# TED Conversations

This conversation is closed. Start a new conversation
or join one »

## how predictable are we?

using statistics and probability theory, humans' actions can be predicted to an extent, for instance, a recent twitter analysis by a university correctly predicted with 90% accuracy when a bunch of people in new york became ill, up to 9 days in advance.

it's this kind of high accuracy rate that has me thinking the famous "what if" question, of, what if accuracy improves to the point that our every action can be pre-determined with high accuracy rates, up to weeks in advance?

i've attempted to explain this question years ago, but nobody bit the lure.

accuracy can and only will get better, as better algorithms are discovered and adopted, and even learnt by a computer program.

ironically, it would be cybernetics that has the possibility of reducing our predictability.

as always i'm interested to read your thoughts!

so, what if?

+1
Share:
• #### Barry Palmer50+

• 0
Aug 15 2012: it's this kind of high accuracy rate that has me thinking the famous "what if" question, of, what if accuracy improves to the point that our every action can be pre-determined with high accuracy rates, up to weeks in advance?

Weeks in advance would be completely impossible. Even theoretical computers need input. To predict weeks in advance you would have to predict all of a human's sensory inputs weeks in advance, and that would require a computer model of the entire globe, possibly the solar system.

THEORETICALLY, if a computer could obtain all of the sensory inputs of a human, it could predict the human's actions. However, the reaction time of a human is measured in milliseconds. Sensory inputs can change a humans' behavior just that fast. So even theoretically, the prediction would precede the actual action by only milliseconds. Can you even describe that as a prediction?
• #### griffin tucker10+

• 0
Aug 15 2012: the theoretical computer would not need all inputs (including all human's inputs) just as you don't need to know what would happen if you pour a jug of water into a glass, you know that it would fill up with water. not all inputs in the solar system are required for this.

also, quantum computers have become blazingly fast. i've heard of a quantum computer that processes data at an _astronomical_ rate, although unverified, but only unverified because there is no computer to compare results with.
• #### griffin tucker10+

• 0
Aug 10 2012: http://itunes.apple.com/au/album/the-all-seeing-eye-project/id258075338

'The Beholder,' track 5, is only available if you purchase the album. pretty much sums up the fear of determinism by most humans.

http://bit.ly/Nxq0dw (free reg) new scientist article on 'the algorithm that runs the world.'
• #### Obey No1kinobe50+

• 0
Aug 10 2012: Foundation by asimov
• #### griffin tucker10+

• 0
Aug 10 2012: yes, i'm familiar with the _fictional_ book.
• #### David Hamilton50+

• +1
Aug 10 2012: I think the only thing they'll never properly predict is art, and confrontation. As a writer, some days I wake up in a particularly emotional state, for seemingly no reason at all, but then I try to stay there, and use it to write something. Sometimes I try to force myself into an altered emotional state, in order to percieve a character, or a debate from an angle I am unable to normally.

I don't think you're going to get a computer to waste it's time on nonsense like that : p
• #### griffin tucker10+

• 0
Aug 10 2012: nowadays, a human can be connected to a computer via electrodes, and the information that is seen by that human can be duplicated (admittedly, not perfectly, early days yet) so eventually i imagine that a computer won't even need electrodes to do this, just by analysing the outside of your body with a camera.

in this sense, if you were filmed while you were sleeping by a computer (sounds creepy, yeah?) the computer could see what you were dreaming, and determine what kind of emotional state you will be in according to your dreams when you wake up.

personally, i don't think i will get a computer to waste time on nonsense like this, but i'm saying it's a possibility. i respect that you may disagree.

EDIT: http://youtu.be/nsjDnYxJ0bo -- brain vision reconstruction, also it's via fMRI not electrodes
• #### David Hamilton50+

• +1
Aug 10 2012: I'd rather it wasn't... I'd like to make this hobby a job : p
• #### griffin tucker10+

• 0
Aug 10 2012: i honestly don't think it would stop you, real or not.
• #### edward long100+

• 0
Aug 9 2012: Some things are predictable because they follow a set of rules.For example I will ask a question and you will be able to accurately predict the answer before you finish reading the entire statement, here goes:What punctuation mark is at the end of this rather contrived query which seems to contain significantly more words than necessary to make its point which is mundane at best? Other things follow no rules and are virtually impossible to predict. For example: What number is missing in the following sequence? 17 2 33 __ 9 R? Arbitrariness and chaos make accurate prediction a crap shoot.Have I completely (and predictably) missed the whole point of the question? Thank you!
• #### griffin tucker10+

• 0
Aug 9 2012: predictions happen on a very small scale, yes. another example would be knowing what happens when you pour water into a glass. the glass will fill up with water. a computer could analyse this kind of prediction.

