Cost of storing all human audio-visual experiences | Hacker News Reader

Cost of storing all human audio-visual experiences

lists.w3.org

83 pointsterrahawkes39 comments

I know it may be a little tangential to the original point, but I reckon you could cut this down considerably if you had more accurate psychophysical models than we do. Humans are probably not actually processing anywhere near 1.25 mbps of input, and as far as our actual experience, we're discarding the vast majority of the information we even take in.

Consider, for example, that if you could perfectly predict the exact path of a person's fovea as they watched an SD movie, you could aggressively compress everything more than a dozen pixels away from that. A few pixels further, and you wouldn't even need to store chroma information at all. Let's say you can do that at 100 kbps, and the pixels on the fovea are 1 mbps. That would put your total bitrate at about 105kbps for video. If you don't optimize audio at all, that brings the cost down to about $170B/year.

And that's not even with new psychophysics; it's just better input. Sure, playing it back would be hard, but in principle, you could fool a brain into thinking it was experiencing the data firsthand. I think it's reasonable to think this is still a massive upper bound. It's not unreasonable to think that a lifetime of human sensory experience could be encoded on the order of hundreds of gigabytes, if not less.

If I were recording my life, I would not want to discard that, however -- I don't want to record merely what I actually saw/noticed, but rather what I would have seen or noticed if I had been looking elsewhere in my field of view.

Charlie Brooker did an episode of Black Mirror[0] based on the premises that everyone records every waking moment of their life.

[0] http://en.wikipedia.org/wiki/Black_Mirror_(TV_series)

I wanted to do something like that but it didn't work out very well. Mostly "everything" meant on my computer (audio files are massive!) or phone and I couldn't wear a video camera at work or scan every piece of paper I came into contact with.

But still I find it interesting it's almost possible.

It's interesting to think about what these factors actually mean. The perception of what's "good" quality will continue to shift hugely. Life expectancy, hopefully, will go up. Hard drive prices actually haven't come down again after the floods, so I'm guessing they'll stay artificially inflated for a long time - especially considering storage density won't increase in the coming years as we seem to hit a wall there as well.

So that's 60€ or $60 per TB for the foreseeable future. Let's say we're using a well compressed H264 stream with a 500 MB per hour data rate for 16 waking hours per day => 3 TB or 180 USD per year and person. Plus electricity and, realistically speaking, bandwidth costs.

Still, it would be nice to have a searchable experience database like that. Ideally I'd want that with a cybernetic eye ;)

> especially considering storage density won't increase in the coming years as we seem to hit a wall there as well

http://www.southampton.ac.uk/mediacentre/news/2013/jul/13_13...

There is any number of similar articles you could have linked to. However, there always seems to be a huge disparity between sensationalistic storage innovation announcements and what's actually ending up on the market.

Also, just snidely posting a link that a good number of people here already know without taking the time to write an actual comment seems kind of lazy.

how exactly does a well compressed H264 stream at 500 MB per hour work out to 3 TB a year? Wouldn't a years worth of video and audio come closer to thousands of terabytes?

He assumes you only record waking hours and that you are awake 16hrs a day so:

16 * 365 = 5840hrs * 0.5GB per hour = 2920 GB or 2.9 TB. Add another 8GB for a leap year it still doesn't get you over 3TB.

It's funny, because in 1000 years, this is an anthropologists dream come true (assuming the storage can be preserved that long by eventually copying to a less volatile medium).

Is it worth sacrificing our privacy now, so that their perfect glimpse of us in "the stupid ages" teaches them enough to learn from all the mistakes we keep repeating?

And as long as it doesn't have DRM (although I assume they'll be able to break it easily).

Have we learnt from the middle ages, or do we find the circumstances too different from now to draw authoritative conclusions? (honest question)

Yes I think in some regards we definitely have - like taking science as fact rather than religion as fact - is mostly the status quo. That certainly wasnt the case 1000 years ago.

Depends where you live. I've toured entire counties that seem to think Al Gore is the devil for daring to say their hummers are melting ice caps.

But on principle, those are the kinds of people that didn't study history. Or much of anything, at all. So we do have evidence we learn from our history, just that we don't inspire enough people to learn about it.

Man, this brings up a lot of privacy questions. Especially around who gets to look at what. I mean is this a personal archive that only unlocks after your death? Is it open to anyone? Is it available to law enforcement?

