Take a bunch of fibrous, cellulosic material, pound it into a pulp and then squeeze it into very thin, flexible sheets of material. Let them dry.
Then, take some form of pigment or dye, and with a very fine stylus impregnated with the dye, visually encode the data on the cellulosic sheets using a set of graphemes. Each grapheme would roughly represent a phoneme in a spoken language.
It would take quite a while to encode all that data. I'd suggest building some type of mechanical to automate the task of transferring dye on to the cellulosic sheets. I'd also want to bundle these individual cellulosic sheets into stacks of 200-500 for organization's sake. I'd probably cover them with a more durable material such as animal hide or perhaps a thicker layer of cellulosic material.
I'd then take all these bundles of data laded cellulosic material, and I'd build a structure to protect these bundles from the elements. Developing a cataloging or indexing system for these bundles shouldn't be too hard. I'm sure it's been done before.
Regardless, you could either preserve the materials or let the public have free access to the information. You'd run the risk of damaging the data, but if you had a mechanical replication system, you could simply make multiple copies of each data set, and ensure the safety of the data that way.
Sheets of fibrous, cellulosic material should last several thousand years if kept in the right environment.
You know. Now that I think about it. It's probably much too complex a system to handle something like that. I really don't think it would work.
Normal A-format paperbacks use acidic wood pulp for paper, but acid-free paper doesn't add a whole lot to the cost. So we get roughly 10Gb/ton, and the whole kaboodle comes in at roughly 500 tons. As the density of wood pulp is close to that of water, this approximates to a 10 x 10 x 5 metre lump. Quite a large time capsule, but not unmanagable :)
However. If we posit the availability of decent optical scanners in 50 years' time, there's no reason not to go denser.
We can print the data as a bitmap at 300dpi and be reasonably sure of retrieving it using a 2400dpi scanner. Let's approximate 300dpi to 254dpi, and call it 10 bits/mm. Printed on reasonable quality acid-free paper, we can get 100mbits/square metre, and that square metre will weigh around 80 grams (it's 80gsm paper -- same as your laser printer runs on). Call it 1.1 mb/gram. At this density, we can print the whole 5Tb (or 40tbits) on just 40 tons of paper, without compression; with compression, call it 20 tons. That's a 2.71 metre cube; just under 9' on an edge.
This assumes a monochrome bitmap. If we go to colour and use archival-quality inks we can probably -- conservatively -- retrieve one bit each of red/blue/green per dot, and it's probably not unreasonable to expect to get a whole byte out in place of each bit in our mono bitmap. So we can probably shrink our archive to roughly 2.5 tons of paper -- a pallet-load.
I've tested it out myself, and it's only after you start to crumple it together that it stops working. I tested it with an inkjet printer though. A laser printer may stand up better.
Eventually I want to do the same thing with a 20ft cargo container and a bunch of concrete.
no way that would ever work.
1. Get a 9600 bps modem. Use it to encode your data, and record the output as an audio file.
2. Take this audio file, and split it up into 60 minute segments.
3. Record these 60 minute segments onto two-sided vinyl LPs, 30 minutes per side. This will take about a million LPs.
4. Print on acid-free paper, using ink that will survive 50 years too, instructions on how a 9600 bps modem works. Describe the encoding in detail, sufficient so that someone using the equivalent of MATLAB or Mathematica or something 50 years from now on the computers they will have then could easily write a program to decode a modem signal.
5. Also print and include instructions for making a record player. As with the modem, the important part is describing how the signal is encoded on the LP. They'll have no trouble building a record player 50 years from now. (Assuming they don't just photograph the LPs with the 3D terapixel camera on their iPhone 54, and then write an app to extract the signal from the photo...)
5. Store all of this somewhere. LPs will last 50 years easily in a typical office environment, so you probably don't have to resort to something like a hermetically sealed vault buried in an old salt mine or anything extreme like that.
