Predicting The Future Is Still Really Tough

from the the-silicon-crystal-ball dept

Across many disciplines, it's common for experts to make bold forecasts about what the future may hold. Ostensibly these forecasts are based on science or sound models, but in the end, nobody can confidently predict the future. Even those forecasts that turned out to be correct don't say too much, because the forecaster could've just been lucky. For some time, software companies have been pushing software to help make accurate predictions of future patterns, but it's not clear how effective this really is. A recent study takes issue with this technology, particularly as it relates to predicting human behavior. Although models can attempt to paint a picture of how people will behave on the aggregate (as a group), they can't say much about the actual individuals that comprise the group -- human behavior is simply too variable to be reduced to an average or a smooth bell curve. What's more, decision makers don't know how to use or understand the data that they receive, which further compounds the problem. Predictive technology will remain an interesting area of further study, but it's foolish to think we're getting to the point that we can see the future.


Reader Comments (rss)

(Flattened / Threaded)

  1.  
    identicon
    Jonas, Jul 19th, 2007 @ 8:47am

    Where's Hari Seldon when you need him...;)

     

    reply to this | link to this | view in thread ]

  2.  
    identicon
    dorpus, Jul 19th, 2007 @ 8:48am

    Central Limit Theorem

    Human behavior may not fit smooth bell curves -- but they may fit gamma distributions, Poisson processes, curved exponential distributions, or Markov chains rather well.

    Would you call me crazy if I said that the time between deaths in the state of Alabama follow the Poisson process predicted by theory almost perfectly? Even the dips and spikes can be explained by the human tendency to record times in multiples of 5; the dips at 0 and 1 correspond to calling center capacity.

    http://www.mixedasians.com/member_pics/haplotype_1183490705.gif

    Would you also call me crazy if I said that death rate is an almost perfectly linear function of temperature? Even the spreading of points near the end of the lines is consistent with theory; with fewer readings at the temperature extremes, more variation is expected.

    http://www.mixedasians.com/member_pics/haplotype_1183414347.gif

    So yes, seemingly unpredictable events like death may turn out to have very predictable patterns. There is no correlation between our intuition of "predictable" events and the pattern of their occurrence.

     

    reply to this | link to this | view in thread ]

  3.  
    identicon
    Anonymous Coward, Jul 19th, 2007 @ 9:31am

    Re: Central Limit Theorem

    particular events are one thing.

    but again, like the article says, behavior is completely different.

     

    reply to this | link to this | view in thread ]

  4.  
    identicon
    Daniel Berch, Jul 19th, 2007 @ 10:13am

    Predicting Normality vs. the Extreme

    Interesting... I was just writing about this on my blog ( http://danielberch.wordpress.com/2007/07/19/decision-analysis-part-2-predictionor-wild-guess

    Essentially, the question is: are we trying to predict normal events or extremely unlikely events. Our ultimate goals should determine what type of approach to use.

     

    reply to this | link to this | view in thread ]

  5.  
    identicon
    Dr E, Jul 19th, 2007 @ 10:35am

    Some Human Behavior Is Highly Predictable

    The comment about human behavior not being reducible to smooth bell curves is not true. Actually, most measures of human behavior fit bell curves very well when the sample of individuals is large and randomly collected. It is not the variability in human behavior that is a problem.

    The claim that human behavior is unpredictable (at the level of the individual) is too broad a brush. Some aspects of human behavior can be predicted with great accuracy, at the individual level. For example, in the US, were I predict that drivers of cars will drive on the right side of the road when they pull out of their driveways, I would be right almost 100% of the time. Whether human behavior is predictable depends on the particular behavior, the nature of the measures used on which to base predictions, and the precision of the prediction. A common demonstration in psychology classes to prove how easy it is predict human behavior at the individual level is this: When taking about prediction and control of behavior, I would right down the following predictions, "Walk to front of room", "Sit on the table in front of class". I would then ask students whether they thought human behavior was predictable. Many would say they didn't think so. I would then say I would like prove them wrong. I would pick a student at random and say I want to do a demonstration that proves I can control your behavior. I would ask the student to come up in front of the class so we could run the test. I would ask them to sit on the table while I prepared the test. Every student in every class did as I requested. After they sat on the table, I would show the class my written predictions. I was right 100% of the time. Behavior can be predicted when researchers have knowledge of the specific situation and know the effect of that situation on people. The situation has to be one that controls specific behaviors. College classrooms exert great control over people. Students are all sitting. They are all dressed. There are clear consequences for violation of the social rules. Under these specific conditions behavior can often be predicted. Performance on final examples is often highly predictable from performance on midterms (the correlations often are .7 and above).

