to a “sophisticated AI agent,” with little capability to differentiate between good and bad data. So, either the sensors and other mechanical devices in the system need to be smarter so as to recognize their own fallibilities or the central brain has to be able to recognize when it is fed bad stuff. There is a lull in engineering education that has moved in the direction of high specialization, without an overall understanding of systems design and risk. This is going to surface many issues across industries from transportation, manufacturing to healthcare.

Second, the human is still the best-known risk mitigator, with her brain fine-tuned over the millennia to sense and avert danger. In transportation, disengagement has to be a fundamental aspect of design. Although it could be tempting to sit back while an aircraft takes off and lands or to read “Harry Potter,” while behind the wheels of an autonomous terrestrial vehicle, these actions are ill-advised. The human has to expect the machine to ill behave and be at the very least ready to receive complete disengagement at any point in time. Excited engineers may think otherwise, but we are nowhere close to fail-safe AI. Let’s not kid ourselves – writing code is easy but making it work all the time, is not. Educational institutions will do a disservice for the next generation of engineers if they impart the idea that AI is human devoid.

Transportation is just one industry. The problems witnessed, span across every industry today. For example, in healthcare, AI is slowly percolating but the designers have to remember that there are weak links there too. Ironically, in the provider arena, the weak link is the human, who “inputs,” data into aging databases, sometimes called Electronic Medical Records (EMR) systems. Designed by engineers, with no understanding of healthcare, a couple of decades ago, they are receptacles of errors that can bring emerging AI and decision systems to their knees. If one designs AI driven decision systems in these environments, she has to be acutely aware of the uncertainty in inputs caused by humans, who are notorious in making mistakes with computer keyboards (or even voice commands) and database containers designed with old technologies. So, designs here need to systematically consider disengagement when the AI agent is unable to decipher data.

In manufacturing, led by data collection enthusiasts from the 90s, older database technologies, sometimes elegantly called, “Enterprise Resource Planning (ERP), systems, dominate. They have been “warehousing,” data for decades with little understanding of what it means. Now, “Hannah,” and her cousins, seem to have gotten religion but again, there is a problem here. Cutting and dicing data to make pretty pictures for decision-makers, does nothing to improve decision-making or to mitigate risk. The weak link here is the technology, designed and maintained by those who believe businesses are run by the collection, aggregation, and reporting of data. Unfortunately, successful businesses have moved on.

AI is a good thing, but not in the absence of logical thinking and systems design. Intelligence is not about the routine, but the ability to act when encountered the “non-routine.” As the software and hardware giants sell their wares, in the cloud and elsewhere, they have to understand the perils of bad and rushed technology. It is great to fool a restaurant by simulated voice, it is great to demonstrate that “machine learning,” on Twitter will create a racist and it is great to tag and collate music fast, but none of these activities is going to propel humanity to the next level. Being good in search, operating system design or good hardware, do not automatically make these companies, “experts” in an area that is erroneously called, Artificial Intelligence. There is nothing artificial about intelligence. Machines have the potential to be a lot more “intelligent,” than humans. If anybody has any doubt, just take a look at the nation’s capital and imagine a scenario of replacing the policy-makers with dumb machines. They will likely perform infinitely better. For the rest of us, the reality is still an important concept and there, we have to make sure the developments are in a beneficial direction.

Intelligence, ultimately, is about decision-making. Humans have been pretty good at it, barring a few narcissistic and mentally ill specimens in full view. They had to survive the dicey world they were handed when they climbed down the tree and stood upright in the African Savannah for the first time. Bad decision-making would have instantly eliminated them. They survived, albeit with less than 10K specimens through a harsh bottleneck. Later, single-cell organisms almost wiped them out on multiple occasions but they survived again. Now, they encounter a technology discontinuity, something that is so foreign to their psyche, the dominant reaction has to be, rejection. And, for the most part, it is. But their brains have morphed into a quantum computer, able to think about possibilities. This could be their Achilles heel, but then, life is not worth living without taking risks.

Educational institutions, still chasing the latest trends to make money, have the ultimate responsibility to get this right. To bring humanity to a level 1 society, we need to move past our instincts, created by tactical objective functions driven by food and sex and embrace intelligence. It is likely that machines will lead humanity out of its perils.




Saturday, March 16, 2019

Micro customization

Recent news (1) that a gastric resident delivery mechanism can deliver reliable, sustainable doses of agents for the long term is important. Innovation in chemical agents has moved ahead of mechanisms that would deliver them at the right time, in the optimum dose, by the best route and to the most receptive site. The ability to optimally deliver the agent is likely more important than the agent itself. In the absence of such delivery mechanisms, manufacturers have stuck to the original blue print - mass manufacturing of pills in a singular dose that shows the best therapeutic index in the population. Personalized medicine, thus, has remained elusive and more importantly, outside the business models of manufacturers.

