linkedin facebook twitter rss

05 Sep Fuzzy Idea Wars in Always Never Land

Smart Phone Open Mind

Joe Roushar – September 2015

Trusted Rx

How should a digital device answer a “Should I…” question? “Should I put on my left shoe first, or my right?” “Should I take the alternate route to avoid traffic? “Should I get a more fuel-efficient automobile?” “That stock price is lower, so should I buy now, or is the market likely to slide further?” “Should I vaccinate my newborn child?” While some answers to prescriptive questions can be benign or trivial, many could be more consequential. Thus the best approach is to cover your bases by pointing the asker to validated references and issuing major disclaimers, such as: “This system will not recommend specific investments. You should contact a qualified investment advisor to help determine investment strategies suitable for your unique situation.”

But is it possible to devise a system that would do a better job than a human in guiding investors, just as riding in a self-driving car may be safer than one with a human at the wheel? It may be possible, but perhaps not desirable. If you had 10,000 copies of a mobile app with the same logic guiding investors, the app itself could affect markets and create risky and transient bubbles. Sometimes fuzzy statements are better than absolute.

Understanding Context Cross-Reference
Click on these Links to other posts and glossary/bibliography references


Understanding Context Intro #10

Intro Icon


Table of Context

VaccinationThe opposite sides in the vaccination idea wars seem to be intransigent in unhealthy ways. The conspiracy theorists and fear-mongers lined up against vaccinations warn of unknown side-effects and suggest that autism may be connected to vaccinations. Many in the medical community, especially epidemiologists point out that vaccinations have nearly eradicated some diseases such as Small Pox and Measles, and use that as a basis for suggesting that all vaccinations are always good. I heard an epidemiologist suggest that even though the flu vaccine for the 2014/2015 flu season targets a strain that is not common in this season, and has virtually zero immunization effect for the strains that are active, that everyone should get it anyway to get in the habit. He went on to engage in some fear-mongering of his own, suggesting that people who choose not to immunize themselves and their children endanger the health of the broader community. This may have a shred of truth in it, but seriously, trafficking in shreds of truth is highly unprofessional.

For full disclosure, I need to confess that I have been lavishly vaccinated and have not been afflicted with measles nor rubella. I have, when younger, been afflicted with horrible allergies. But rather than identifying the source (probably pollens) and getting treatments, I have let my immune system deal with it. The result: over the years I have developed immunities to allergies that used to seriously mess me up. Is this approach best for all people? I think not. I think “always” and “never” tend to be traps that impede proper understanding.

Another trap in idea wars that wants a fuzzy response is the ambiguity between causation and correlation. Yes, many parents have identified symptoms of autism soon after their child is immunized. Is the immunization the cause or trigger of the symptoms, or is it just the time of life when those symptoms first become apparent whether the child is immunized or not? If the condition was present but not evident before the immunization, was it the contents of the needle that triggered it, or the trauma of the needle stick itself? Until causal links are scientifically established, it is completely undisciplined and wrong-headed for a system or person to infer that the temporal sequence of events or phenomena prove causation. The last thing we need is wrong-headed systems! provides some useful statistics:

  1. The Centers for Disease Control (CDC) estimated that 732,000 American children were saved from death and 322 million cases of childhood illnesses were prevented between 1994 and 2014 due to vaccination. [45]
  2. About 30,000 cases of adverse reactions to vaccines have been reported annually to the Vaccine Adverse Event Reporting System (VAERS) since 1990, with 10-15% classified as serious, meaning associated with permanent disability, hospitalization, life-threatening illness, or death.

Virus Illustration The US CDC recently announced that polio immunizations given in the US from 1955 to 1963 (I was born and vaccinated in 1959) were unknowingly contaminated with Simian Virus 40 (SV40), a potentially cancer-causing virus. The CDC has correctly stated that there is no study showing a causal link between the tainted vaccine and the increased incidence of cancer, but they failed to mention that after a certain number of years, time in which such a virus may survive and reproduce in a living host, establishing a causal link is virtually impossible. Their mentioning the absence of documented evidence seems to imply that folks shouldn’t worry about it. They also failed to mention that SV40 is believed to damage the p53 tumor-suppressing gene in humans which would indirectly increase the incidence of some cancers.

I’m not saying the vaccine makers were at fault for delivering unknown carcinogens in their product. In fact – I’m glad I never got polio. Questions like this are just plain complex! I love Richard Feynman’s video on knowing versus understanding in which the fuzziness of things leads to completely different possibilities: (Knowing Versus Understanding).

If, in a highly free society with universal information availability, well-funded government organizations can obfuscate facts so easily and casually, how can anyone trust some tech startup to deliver apps that a person can trust for advice? Answer: you can’t. Thus, any system that wants to provide question-answering capabilities can never say “never” or “always” and SHOULD NOT dispense authoritative prescriptive advice – only possibilities.

This brings us to the question of fuzziness in computing systems. One of the most difficult technical questions I have had to deal with in my work to improve automated language understanding is: “Where can we best apply fuzzy logic and processes to achieve the best possible results?” In human language processing and communication, opportunities for non-deterministic processes and outputs are ubiquitous. In my post on “Unlocking the Power of Unruly Systems” I mentioned how Google has programmed it’s cars with general rules that help it respond well to situations that it has never encountered before and weren’t specifically programmed in, such as ducks, deer and gandmas in wheelchairs. Human responses to new situations and even paradoxes are very dynamic and involve inherently fuzzy processes.

Knowledge and Fuzzy ModelsLanguage is fuzzy coming in and there is little harm in keeping it less prescriptive (deterministic) coming out. Consider my post on Knowing, Thinking and Believing, and my assertion that machines could (or should) follow the “Japanese model of being polite in its assertions.” I want systems to be a lot smarter than they are today, and paradoxically, the answers that smarter systems provide are likely to be less cut and dried, and more subject to individual interpretation. The machine that never says “never” nor “always” may end up being the smartest machine around.

Click below to look in each Understanding Context section


Comments are closed.