    With that said, however, it should be noted that most research conducted by psychologists and psychiatrists is NOT designed to make point predictions. The research not only deal with averages over many "equivalent" examples (different people), something that is common in all sciences (let's see a physicist predict the trajectory of a particular particle), it also is rarely parametric (unlike physics). Thus, when studying say, the relationship between drugs and future crime, researchers almost never attempt to fit particular theoretically derived functions relating the amount of drug use (but how measured?: frequency of use, dosage per use, type of drug, mixture of drugs, etc.) with the "amount" of crime (but how measured?: frequency of arrests, most serious crime committed, number of times convicted, mix of crimes, etc.). The lack of parametric experimental designs and the lack of agreement about how to measure key conceptual variables is a major reason more aspects of human behavior is not more predictable. Not the fact that human behavior is variable.

    Another reason human behavior is difficult to predict is the fact that the situations in which humans find themselves is highly variable and the fact that a good portion of human behavior is under the control of situations in which we find ourselves. Crimes occur in many different specific situations. If the situations do not reoccur, the behaviors controlled by those situations will not reoccur. This is why simple measures (e.g., drug use, family structure, etc.) do not do a very good job of predicting. They ignore the specific criminal settings.

     

    reply to this | link to this | view in thread ]

  6.  
    identicon
    dorpus, Jul 19th, 2007 @ 10:51am

    Re: Re: Central Limit Theorem

    Human behavior has finite variance. When narrowed to suitable contexts, individual behavior may become very predictable. How people behave at an airport, a Wal-mart store, may be very predictable down to the individual level.

     

    reply to this | link to this | view in thread ]

  7.  
    identicon
    suv4x4, Jul 19th, 2007 @ 10:57am

    You are unique, just like everyone

    "human behavior is simply too variable to be reduced to an average or a smooth bell curve."

    The grand illusion we're unpredictable: it's hard to accept you're a little fish in a flock of clones. But you are.

    Ever wondered how come exit poles work (most of the times reliably)? Wow, so we're not so unique: a large majority of the population mirrors a small representative group picked out semi randomly.

    Sucks doesn't it. But maybe you can be non-conformist and wear emo white/black makeup and laugh at people being similar. That just puts you in another large group of look-alikes.

    No escape from this trap.

     

    reply to this | link to this | view in thread ]

  8.  
    identicon
    MrGreenthumb, Jul 19th, 2007 @ 11:16am

    one in a million

    One of my favorite was to conceptualize the liklyhood of being similar to another individual is to simply look at the numbers.

    If you are "one in a million", there are 6,602 people just like you.

    Which is to say that you are unique and different from 999,999 other people- unlikely.

     

    reply to this | link to this | view in thread ]

  9.  
    identicon
    The Man, Jul 19th, 2007 @ 11:37am

    Predictions in 1900

    If you would have asked someone in 1900 to predict what they thought a major problem in 2000 would be you would probably get the answer, "where will they get enough horses for the population increase?" and "How will they fix the pollution problem of all the horse shit in the streets?"

    Would someone from the 50's be able to predict that almost all power in Europe. Anyone in the 70's think about video on a cell phone?

    you can not predict the future. Computer models are just humans puting in code from guess they make.

    That is why Global Warming is BS. Sure there is a slight warming trend, but will it cause global devistation? How are they predicting that except of wild ass guesses? To think that we right now can even imagine what type of fuel and vehicles people will use is crazy.

     

    reply to this | link to this | view in thread ]

  10.  
    identicon
    Overcast, Jul 19th, 2007 @ 11:55am

    Ha!!

    Maybe the future's hard for you people without shamanistic bones to throw onto the ground! Just a toss and the bones tell me what's goin' down tomorrow.

    These software companies could do to invest some money in bones, crystal balls, or even various entrails and a decent soothsayer. I have those and some mystical crystals I would be more than willing to sell them.

    Only $4200.00 for a nice set of bones, how could they pass it up??

    hehe

     

    reply to this | link to this | view in thread ]

  11.  
    identicon
    Anonymous Coward, Jul 19th, 2007 @ 12:15pm

    4201 my bid

     

    reply to this | link to this | view in thread ]

  12.  
    identicon
    Anonymous Coward, Jul 19th, 2007 @ 1:53pm

    Sure it may be possible to predict the behavior of people on litte things like who do you vote for since your choices are very limited (the number of candidates + 1 for the choice not to vote) but for something that vastly more possibilities.

    If I walk into a video game store where there are 300 different games (assuming I have the required console/pc to play them all) how do you predict which one I will buy? It may be somehow possible to narrow it down to a specific genre or two which would reduce the selection from 300 to about 20-40 games but how do you get the final prediction of which one I'll buy?