This about this:

Your entire sex life is replayable. This includes being with someone or solo action. Do you want you grandkids watching that?

No longer are we contemplating whether someone committed a crime, we would just watch the replay. Would you even commit a crime if you knew it was being recorded? You wouldn't need witnesses, just their feeds. Heck, you don't need forensic evidence really - if you never find a murder weapon who cares? It's on video! The innocent always have an air-tight alibi. It's on video!

Would not allowing law enforcement to look at your 'life feed' automatically make you a suspect?

Robin Williams played in The Final Cut[1], which is a film that explores the idea of making a post-death movie called a "Rememory," usually shown at the deceased's memorial service. Williams plays the role of the editor of the Rememory, and he faces the internal and external dilemmas associated with uncovering all of a person's gritty life experiences.

[1] http://en.wikipedia.org/wiki/The_Final_Cut_(2004_film)

This post reminds me of a cool project, where the author is recording a single second of video every day, for the rest of his life.

http://www.ted.com/talks/cesar_kuriyama_one_second_every_day...

assuming somekind of google glass attached to a giant world wide data storage. Could it be compressed? (which got me to thinking about compression and the larger the size of the data stored the more likely the chance of larger and larger streams being duplicate. How many parts of the image do I have to store of my desk. Also, could we limit it to interesting moments(how many times do I have to record me eating cheerios in the morning.)

Presumably you could also use lossy compression across users, especially for the boring parts. I imagine that, within the confines of my bowl, my milk and cheerios look pretty similar to yours. Do the actual differences matter, as a record of some particular breakfast experience? I suppose only if they float into some surprising configuration.

Dedupe for life experiences.

Whenever you have many recordings, the thing to do is building a 3D model of the world from it and then keep the camera movements for each camera, if you still want them.

I guess this or something similar to it will soon happen at large sports venues, say at the Super Bowl. It will start with the field, but if technology ges better and cheaper, they could include the audience, too.

Imagine buying an improved version of your memory: a fully immersive recording of your view of the game, with some improvements. For example, that guy that jumped up, obscuring your view of the winning touchdown? In the recording, he happens to sit a few inches to the left. View that a few times and wait a few decades, and you won't even remember that you didn't see that famous moment in sport.

And yes, that can be used for other purposes, too. Imagine that I appear in thousands of photos at some time and place while in reality, I was robbing a bank at the other end of the world.

"If storage prices continue to develop as they have since the 1980s," bzzzt and thanks for playing.

Remember this truism, "In real life, exponentials are s-curves."

That said, there are folks today who record 24x7 stereo 44khz as a 'life diary.' It isn't that hard to store these days and compress down. One could imagine adding video to the mix.

>bzzzt and thanks for playing.

There may be tens or hundreds or even thousands of years before that exponential curve looks like an s-curve, making his point more relevant that yours.

Please compute for me the number of bits per atom you will be storing if current trends continue for a thousand years.

Please compute for me when we'll reach the peak of the exponential growth, and move to the "top of the s-curve". I realize it will happen at some point. I don't pretend to know when.

Do you really think it will occur within 10 years, which is what the author uses for his extrapolation in the article, and which would justify the parent commenter implying the back-of-the-envelope calculation is wrong (with a snide comment)?

"I realize it will happen at some point. I don't pretend to know when."

Partial knowledge is possible. We can be quite comfortable ruling out "thousands" of years. It would require a total, utter rewrite of physics. While this can never be 100% ruled out, I do tend to think that when that's your last defense, you've gone too far. If people a thousand years from now look back and laugh at me for thinking that it's impossible to store multiple petabytes on a single hydrogen atom, I hope they'll at least have the courtesy to admit it was by far the most likely hypothesis given the quite substantial amount we know about particle physics right now.

Well I've been pretty involved in storage for the last 10 years, 5 of them at Network Appliance and then later another 4 at Google dealing with their storage scale. And worked with both Seagate, Fujitsu, and to a lesser extent Hitachi as they have worked to increase the density of what can be reliably recalled. More interesting has been watching the struggle of thermal noise and the ability to push past areal densities of 300 - 400 Gbits/inch^2.

There are certainly nano-scale technologies which seek to store information in the 'spin' of electrons (I'd be hard pressed to see you get better than that) but may become impractical if the network bandwidth gets to the point where the size doesn't matter any more. Specifically, what matters to the end consumer is that they can get what they stored into something which can use that information. If it comes from disk across the room that is just as good as a local on board disk if the bandwidth is the same.