It is software which, when loaded into the Speccy's audio In port, does funny lightshows in time to the album's songs.
BTW, the LP speed doesn't matter, the Speccy picks it up anyway.
Pretty innovative. And the best thing is you can now get the LP in a TAP-style emulator format! So it survived over 25 years.
Removing that constraint and completely ignoring cost I'd also setup a low-risk savings account with $1M in it and put the data on S3 and Rackspace Cloud. I'd store access credentials in the capsule. Odds are pretty good one of those 2 will be around in 50 years (and you'll have a chunk of money left over in interest).
Try to keep everything ASCII, with really good text descriptions of data formats.
Realistically 50 years is not a long time: I would bet we'll still have legacy access to USB, SATA, and probably ext3 & NTFS (though probably not IDE). Tons of computer folk who used these technologies will still be alive to work them. English will still be the primary language in the US.
An interesting problem is what to do when the timescale allows these things to change. What if nobody remembers USB, or what spinning platters are. Or the English language?
Some people claim than MO media and DVD-RAM can guarantee 30 years, but this still is an estimate, they have not been around long enough to actually know.
The only "reliable" way to store digital data for more than five years known today is to copy them to new media well in advance of the old media loosing them, and even that is difficult if the amount of data is growing faster than the the storage technologies get faster. (I don't know if I should trust Eric Schmidt, but a few days ago he claimed that currently humanity generates as much data every two days as it did up until 2003, http://techcrunch.com/2010/08/04/schmidt-data )
If I was to store on magnetic media, I'd do it in a way that allows for some data loss (like usenet does with .par2 files). If you can stand to lose some of it, just pad it with enough redundancy for recovery and you'll be fine.
I think that this is luck; I found a batch of 8-year-old floppies a few years ago, and more than 2/3 of them were unreadable.
Edit: answered my own question. There are several kinds of toner (I had never heard of liquid toner), but some kinds are just fine:
"Toners composed of stable resin materials and a stable pigment such as carbon black are capable of strong bonding to the paper surface. Copies using these toners and printed onto permanent or archival quality paper can be considered permanent and suitable for long-term storage."
-- National Archives Australia
Spansion quotes 20 years minimum. Under optimum conditions (whatever they may be) and with an adequate amount of redundancy, 50 years should be achievable.
Spansion single-bit-per-cell floating-gate flash devices are designed to provide 20 years of data retention after initial programming when exposed to a 125° C environment.
Spansion two-bit-per-cell MirrorBit flash devices are designed to provide 20 years of data retention after initial programming when exposed to an 85° C environment.
Both MirrorBit flash and floating gate flash are guaranteed to provide 20 years of data retention after initial programming when exposed to a 55° C environment. MirrorBit flash is guaranteed to retain data for up to the minimum guaranteed cycles (10,000). [F]loating gate is guaranteed to retain data after the guaranteed minimum of 100,000
Have a child.
Name your child that string of letters.
Now preserving your data is the government's problem- they have to produce a birth certificate and keep track of him/her in their databases.
The thing about government is, all the laws apply to you. But they can do anything they want.
In other words, the reason we don't use (say) the MD5 hash system is not because computers are able to break the 32-bit hash system, but because people have discovered flaws in the MD5 algorithm that means it doesn't give '32 bits of strength'. In this case it's not the hardware (CPU clock speed) that gets better, but the software (programmes that break MD5) that gets better.
Hence, you could use Moore's Law to predict what computers in 50 years time will be like, but you can't know what mathematical techniques will be like in 10 yet alone 50 years. Your encryption system you use might get broken in 10 years time.
But even when it was published, people were saying a 56-bit key was too small, which they aren't saying of modern cryptosystems (to my knowledge).
For only 50 years, I'd probably risk making many thousands of DVDs and CDs, using different manufacturers and drives. Store with tons of redundancy and ECC, don't use inner / outer tracks for anything important, etc.
Also, are all of the data equally important? You can afford to store the more critical pieces in more expensive and less compact, but more robust formats.