     

    reply to this | link to this | view in thread ]

  13.  
    identicon
    dorpus, Jul 19th, 2007 @ 2:01pm

    Re:

    You're right, it would be hard to predict which exact game you will buy.

    On the other hand, if we had access to, say, your credit card records, we may be able to predict the timing of your next purchase fairly accurately.

     

    reply to this | link to this | view in thread ]

  14.  
    identicon
    Charles Griswold, Jul 19th, 2007 @ 5:11pm

    Human Behaviour...

    Brian: Please, please, please listen! I've got one or two things to say.
    The Crowd: Tell us! Tell us both of them!
    Brian: Look, you've got it all wrong! You don't NEED to follow ME, You don't NEED to follow ANYBODY! You've got to think for your selves! You're ALL individuals!
    The Crowd: Yes! We're all individuals!
    Brian: You're all different!
    The Crowd: Yes, we ARE all different!
    Man in crowd: I'm not...
    The Crowd: Sch!

    Monty Python quotes for the win! :-)

     

    reply to this | link to this | view in thread ]

  15.  
    identicon
    SailorRipley, Jul 20th, 2007 @ 6:05am

    Re: Central Limit Theorem

    Even if death has a mathematically nice distribution, this won't do jack shit to help you predict even one individual('s) death, so your claims, even if true, have no relevance

     

    reply to this | link to this | view in thread ]

  16.  
    identicon
    SailorRipley, Jul 20th, 2007 @ 6:07am

    Re: Some Human Behavior Is Highly Predictable

    no offense, but if I was sitting in your class, I would think, and probably tell you, that your demonstration was quite lame...you might as well have written down on the cards "I will be asking a student to come to the front of the class" and "I will ask a student to sit on the table"

     

    reply to this | link to this | view in thread ]

  17.  
    identicon
    SailorRipley, Jul 20th, 2007 @ 6:12am

    Re: You are unique, just like everyone

    geez, what an excellent example...

    here's how exit poles work: you take a sample group that is suppposed to be sufficiently large to

    1) be representative for the entire group of voters
    2) prevent any disproportionate representation of a subgroup

    the fact that exit poles predict the outcome rather accurately has nothing to do with people being similar or not, it just has to do with math and statistics...

    if people's behavior was predictable, one wouldn't need to do the exit polls...

     

    reply to this | link to this | view in thread ]

  18.  
    identicon
    Anonymous Coward, Jul 20th, 2007 @ 6:14am

    Re: Predictions in 1900

    nice way to prove your retardedness

     

    reply to this | link to this | view in thread ]

  19.  
    identicon
    dorpus, Jul 20th, 2007 @ 10:01pm

    Re: Re: Central Limit Theorem

    Even if death has a mathematically nice distribution, this won't do jack shit to help you predict even one individual('s) death, so your claims, even if true, have no relevance

    Depending on the circumstances, it is possible to predict patient outcomes with high probability. If a patient's death is certain, or has a high probability, they will receive lower priority than those that have hope. Health care outcomes are less certain because we already exclude those whose death is certain -- e.g. patients with no chance of surviving an organ transplant are not given them in the first place. Those who have some chance are given the organs. Those whose survival is certain are not considered for mortality statistics, hence the mortality statistics appear less certain than they really are.

     

    reply to this | link to this | view in thread ]

  20.  
    identicon
    dorpus, Jul 20th, 2007 @ 10:04pm

    Re: Re: Re: Central Limit Theorem

    e.g. if you are a healthy young male with a cold, there is a 99% probability that you will be better within 2 weeks.

    Such statistics aren't publicized much because they are too common sensical, but it is a part of the vast knowledge of very predictable outcomes.

     

    reply to this | link to this | view in thread ]

  21.  
    identicon
    dorpus, Jul 20th, 2007 @ 10:13pm

    If you post on this forum

    There is a very high probability that you live in a place where there is at least one chair, two tables, paper towels, running water, and electricity. You probably don't eat mice on a regular basis, you've had at least one soft drink in the past year, and riding a car is not an amazing experience for you.

    On a worldwide basis, that puts you in a relatively select group of people, and I have narrowed down your behavior considerably.

     

    reply to this | link to this | view in thread ]


Add Your Comment

Have a Techdirt Account? Sign in now. Want one? Register here
Get Techdirt’s Daily Email
Save me a cookie
  • Note: A CRLF will be replaced by a break tag (<br>), all other allowable HTML will remain intact
  • Allowed HTML Tags: <b> <i> <a> <em> <br> <strong> <blockquote> <hr> <tt>
Follow Techdirt
A word from our sponsors...
Essential Reading
Techdirt Reading List
Techdirt Insider Chat
A word from our sponsors...
Recent Stories
A word from our sponsors...

Close

Email This