Exabytes

New York Times Piece

Some more very exciting news occurred while I was away on vacation up north temporarily enjoying being off the grid. Recently, Nick Bostrom's Simulation Argument was discussed in a New York Times article by John Tierney. My historical simulations paper was mentioned in the NYT article and in Tierney's accompanying NYT blog here and also here. Tierney's view is that the chance of our being in a simulation is about 50/50 which is more sanguine than Bostrom's estimation of 20%. Also, there was an item on MSNBC. 

Simulation Argument

A couple of days ago I received a very pleasant surprise in the form of an e-mail from Nick Bostrom of the Future of Humanity Institute at the University of Oxford informing me that he has added a link to my historical simulations paper on the scholarly papers portion of his Simulation Argument site. This is very encouraging and comes at a very opportune time as I am gearing up for some research in the area of prediction markets, a topic in which Nick's FHI colleague, Robin Hanson, is a pioneer. More news on this coming soon!

Ludium 2 Wrap-up

I went to the Ludium 2 conference in Bloomington, Indiana last weekend. It was based on a game of a 19th century-style political convention with the object of coming up with a platform of proposals for the virtual world industry and electing a Speaker representative who would put forward the platform to political candidates in the real world. The final platform is here. As you can see, some of it seems to be based on the preservation of the status quo, e.g. freedom of expression for designers, internet neutrality, shielding of game developers from liability for actions of players, fair use, and there is also the predictable grab for research money, but there is some cutting edge stuff too, such as the player's bill of rights, (free speech, right to assemble, to organize), that I fully supported of course. The convention game machinations themselves were absorbing although at times distracting and a bit frustrating. There was an attempt to merge two voting regions (Palvi and Sysland) and then announce it as a fait accompli without first making a motion on the floor to approve the merger. This stuck in my craw, especially when Ted Castronova, the gamemaster, said from the podium that it would not affect the final platform. Well, in mergers like this, there is a good chance that it will end up in an oppression of the minority by the majority and also limit the debate due to enforced homogeneity of opinion within the merged group. I raised this point and it was put to a vote on the floor and the merger was narrowly defeated, vindicating my opinion. For this I was rewarded with a Twitter. It turns out that the merger was part of plot to ensure the election of an academic (Thomas Malaby of U. of Wisconsin) as the Speaker rep rather than an industry type (Corey Bridges of Multiverse). Apparently, there was a perception that industry persons are being given a hard time now in the U.S. as spokespersons for interest groups. Personally, I liked Corey better and voted for him since the class clown who just obtained $4 million in venture financing for Multiverse appealed to me (it had nothing to do with the several hours spent with him and Jerry Paffendorf et al drinking at Nick's Bar which was after the event!). However, Malaby was ultimately elected. Best of luck to him during his term of office. The District I was in (Palvi 2 which was composed of three lawyers) won a prize for getting some of its recommendations into the final platform.

On the whole, Ted Castronova put on a productive and fun conference, but I have two suggestions for the next Ludium (2009?): 1) abandon the convention format and instead determine issues through a prediction market, which should be less subject to rule bending and 2) drop the U.S. centric approach and focus on the issues from an international perspective. I'm not advocating cyber-separatism or pie in the sky like a seat on the U.N. for virtual worlds, but ultimately there should be an international treaty on virtual world issues, just as there is for intellectual property.

Stack or Snack?

Check out this video. Apparently, the stacking problem can be solved by snacking! BTW, if Second Life has a million residents, then Third Life should have a trillion and Fourth Life a quintillion etc. Good viral marketing strategy.

Information Plague

Here's a frightening policy paper from IBM estimating that by the year 2010 the amount of the world's digital information will double every 11 hours! This means that it would double an astounding 796 times in 1 year. According to my calculations, we would never reach the end of the year. Conservatively assuming that we start in 2010 with a total amount of digital information equal to about 100 exabytes, then about 110 days into the year, (April 20, 2010), we would reach the maximum computational storage capacity of the entire universe which is 10⁹² according to Seth Lloyd (unless large scale quantum computing is operationalized by that time). 

Get a First Life

Here's the lighter side of historical simulations, and the review. Enjoy! Did you ever  wonder, though, when it was that you clicked on the End-User Licensing Agreement?

Milgram Redux

In my recent paper on Historical Simulations, I noted that it would be possible for a future society around 2050 to run completely realistic simulations designed to test AI for ethical flaws. They could overcome the ethical problems associated with the Milgram Experiment by obtaining informed consent from the AI and then doing the appropriate memory wipe of the consent process.

Now we see that first step has been taken along the road towards this scenario. Researchers at University College in London have re-created the Milgram Experiment in virtual reality by requesting participants to administer electrical "shocks" to a simulated "subject". These researchers claim that the participants reacted as if the experiment were as real as the original 1961 experiment, despite the fact that they clearly knew that the individual to whom they were administering the "shocks" was merely virtual.

I expect that the results of this recent attempt to re-create the Milgram experiment will eventually be seen as somewhat feeble, given that the participants were obviously aware that they were just giving simulated shocks, which eliminates the deception necessary for the experiment to carry any weight.  If someone is ordered to toss a Barbie doll over the edge of a cliff, it doesn't mean that they would likely do the same if ordered to do so to a live human being. Eventually, this dissatisfaction with the virtual Milgram Experiment will lead to impetus to conduct it, or something similar to it, with AI in a completely realistic simulation in which the AI is not aware that it is actually in a simulation. To avoid the ethical problems associated with the lack of informed consent, the appropriate memory wipe of the consent process would be used, as I outlined in my paper. 