I think the real enemy is obsolescence, and that keeping the data simple (and providing decompression programs and indices in easily understandable formats) is likely more important than worrying about bit-dropout, which seems largely manageable over your specified time.
For 500 years, I'd print it, or micro-inscribe it. (One problem with printed matter is that it has other inherent value, e.g., fuel for heating the yurts of cold barbarians).
For 5K years, micro-inscription and (if you are worried about technological crashes) an archive in the sky. You could populate a host of satellites in various orbits, timed to re-enter at intervals of (say) a decade over a few thousand years (hard to be exact with atmospheric drag and climate change, but you get my drift). Getting something from orbit down to the ground is not hard, getting /noticed/ and picked up as an interesting artifact is probably harder.
For 5M years, add a metric buttload of ECC and stick it in the DNA of some critter that doesn't get out much. A bottom-feeder in a radiation-shielded environment would be cool. Say, a lobster.
You would also need some mechanism to signal people in the far future that the lobsters were data carrying devices. Otherwise they wouldn't have any reason to randomly decode sea creatures. Perhaps you could program the lobsters to develop spots on their shells every century which denote the first 10 prime numbers.
The story for CDs seems to me to be similar and they are still popular everywhere. I give them at least another forty to fifty years.
As you say, you can use forward error correction to preserve the data. The hard part is describing the data format to the reader.
From the ATATA...intro, a reader knows they have a message. But now they need to know how to interpret it. You need a way of encoding information (english text?) in DNA and you also need a way of describing that encoding mechanism in DNA too...
Basically, over 5k years you should look up all the protocols SETI people have thought up. And/or re-read Gödel, Escher Bach.
Just bury 250TB worth of SSD storage, along with a device that activates every year and copies from one 5TB block to the next. Any single SSD will only be in use for a year. If the drives can survive 49 years before their first use, it will work.
Storing the data in some ridiculous format is just going to discourage anyone from ever reading it. I'm sure the people of tomorrow have better things to do than OCR millions of sheets of paper just to see grandpa's porn collection.
Sure, we still have the technical specifications for how to build it, but manufacturing the individual components would be a giant pain in the ass.
Regardless, I think it's likely that we will be better at reading 2010 media in 2060 than we are at reading 1960 media in 2010.
And currently, it's fairly easy to get data off of LPs, and they're actually somewhat popular for musical releases. There are devices which let you record LPs into computers, even. So perhaps not quite the best comparison.
(A minor nitpick: it being a SSD doesn't matter. The interface used to connect to it does. There is no difference in the hardware used to talk to a SSD or a HD, as long as they're both using a SATA connector. Note that I refer to the entire unit, what you get if you buy a HD or SSD, not to the actual internal systems that unit uses to access and store data. Those are obvious different, but the computer talking to it doesn't have to know about them.)
That's actually a pretty good thought - whatever it is, label it as porn.
People have spent millions of dollars restoring vintage erotica films.
Supposedly one holds 13,000 pages of text in human languages. If we assume your data is similar text, and one page is 58 lines of 66 characters (as are plain text IETF RFCs), you'll need:
(5TB / (3828 bytes)) / 13000 = 110473 disks
2. Put a decimal point in front of this long string. The result will be a rational number between 0 and 1. Call it x.
3. Get a titanium rod exactly 12 inches long.
4. Using a fine laser, etch a line in the rod precisely 12x inches from the end.
5. Done. Precise, durable, elegant, compact, and green.
Another problem is that any rod etched in this way will have two decodings. :-)
Of course, n bytes can store a number up to 2^(8n), not up to n. Thus, the number we'd be recording has order of magnitude (50 terabytes/bit)×log/log ≈ 1.20e+14, so we'd need to distinguish that many digits after the decimal point.
Conversely, as bayes pointed out (http://news.ycombinator.com/item?id=1585498), if we have a resolution of 10 digits after the decimal point, then we can handle about 10*log/log ≈ 33 bits.