It may be changing. Ironically, providers have moved ahead of other participants in the healthcare value chain, in the implementation of personalized medicine. Recent advancements in Artificial Intelligence and the availability of abundant data have better  positioned the providers to understand, treat and manage patients, individual by individual. If delivery mechanisms improve and become individually customizable, we can rapidly move into the next level of personalized medicine. Here, we can envision devices that can measure, decide and disburse micro doses to assure optimum delivery and complete compliance. Intelligent devices could be just round the corner, taking advantage of IoT. With embedded intelligence on board, such devices can not only operate as initially primed but also self learn and adjust over time. A couple of decades from now, medical professionals will likely view the current regime to be completely archaic.

More generally, any business that is driven by scale, a blind adherence to singular specifications, will have great difficulty to survive in the future. Technology is readily available, not just for mass customization but rather for individual intervention. This is a regime change that will affect every industry and every business. Getting ahead of this rapid transformation is a necessary condition for success.

(1) A gastric resident drug delivery system for prolonged gram-level dosing of tuberculosis treatment. Verma et al. http://stm.sciencemag.org/content/11/483/eaau6267

Sunday, March 10, 2019

Type 2 Debate

A recent article (1) brings the current diagnostic and treatment regimens for those who are deemed to be at risk for progressing into Type 2 Diabetes, into question. The arguments are fair and the conflicts are clear. But what the article misses is the risk/return trade-off for the wretched condition. The total cost of diabetes related cost in the US alone is fast approaching $0.5 Trillion. With India and China running fast to the precipice, happily feeding on Western food, we do have a significant issue to deal with.

So, there are two important questions. First, do we have a reasonable idea of the risk of incidence and progression? Age old heuristics such as fasting blood sugar has been shown to be utterly useless.The current golden metric, A1c, is likely equally flawed. The answer to this question appears to be an emphatic no. And second, are the people in charge of prescribing the thresholds for diagnosis and prevention influenced by factors outside science. Unfortunately, the answer to this is likely yes.

Medicine has always shunned data and analytics in making decisions. This has set them so far behind, the US spends close to 20% of its GDP for worse outcomes. And aging regulators, still using century old statistical measurements to make decisions, makes this worse. The manufacturers, bloated with conventional statisticians, ever ready to prove what the regulators seek, are progressing backward in time. So, it is not surprising that we have ended up in a confusing muddle regarding a disease state that affects close 20% of the world population.

Now, what? It is easy to show statistics that claims only 2% of the pre-diabetics progress into diabetes. Since we are unclear as to the "precise," demarcation between pre and post diabetes, a more relevant question is what the total cost is on the system. A few thousand years ago, humans invented agriculture - and it clearly had an impact on their health. Meat eating humans for millennia, moved to stuff their bodies are not designed to process. In the last hundred years, they had too much to eat and that will certainly create problems not only for their organs, efficiently designed for very little food but also their infrastructure, designed for a slim body. Half the world's population now exceed the parameters of original designs.

Medicine has to embrace modern technology. Doctors should not be simply following rules put down blindly, regulators have to open their windows and realize that the "p-value," is a dead construct and manufacturers need to understand that their job is not a constrained optimization based on what the regulators think.

It is time for the industry to move on.


(1) https://www.sciencemag.org/news/2019/03/war-prediabetes-could-be-boon-pharma-it-good-medicine

Wednesday, March 6, 2019

Health

News (1) that a group of scientists is attempting to build up a health system in Madagascar from scratch, driven by data, analytics, and pure heart is welcome news. Health, a non-tradable asset, has kept modern humans bottled up. Electing dumb leaders has taken them further behind. The economics of health is a complex question and till scientists and medical professionals take it on, we will swirl around in political ignorance and impasse.

Humans did not have to worry about health until recently. Their able bodies were eaten by cats or crushed by heavier animals much before any required intervention against single cell organisms or auto-immune diseases. It is different now. They are living way past the design horizon. To make matters worse, they have been carrying unwanted weight, putting unexpected levels of stress on their joints. Their organs have been failing because of overuse and their infrastructure is crumbling, not able to hold them up. Modern medicine has acted as a band-aid to prolong life but it has not contributed to maximizing the utility of the individual or society.

A big part of this issue has to do with the inadequacies of systems that apparently intelligent people designed. When there are no markets, incompetence rises. When incentives are misaligned, humans behave as they are expected to, maximizing short term returns. And, humans have a tendency to gang up on the weak and the weary as they learned through the eons of evolution. It is a perfect storm - incompetence, ignorance and non-market designs - that could sink humanity.

It is time to step out. Experiment, revolutionize and let information and technology guide future actions.

(1) http://science.sciencemag.org/content/363/6430/918