if i take your sequence of numbers and try to guess the right one, i would most likely fail miserably. a computer, however, could tell you. even if the computer is wrong, the computer can 'learn' from the mistake considering it learns the correct answer. it would be probable that after a few times of putting in a missing number sequence that it would get it right.

this kind of learning from mistakes is exactly how prediction algorithms work, or rather, the algorithm predicts multiple virtual environments that each have a possible correct answer, and eliminates the incorrect environments until there is 1 possible environment left, which will be, in effect, the future, before it happened.

i'm saying chaos and determinism can exist at the same time, however contradictory this may seem. basically it's like not predicting the finer details, just the bigger picture. the finer details could be predicted, but it would simply take too much computing power that could be used for other resources such as predicting further into the future for a given environment.

einstein was big on determinism theory, and so am i.
• #### edward long100+

• 0
Aug 9 2012: I hope I am not diverting you from the deeper idea, but the sequence problem is sufficient to occupy my intellect, so permit a follow-up scenario; I created the sequence as randomly as I am able. I have no idea what the missing number is, but here and now I will declare it to be 230.5436. Are you saying a computer could have psychically "anticipated" the missing number before I typed this post?
• #### griffin tucker10+

• 0
Aug 9 2012: short answer: yes.

long answer: yes, but not psychically. logically speaking, every number possible (including decimals) up to tens of thousands of numbers could have been determined to be in the sequence in each possible environment. many, many times more depending on how fast the computer was.

also, based off what i wrote, an algorithm with enough information about your attitude (how you react to different situations) and what you know, could determine that you were going to say something along the lines of "i don't want to dis-sway you, but..." as well as the number you were about to declare as 230.5436 .

the accuracy of the type of words you use to explain a question (everyone uses their own terminology to an extent) would only be as valid for as much information the algorithm had about your choice of words.

it's not magic. it's science.
• #### edward long100+

• 0
Aug 10 2012: Unbelievable! Literally. The key phrase is: "with enough information about your attitude etc.". That is a contingency which could never be reality. By the way I have not decided what the real missing number will be, so the computer was wrong in predicting 230.5436. Did the computer predict I would do that? Sounds like mumbo-jumbo to me Mr. Tucker.
• #### griffin tucker10+

• 0
Aug 10 2012: the _theoretical_ computer could contain every phrase that you would ever type. consider it like the infinite amount of monkeys producing the greatest works of shakespeare on type-writers conundrum. once the infinite amount of monkeys have actually produced the greatest works of shakespeare, it would be finite.

a _quantum_ computer can store, manage, and calculate quite large numbers, at a very rapid pace. with this kind of calculation ability, producing the greatest works of shakespeare in the same way monkeys would, would be faster. many, many times faster.

the contingency of knowing about your attitudes, what your environment has been or will be, is being determined right now by advertising agencies. TED.com itself uses something called google analytics, which analyses behaviour of each user (supposedly anonymously) to the point where if google wanted to, they could show an advertisement to you that you would find very hard to resist, all because of the information that google knows about you.
• #### Robert Winner100+

• 0
Aug 10 2012: This conversation is using the same tools and arguments that I just read in the Free Will conversation. I could not validate that argument any more than this one. Perhaps to define probalility would help. I do not have to apply advanced math to know if I touch a hot iron it is gonna leave a mark ... no probally about it.

Until you decide what the number will be the system will continually be wrong. If I never decide does the machine continue to work the problem.

If Mr Tucker is right the "chance" just left the "gambol" and he will soon be the next winner of the Irish Sweepstakes .... as well as all other games of chance. And that will be the proof in the pudding Mr Tucker.

One last question. Will this eleminate the random generator?

All the best. Bob.
• #### griffin tucker10+

• 0
Aug 10 2012: hi Bob.

a random number generator used to alter one's actions would slow down the prediction ability, but not totally erase it. i imagine that at first truly random number generators would be useful in this way, until an even faster computer comes along.

• #### Debra Smith200+

• 0
Aug 9 2012: Welll Griffin, a couple of years ago I was at an infection control conference in Toronto where an epidemiologist said that they can now predict which diseases will be in Toronto hospitals in 6 weeks time based on the flooding of the Don River in Toronto. So I am not so impressed with your Twitter report.