(EDIT: To be clear, it's the same person posting; I just can't stand to wait 154 minutes to correct the error.)
But in ours, where stuff is made of atoms, I can't see you positioning the mark on the rod any more precisely than the width of an atom, which I think is about 10 to the -10 meters. So I'm guessing you could only encode 30 or 40 bits, even with super-advanced etching and measuring equipment.
I also think things like NASA datasets, other govt agency datasets, etc, should be placed on torrents for anyone who wants to make a copy. Let the self-replicating nature of the internet serve as the backup backup plan.
If you put those Apollo datasets online, it's a guaranteed certainty that some hacker somewhere will have them in 50 years.
If so, you could maybe give use some more information on the constraints involved(although i must admit thinking about it without any constraints is fun, too)
The constraints were chosen in order to remove the easiest answers (file sizes, period of time, etc).
Ultimately I think it's an unsolved problem that will become more important over time. My family can has photo albums from over 50 years ago but that doesn't have the kind of bandwidth we need for larger datasets (audio, video, etc).
So I guess it's just a thought experiment I thought was interesting.
Ideally some kind of artificial intelligence would come about sometime in the future to assume the role of data keeper - hiring people to do any work it couldn't do from within the computer and running off some kind of fund that had been set up. Maybe one day there will be a market for creating intelligent services like this, I hope I have something to do with them.
For example, even just sending a laser to the moon and back is quite sophisticated.
"Laser beams are used because they remain tightly focused for large distances. Nevertheless, there is enough dispersion of the beam that it is about 7 kilometers in diameter when it reaches the Moon and 20 kilometers in diameter when it returns to Earth. Because of this very weak signal, observations are made for several hours at a time. By averaging the signal for this period, the distance to the Moon can be measured to an accuracy of about 3 centimeters (the average distance from the Earth to the Moon is about 385,000 kilometers)."
Thus, even if you could get a mirror 25 light years away, the light wouldn't hit it or bounce back.
use 1200 dpi printer to print b/w dots this gives:
8.3 * 11.7 * 1200 * 1200 bits/page = 17 MB/page
5 TB are 142 Books with 1000 pages double sided each (size of a small personal library)
A4 page, which has 210x297mm, say a border of 10mm all around non-printable area, gives 190x277.
Say we print at a resolution of 0.5mm, and in 16 shades, so we get 4x4 = 16 bits per mm, or 2 bytes. That gives 105,260 bytes per page. Probably we can squeeze more bytes in than this, but let's allocate those to redundancy and error correcting codes.
So for 5TB (in hard drive size, 5e12) would take about 23.75 million sheets of paper printed on both sides. At 5g per sheet, that's about 119 metric tons.
1 ream (500 pages) of the 75g/m^2 A4 paper I have beside my printer here is about 50mm thick. Say 215x305x55mm including some slack for packaging, is a little over 3.6 litres in volume; total volume for 23.75 million sheets is 171,285 litres, or 171.285 cubic metres.
A room with a ceiling height of 3 metres would need to be only 8 meters on each side to store this. Of course, the room shouldn't have any windows and should be at the appropriate humidity, etc.
The cost of the paper, assuming $5 per 500 pages, is less than a quarter of a million. Much more forbidding is the labour and temporary acquisition of printers required to transcribe the data to paper. A good printer @50ppm would take nearly 2 years, assuming zero downtime and very quick paper and toner changes. To do it in more reasonable time and with more reliability, you'd want a bunch more printers; and of course, you'd need to hire the people to do the work of shuffling paper and carting it around, but I'd bet you could probably do it for less than the cost of the paper, particularly if you did it in a cheap labour country.
size of one pixel: 0.5mm * 0.5mm= 0.25 mm2
1 page can then hold: 52630/0.25 = 210520 pixel
1 pixel is 2 byte thus 1 page is 421040 byte is about 0.5 MB/page (or 1 MB if you print double sided)
5 TB= 5000 GB = 5,000,000 MB = 5 Million pages
this is 5000 books with 1000 pages each.