The effect this reality is having on storage is that fatter (and increasingly more fragile) drives, are becoming less useful to consumers than larger but more reliable storage attached via a protocol (be it iSCSI, iCloud, S3, or NFS).

So once people are unwilling to pay to carry it around, the ability to recoup your investment in making it possible to carry around a device that can read electron spin is less and less likely. And your s-curve will become clear, when for the last 5 years the drives have all been at most 6 maybe 10TB and the price of those doesn't seem to fall all that much.

I realize disk growth has been phenomenal "your whole life" as CPU performance growth was for most of mine, but CPU performance growth has kinda sputtered big time. We've been multi-coring for a relatively long time now. Storage systems (especially random access read/write systems) are in the same boat. But the glory days are over.

I am curious to see if storage will have its 'multi-core' moment. That will be interesting if it happens.

>I am curious to see if storage will have its 'multi-core' moment. That will be interesting if it happens.

Thank you for the insight and explanation.

I suppose I took issue with your above comment as I saw it "missing the point" of the author; if storage capability keeps growing (even +/- its current rate), it will be extremely cheap by modern economic standards to record...everything.

My bet is still on a break-through in write-once storage, which wouldn't just be useful for life logging but could also work wonders with a storage system like Datomic (with persistent/purely functional data structures.)

I read an interesting paper in Science about backing up data to DNA as a future archive scenario. At the time I hadn't realized you could just store DNA by drying it out and putting it on an index card, I had visions of cryo vats of test tubes :-). Such a system would be a 'write once' sort of solution.

Indeed. That's exactly the kind of thing I'm talking about, although from what I've heard access times for DNA are still a bit steep though ;)

Episode 3 of this tv show is very relevant: http://en.wikipedia.org/wiki/List_of_Black_Mirror_episodes

What about the cost of sharing all human audio-visual experiences? Because simply storing them, while cool and/or terrifying in and of itself, will leave a lot to be desired.

I think a typical DVD is more like 4.5 Mbps...

That's true, but the OP said "Typical DVD-quality SD bitrate". DVDs use MPEG-2 encoding, but there are much better codecs available today, such as H.264, which can achieve equivalent quality at substantially lower bitrates.

Opinions vary on what the "best" bitrate for transcoding DVDs to H.264 is, and it depends on what your content is too. In a quick search I found people claiming as low as 500kbps(+128kpbs for audio)[1], which is half what the OP uses, all the way up to 2500kbps[2], double his figure. All of which leads me to believe that 1.25Mbps is a pretty reasonable napkin number.

[1]http://forums.macrumors.com/showthread.php?t=842536

[2]http://www.networkedmediatank.com/showthread.php?tid=6867

Once whatever successor to Google Glass becomes ubiquitous, I can see this becoming a reality.

This is an advertising research agency's wet dream come true, unfortunately.

I wonder, with personal income on the decline this past decade, will advertisers spend less money on adverts or more on people who have more disposable income from mined data?

Doubt it. If firms could analyze and deconstruct the purchasing decisions and preferences from such 24/7 feeds, it will just allow them to more efficiently funnel money from the consumer to the top. The lower income market is intensely profitable, mostly due to the limitation of choice a disenfranchised person has.

It's disconcerting that, literally knowing everything about you, puts advertisers in a position to accurately target you in insidious ways.

Advertising is incredibly good at convincing people that they need [product] as it is. Assuming the 'rational actor' theory ever treaded water, this sinks it.

It does seem like the logical progression for Google and friends. What will be interesting is how such invasive monitoring will be pitched as innocuous. People seem to be a-okay with corporations mining their data, though.

>The lower income market is intensely profitable, mostly due to the limitation of choice a disenfranchised person has.

From what I've seen, this has actually started to attract foreign owned corporations in some industries like food (snacks in particular), I'm sure that's not the only industry.

>It does seem like the logical progression for Google and friends. What will be interesting is how such invasive monitoring will be pitched as innocuous. People seem to be a-okay with corporations mining their data, though.

Agreed. From someone trying to get into the data mining space, I feel like the "positive" direction to go in is to basically admitting that there is no putting the lid back onto this box (simply too valuable to many services and provides utility for people can't be botherd to learn more outside of how to press buttons) and trying to make data more valuable to people on an individual level. (instead of just for sigint agencies and their outsourced private equity owners portfolio companies like BAH).

Reply on news.ycombinator.com