On another thread, I said that I think such algorithsms will improve but I highly doubt that one will ever be able to predict thespecific circumstances that make me go to the store because I think it is weird myself. As it turns out - no matter what else I run out of, I never go to the grocery store until I am out of lemons. Surely that is weird enough to count as an unaccountable individual difference. At least I have no hope of statistically predicting it with a multiple regression.
• #### James Zhang30+

• 0
Aug 9 2012: lol twitter report is too weaksauce dang
• #### griffin tucker10+

• 0
Aug 9 2012: i'll reply to this post in the morning. going to bed now. zzz...
• #### griffin tucker10+

• 0
Aug 9 2012: hi Debra, always good to talk to you.

the twitter study is different to the study you mention in at least two ways:

1) the twitter study doesn't mention what kind of flu or illness occurs.
2) the twitter study is accurate enough to predict when an individual will get sick
http://youtu.be/3S2rq2SKTSw

basically what i'm trying to explain is, algorithms are being developed by humans, of which, learn themselves. for instance an algorithm could be developed where all that is needed to start it is to insert a single algorithm that is more than 50% accurate, from there it will develop it's own additional algorithms, based off data that is seemingly irrelevant.

it is quite a challenging thought to be thinking that your own actions can be pre-determined. 'normal' reactions are usually fear, disbelief, or even anger. humans already pre-determine on a small scale, depending on how well they know a person. i've known twins that finish off each other's sentences, and i bet you could think of more examples.

you can buy fridges now that actually order more milk for you when you're running low, and detect patterns of when the fridge opens to conserve energy at the right times, including the occasional mid-night snack.

these sort of algorithms are constantly being improved. what i'm saying is they can and only will improve, as more data (information) becomes abundantly available, so much information that humans won't know what to do with it, but machines could definitely use.

while i don't think it is possible (yet) to determine when you will go to the store, it could be statistically analysed using current algorithms and enough data about your life what day you will go to the store a few days in advance.
• #### Debra Smith200+

• +1
Aug 9 2012: Fascinating Griffin! Maybe all they will have to do is analyze my TED writings. I am open enough.
• #### griffin tucker10+

• 0
Aug 9 2012: perhaps not solely off your TED writings, but if it were, i'd imagine the algorithm would time how long you are away between posts to initially determine when you were actually at the store, then analyse after that your typing words per minute score (determined from timing between rapid succession of posts) that could determine which minerals in your body are low or high (not much potassium, for instance) based on changes in your typing wpm, then predict not only when you were going to the store, but particular mineral-rich foods you were going to buy.
• #### Debra Smith200+

• 0
Aug 9 2012: How exciting, do not forget to factor in the strokes and the fact that I type alot to bring my left hand back to 100%.
• #### griffin tucker10+

• +1
Aug 9 2012: i'm not going to try. it would be an invasion of privacy in my opinion, which is an EXTREMELY important point as mentioned by Jan-Bernd Pauli below.
• #### Lejan .50+

• +1
Aug 9 2012: At least George Orwell's predictions are about to become right.

Within the whole talk no mention of privacy rights at all! What a happy new world we are facing!

I just recently installed a Firefox plugin called 'Ghostery' and from that moment on I became aware for the first time how many websites were already and invisibly tracking my web browsing behaviour for their behavioral statistics ...

On this very TED site this plugin blocks: ChartBeat, Google Analytics, NetRatings SiteCensus and ScoreCard Research Beacon on any new site opened. For those who are interested who is behind this and for what reason, just follow the links at the end of this comment.

So my question rather is, do we really wan't to become predictable and by whom?

ChartBeat
http://www.ghostery.com/apps/chartbeat

NetRatings SiteCensus
http://www.ghostery.com/apps/netratings_sitecensus

ScoreCard Research Beacon
http://www.ghostery.com/apps/scorecard_research_beacon
• #### griffin tucker10+

• 0
Aug 9 2012: i've also discussed this topic before with out anyone responding. it's my opinion that someone or something will eventually determine our future actions, however, it is possible to become less predictable.

if we were to become predictable, privacy would become something of the past. working conditions world-wide could get better as a result. everyone would be held accountable for their actions.

to become less predictable would require cybernetics to calculate truly random algorithms for future actions, but even then, a fast enough super-computer could determine each possible random algorithm that the cybernetics could possibly develop. i imagine only the elite would have such technology at first.

there would be a time-frame window that cybernetics could make humans less predictable, until such a time when people either accept that privacy is most definitely something of the past, or that people don't care about their privacy.
• #### Lejan .50+

• 0
Aug 9 2012: In my opinion everyone should be held accountable for their actions, yet not by statistics!

We could escape this new form of inquisitiveness for a while by avoiding or supressing digital traceability and the connection of collected data to ourselfs.

As I do not believe in determinism, there is a difference for me in 'prediction' and 'probabilities' of behaviour.

Regardless all data stored and analysed about me, no algorithm would ever be able to 'foresee' my actions in advance. It may statistically 'guess' where I may be at a certain day to a given time as long enough data is available on this my schedule. And it would even be quite accurate as long as my life 'swings' within a certain rhythm.

But predictions on 'future actions' and if I did not understand you wrong, would need to include some sort of 'prevision' or 'clairvoyance' on which I have no valid evidence on its existence.