I intentionally chose a pretty big pixel for redundancy reasons rather than trying to be clever and working out a code etc., but the link in the reply to my post looks more worked out.
Start by looking at these guys:
I've heard that the LDS church / Vatican have both been interested in the archival media, and they have a pretty good long perspective, so might be worth checking with technologists in that realm.
Even if it didn't, if you compare the costs of many other storage techniques, they're probably equivalent to jerry-rigging a CD player to read back these disks. The difference is the cost is shifted to the reading and not the storage.
Your main risk is the shonky assumptions of the "archival" CD manufacturer. Not that I know what those shonky assumptions are, but I have a vivid memory of the hosts of Tomorrow's World demonstrating the durability of CDs by spreading jam on one, then wiping it off.
Also I'd claim that your 286 has higher build quality and it's less sensitive than modern computers.
Another proven way to communicating your knowledge through thousands of years is to start your own ethno-religious group/nation, like for example Jews.
If you want to combine both approaches - try Scientology ;)
Humanity is changing much slower than you may think. The big leap was electricity, and that is well behind us. I doubt the next big leap (nanotech) will become widespread for a hundred or more years.
2. Nanotech is much closer, than you think!
Toshiba is already working on 16nm semiconductor process. The transistors will be of the size of atoms.
Intel Nehalem is already 32nm process, so 16nm is two tic-toc cycles aways.
32 nm — 2010
22 nm — approx. 2011
16 nm — approx. 2013
11 nm — approx. 2015
You will need nanorobots to build chips of the NEAR future.
My startup in the nanotech area, so I know what I talking about.
A modern laser film recorder is capable of a resolution of 4096x3112 and 10 bits per pixel, so that's about 16MB of data per 35mm frame with black and white film.
After 50 years you can OCR the data etc (or ask your personal robot to do it for you) and print it using a variant of TeX/LaTex. This has already survived for 34 years, so another fifty years is almost guaranteed;) Knuth predicted some years back that TeX will last for about 100 years.
Mylar "paper" tape, or use some other plastic that's known to have serious archival qualities.
Bulky, but if stable (enough) in the presence of water it'll survive various failure modes that would kill acid free paper. Of course you could etch Mylar or some other stable plastic to gain greater data densities like with the suggestions for paper. Just pick a plastic we know is seriously stable from actual experience, like we know with acid-free paper.
We also have such experience with emulsion based storage methods (microfilm, fiche, etc.), but those are rather delicate for my taste.
At this rate, encoding a gigabyte requires 2,185 pages. As an aside, this is only 5 pages fewer than are contained in the "Art of Computer Programming" box set.
We can comfortably fit a gigabyte, then, on printed paper, in a 10"x12"x5" box. A terabyte will then fit comfortably in a 10'x10'x5' space. Throw a few of these together to get 5 terabytes. Let's add, say, 1 TB more of error correcting codes. In the unused margins of each page add in some information about alignment, a printing of all the colors used (to try to protect against inks changing color over time) and the page number. All together, this is certainly big, but could probably fit in, say, a tractor-trailer. Throw in some books describing the data format and the meaning of the data, and you're done.
The nice thing about this solution is that there's a lot of existing knowledge about how to build and design optical media, how to build in parity checking, detect jitter when reading, etc. This could be relatively cheap to do in bulk also, if you found the right material (I just mentioned gold because I know it to be chemically stable, but there are probably better materials).
Alternatively burn a 100 or so Blu-rays and get two Blu-ray readers (one on a mobile device) other an external reader that you will attach to the aforementioned desktop.
Or who knows, maybe holographic storage (http://en.wikipedia.org/wiki/Holographic_data_storage) will come around finally and store the 5TB in a toothpick sized gizmo (which might probably run 512 cores of Googple's Chip (in my alternate reality, Google and Apple merge and buy Intel)...