Copybot Goo

The copybot problem in Second Life raises the issue of what happens when these programs, currently used to copy avatars and digital paraphernalia such as virtual cars, are eventually extended to copy entire virtual worlds. It certainly highlights the extent to which simulated worlds could rapidly reach a stage where they outnumber the real one by a factor of billions or more to one. Stacked simulations (sims within sims) could, as I pointed out in my paper, then quickly exceed the entire amount of particles available for computation in the entire universe, thus ultimately destroying the real world in a copybot goo scenario where the entire universe is converted into computing substrate, e.g. computronium. Where this would end is the crashing of the entire stack of simulations, leaving a real world composed of nothing but inoperable hardware, on which nothing runs. Preventing stacking in the first place, by programming a suitable terminal date into the first set of simulations, could avoid this copybot goo scenario.

Question: is a simulation just a copybot's way of making more copybots?

Historical Simulations

   

My new paper on historical simulations has finally appeared in print in the August issue of the Journal of Futures Studies. George Dvorsky has posted some comments about it on Sentient Developments, where it spawned an interesting little thread on the stacked simulation issue, (simulations within simulations) involving Robert Bradbury and others. George quoted an earlier passage (written prior to my paper) from Bradbury stating that simulated realities will run slower than the host reality, therefore allowing at least "trillions of trillions of trillions" of simulations and therefore the universe is unlikely to run out of computing power due to stacked simulations. 

This quote from Bradbury also indicated that if we live in a simulation, then we cannot extrapolate the size of the host or "basement" universe from examining the size of the simulated universe that we inhabit, since future societies might simulate "weird" universes with different physical laws and dimensions than the real one they occupy. George thus concludes that it is arbitrary and unfounded to conclude that our simulation will terminate around 2050 when we start on a widespread basis to create historical simulations of our own. There are several problems with George's concerns:

1) He ignores the point in the paper that there is another reason for the simulation to be unplugged when it reaches 2050, i.e. that a simulation depicting a period after that date would lose the element of deception of its inhabitants necessary to produce genuine behavior for observational and experimental purposes. Robert Bradbury's comments in the thread discussing my paper even concede that it is possible that we could be in a simulation designed "to study evolutionary paths of advanced civilizations up to the point where they begin to simulate the evolutionary paths of advanced civilizations" in which case we may be indefinitely "suspended" (nice euphemism for "terminated") at that point.

2) George's quote states that due to the simulations running much more slowly than the host reality, the universe would be capable of at least "trillions of trillions of trillions of simulations". However, this is merely 10³⁶ simulations, which still doesn't come anywhere close to the 10⁹⁶simulations that would be produced after only 16 generations of stacking starting with an initial one million simulations.

World of Warcraft currently has at least 6 million subscribers, so it is easy to project a starting point of at least 1 million fully realistic simulated worlds by 2050. In any event, Robert concedes in his comments that if the hardware is intentionally engineered to facilitate simulations, (e.g. a nanotechnology or femtotechnology graphics card!) it will run at close to the speeds of the reality being simulated. Presumably if a future society wants to create simulations, then it will dedicate special technology to the task and won't use a generic form of hardware. Robert also notes in his comments that the simulation could develop abstractions (shortcuts or heuristics) that would enable computational speedups which exceed the reduction in speed inherent in the simulation.

3) In his comments on the paper, Robert also notes that there is the issue of whether the abstractions or shortcuts developed in the simulation can be translated back into the reality that produced the simulation. If the rules in the simulation are vastly different, (e.g. the "weird universe"), then extracting the inventions of the simulation may be impossible. Robert then wonders whether the "gods" would intentionally design simulations from which extraction is difficult or impossible. I think that the answer to this question must be in the negative. If a future society goes to the trouble of creating a historical simulation, it must be for the purpose of extracting information from it. If we are in a simulation now (which we probably are), then it is likely not a depiction of a "weird" universe that is fundamentally different from that of the host reality. We can therefore extrapolate from observing the simulated universe that we inhabit what the total computational power of the host universe is likely to be.

I am looking forward to further comments. The stacked simulation problem, which was only a small part of my Historical Simulations paper, could easily justify a paper of its own.

Second Cambrian Explosion?

Consider that Andrew Parker's 2003 book, "In the Blink of an Eye" set out the brilliant hypothesis that the "Cambrian Explosion", which commenced about 544 million years ago and involved the dramatic increase of the number of animal phyla from 3 to 38 in the relatively short period of  6 million years, was due to the development of vision in primitive animals. The exponential increase in phyla was the result of the need for new defense systems in animals being chased by vision-equipped predators such as early trilobites. Put this together with new developments in the area of brain-reading technology which can already enable persons to move computer cursors using just their thoughts, and eventually will lead to "network enabled telepathy" where Internet communication goes directly from brain to brain. What could be the result? A Second Cambrian Explosion where the number of phyla will again increase exponentially, as humans evolve into new forms to defend themselves against the threat of those empowered by the radical new sense of perception.