This would include to tackle the 'Laplace's demon' which, luckily, blurrs within the realm of quantum mechanics.

So let's get even more blurry to fool the elite and whoever this is...
• #### griffin tucker10+

• 0
Aug 9 2012: luckily, it would require a quantum-state computer to calculate the future or past. quantum-state computers aren't exactly easy to come by, but they do exist.

respectfully, i disagree with the idea of 'fooling' advertising companies data collection bins and monitoring agencies. i like to be able to google for something, then have a relevant advertisement moments later surprise me with exactly what i want. this is just my opinion, though.

i don't have the computing power nor programming capacity myself to develop such a prediction algorithm, i just like to explore extreme ideas of the future, however improbable or even impossible.

privacy is a real concern these days. would you rather only the elite watch you, or a possibility that everyone can watch you? in my opinion it would be one or the other, choosing none isn't an option.

that being said, there would be a possibility that 'de-radioactive' zones could exist where upon no computerised or internetworked cameras are allowed, or even radio-emitting devices, therefore making it harder (but not impossible) to monitor those areas.

entering such an area would make it many times more harder to monitor actions, and leaving such an area into a monitored zone would make it harder at first to predict such actions, getting progressively easier and progressively more accurate.

EDIT (4 hours later) why you would wish to make a fool out of anyone, including me, i may never understand.
• #### James Zhang30+

• 0
Aug 9 2012: Cool! I'll check those out sometime
• #### James Zhang30+

• 0
Aug 8 2012: I don't think 100% of all actions can be explained, simply because I think there are infinite (or infinitesimal) number of actions. And we still haven't been able to explain the random "hiccups" an atom can do.

But, indeed, we can eventually explain and predict a lot of things in the future, but randomness or probability will still be hard to explain imo. Like when I spin a coin on the table or roll some dice, are we able to predict the likelihood of chance?
• #### griffin tucker10+

• 0
Aug 9 2012: at the planck level, you're right, it would be near-impossible (or at least a very, very fast super computer required) to predict each atom. that being said, the idea here is to improve upon accuracy, not make it 100% accurate.

for instance, a super computer could analyse each frame in a blu-ray video disc, and predict the last minute's frames according to patterns generated by each individual pixel and their surrounding pixels. each frame has a finite amount of pixels. each pixel has a finite amount of colour possibilities.

of course to do this kind of predicting using current probability theory and statistics would be impossible.

however, if the computer program were to learn it's own method of predicting each individual pixel's colour (programming term: neural networks) then i think it's not only possible, but probable within 20 years.

this could mean you could set up a camera, take a few billion pictures, take away the camera, and let the program determine not only the future of what happened in the camera's field of view, but also what happened before it.

a fast enough computer could determine what happened at both the beginning and end of the universe, if there is one or both.
• #### James Zhang30+

• 0
Aug 9 2012: Could there also be more conditional outcomes though?

Like, the computer cannot really know for certain what will happen, unless this one thing happens that triggers a certain path, and not the other path.

"a fast enough computer could determine what happened at both the beginning and end of the universe, if there is one or both."

Wouldn't this computer take an infinite amount of time to calculate it? lol
Unless we're skipping some frames or something like that
• #### griffin tucker10+

• +1
Aug 9 2012: accuracy doesn't have to be 100%. it only has to be 99.999999999...etc % accurate to form an image of one frame in the future or past. the human eye couldn't detect such a change from this kind of accuracy.

by definition, if there is a beginning to the universe, then it can be seen. i'm not entirely sure if there is an end to the universe, but if there's one there, skipping frames is exactly the method to see it, and there could be other methods too. same with the beginning of the universe.

the computer would only need to be equal to or faster than the universe to calculate the entire universe. considering the big bang theory, the universe started out small, and grew. this then means calculating would only get faster and faster to determine the eventual beginning.

with compression algorithms, a frame could be displayed even faster for a smaller dimension window (such as a 1080p frame) for determination of the future or past.
• #### James Zhang30+

• 0
Aug 9 2012: Hmm, it's definitely an interesting thought, maybe we don't need to know finer details to see the bigger more abstract picture of the skipped frames...
• #### griffin tucker10+

• +1
Aug 9 2012: indeed. an algorithm could be developed (learnt) by the program to determine that every x frames, a difference of y would achieve z, where z is the result after x frames. of course most likely not as simple as this, but the gist is there.

in my mind, you would need all of the surrounding pixels of a particular pixel as well as determinations from the bigger picture to determine the next few frames. i imagine in this way that the further in the future/past you go, that the smaller the reference frame that is of a higher accuracy.

in other words, the further into the future/past you go with a circular image (most efficient for this type of algorithm) the smaller the circle of accuracy becomes. ironically it could be compared to a crystal ball, except this is science, and not dis-proven science at that!