A quick Google of HDD MTBF suggests that 1 million hours (over a century) is wildly optimistic, and typical failure rate is 2-4% per year, possibly as high as 13%. If e is the failure rate (as a fraction of 1), and assuming a constant and independent failure rate over time, then the survival chances for any one drive are:
(1 - e)^50
1 - (1 - e)^50
f = (1 - (1 - e)^50)^n
log(f) = n log(1 - (1 - e)^50)
n = log(f) / log(1 - (1 - e)^50)
I don't know if the bits eventually lose their magnetism over time, so if they do, you may need to spin up the drives every so often and copy to and from drives to make sure the data is still "fresh", but I seriously doubt they'd need to be left on and spinning for the entire 50 year span.
Since the basics of logic will not change and since theoretically any computer can simulate another. Why shouldn't we just keep a hackable computer with detailed visual instructions and specifications? Further, to enable someone to read the specifications we could have a "learning board" with the symbol and the component next to it. Also, we could even have a haptic output with which people can interact with the computer.
Let's assume that we have a nuclear battery made out of Technetium. Now this feeds into a bank of capacitors and high performance rechargeable cells. Slowly over time the batteries are kept topped off and they are "exercised" by the computer. Further, for redundancy 4 or more computing units in parallel could be placed that would wake up sequentially and call the others to check how the entire unit is working. If we keep something like this in a hermetically sealed environment and we use the radiation source to manage the temperature and use passive cooling technologies for letting out heat. It should be able to sit still until someone finds it.
Now the data itself would be stored on a series of solid state devices [edit: a specially designed optical storage medium would be far better, but this is 50 years not 1050.] attached to a display. Why shouldn't this suffice?
Presuming that civilization has not collapsed anyone should be able to read it.
By the way, the Phoenix lander has a DVD that tries to do just this (see: http://en.wikipedia.org/wiki/Phoenix_(spacecraft)#Phoenix_DV...). It even has this awesome intro by Carl Sagan. (hear: http://www.planetary.org/special/vomgreetings/greet2/SAGAN.h...)
[edit: Dyslexic errors.]
The human brain has superior data storage capacity and resilience to any of the mentioned media, and a longer functional lifespan. It can also adapt on the fly to changes in technology and language.
This also makes the data proof against illiteracy and technological collapse, assuming a cultural impetus to memorize the data and teach it to others can be maintained.
If 1MB is too much, try 1KB, and repeat for 5 billion people. There should still be some people left for systems maintenance.
1. Create some type of mechanical which is able to periodically do these tasks:
- create an hard disk (named n)
- plug it to the computer above
- transfer the data from hard disk n-1 to hard disk n
- check the hashes
- unplug hard disk n-1
- reboot the system using the new hard disk
vacuum seal everything, with a packet of silica.. no air at all, moisture removed.. part by part.
1) then vacuum seal the container
2) pack it with closed cell insulation
no light, no air, no corrosion or UV damage.
advantage to #2 is that it would be ruggedized for hits & transport.
For preservation, encode the data in the DNA of a bacterium, replicate it massively and put them in some kind of suspended animation (hand waving wildly).
Getting the meaning of the data is another matter. I wonder whether it will be possible to create species of bacteria that can decode the above DNA and present it directly accessible to human senses - ex: bacteria that change color, form shapes, etc.
Easy isn't it? ;)
How would you store an ever-increasing amount of digital data indefinitely?
 Charles Stross describes it: "Memory diamond is quite simple: at any given position in the rigid carbon lattice, a carbon-12 followed by a carbon-13 means zero, and a carbon-13 followed by a carbon-12 means one. To rewrite a zero to a one, you swap the positions of the two atoms, and vice versa." See http://www.antipope.org/charlie/blog-static/2007/05/shaping_...
I don't think 'use rackspace /and/ use amazon' is a good strategy either, just 'cause if market conditions change enough that one goes under it's likely the other will to... and it's also likely that they could be bought by the same entity. you'd want to put one copy on s3, and then use some completely different storage method, like microfiche or or archival CD roms or something in a safety deposit box.
In a few years we'll have those with 64gb/128gb capacity.
design a few of them with the data just burned in. no capacitors to fail, you could get your redundancy by just making a couple dozen of each. design them to wait a few seconds after power on, and then start spitting out the data and a clock. label the power, ground, serial out and data pins.
given those pins, there isn't anything else you could do with it besides hook it up and see what it has to say. and all you need to read it is a stable power supply, and sufficiently fast analog to digital converters.
Optical media is the only way to guarantee whichever future "creatures" encounter the data, can actually figure out how to access it.
If you are a blind alien (so that you do not understand the _concept of light_) you can potentially still have measuring equipment that can sense the pits in the media and make sense out of the binary data.
(note: self.race == asian)
goes back to playing 30+ year old Atari games
The New York Times used them to make a nickel disk of their archives for a very long-lived time capsule.
Assuming there is no cost limit here, I would go one step further and say use some form of metal. Say stainless steel, aluminum, gold, or titanium. Some metal that is very stable over time and does not interact with the atmosphere readily. Again, use micro-abrasion / carving technology to write data to the materials.
The next question is what format for the data? It depends on whats being stored. The biggest issue is that of "formats".
Lets look at things that last a long time. The English(or any language) is unlikely to change that much in 50, 100, or even 200 years. Words and their meanings will change, but for the most part a native English speaker 200 years from now could read what we write now. Whether or not they understand the usage of the language is a different question. So if its a textual document you're saving, write it in plain English. No abbreviations, etc.
What about media? That gets complicated. If its a static image, perhaps keeping it simple is best. In plain English, write that the following section of data is an image. Each group of three numbers starts at 0 through 255. In procession from left-to-right, the values represent Red, Green, and Blue. Each group of 3 numbers is what we called a "pixel". The image is 300 pixels wide, and 800 pixels heigh(arbitrary numbers for this argument).
For moving images, further expand on the single image description and say every 24 images should be spaced equally and viewed over the course of 1 second to achieve animation.
Sound is something I don't know anything about from a data format perspective, but I would again find the simplest mechanical way to produce a sound and store it in that format with ample verbage describing how to handle it.
After reading other responses that came in while I was writing this, I want to add some more thoughts.
Remember that our technology is ephemeral. We don't really use much tech from 50 years ago, hardly any from 100 years ago, and it just gets worse from there.
Things like microfiche, ssd's, cd-roms, blue-ray, etc are all the more bad ideas for long-term backup. Paper books are a better option than any of these for near-term storage for time periods up to 50 years.
If we want to actually store data in a meaningful way for long periods of time, say over 100 years, we have to keep it simple. Your devices will probably not last 100 years, even if kept in storage under the most secure environment. But in 100 years people will still have eyes, ears, and hands.
We have to look back over history and look at the material types that survive long periods. Stone, and metal to an extent, are very good long-term materials. Cloth and paper are not*. DNA is potentially a usable data store, but is corruptible. Plus you can't read DNA patterns with the eye.
Printing presses, microfiche, film, radio, telephones, television, computers, light bulbs, electrical sockets, toasters, cars, bicycles, electric stoves, speakers, microphones, projectors, etc.
Technology is not nearly as fickle as you'd think. Remember, 50 years ago is only 1960. VHS tape technology is almost 40 years old. Sure, things are moving faster now, but things that work tend to stick around.
But as you say, stone and metal as well as acid-free paper made from rags or mulberry fibres have a proven track record for longevity.
Actually that alone is no proof that clay tablets are very durable. Who knows, there might have been billions of them in circulation, and only a few of them survived. That would be a rather bad track record.
I'd include the definition of a simple machine, and the text of a program written in opcodes for the machine, for decoding.