Modeling the future

Tom Craver has an interesting idea: a Wiki model of the future.

Such models have equations and parameters, and (for example) simulate what will happen to the economy in a year or two. The Club of Rome built such a model, and their model has been very influential.

So, is it possible to build an open-source model? One in which anyone can contribute a bit of insight or knowledge, and the system processes all the bits until it comes out with a useful set of projections about the future?

CRN has engaged in online collaborative projects before, including our groundbreaking scenario project. And a model of the future would certainly be useful to us.

Of course, it wouldn't be just one model. A useful system would run lots of projections using different variants of the model, and then aggregate the projections, looking for patterns in the relation between assumptions and outcomes. This would take a lot of computer power, and could be done on an @home type platform.

The information input into the scenario process would benefit from semi-automated Delphi-like feedback. Different versions of a piece of data would be rated according to how they matched opinions from other users, as well as what effect they had on the model.

Does anyone here know enough about mathematical modeling of the future to say whether this idea might be worth implementing?

Chris Phoenix

Comments

[todd andersen]

sorry i do not, i would like to say welcome back and i look forword to reading your posts chris.

[Chris Phoenix]

Thanks Todd!

Chris

[Michael Anissimov]

I'm involved with two such attempts, neither of which is done yet, but I didn't handle the math end of things, just the writing end. About a third of one model (originally began as an SIAI Summer 2007 Research Project) is overviewed here:

http://www.acceleratingfuture.com/michael/blog/2009/02/the-uncertain-future-simple-ai-self-improvement-models/

The math side of the project is still working on that algorithms involved, but the basic outline is that the model is made up of a bunch of probabilistic nodes that produce values different than the input by running them through two or more nodes. Hopefully this project will be online within a few months, and you can see what we mean.

If you want to talk more about the math end of this, ask Anna Salamon. You can get her email from M. Vassar.

Another project is a work in progress by Peer Infinity, who is creating a wiki on my website. The overview can be found here:

http://www.acceleratingfuture.com/wiki/index.php?title=The_Scenarios_Project

I think it needs a lot of work, and a new approach to presenting the info, but it has some potential. Unlike the SIAI project, there is nothing dynamic or algorithmic planned to happen in the immediate future with this wiki scenarios project, but I understand it's a longer-term goal.

[Michael Anissimov]

Sorry, I meant a Summer 2008 Research Project.

[Dan S]

Let me present arguments against building such a model. While I agree that it would be useful to have one, I think that construction of such model is orthogonal to CRN goals and, very likely, impossible.
1. Economic systems are chaotic by their nature, and predictions rapidly diverge from reality in a few timesteps. Current economic crisis demonstrates that very well. It is even possible that future of economy can not be computed by a less complex system. Really, forecasting in economic and social systems is a big scientific problem without satisfactory solution available.
2. Constructing global economy model from the scratch is a very ambitious project. Years of work, lots of effort, low probability of success. Such projects better be left for expert economists.
Instead of diving into arcane science of economic forecasting, I suggest that CRN should start open source project to develop safe molecular manufacturing system. This is no more complex (more likely much simpler) than predictive economic model and suits CRN purpose much better. If properly designed, such project will raise public awareness of MNT potential, speed up MNT development and ensure maximum safety of system design.

[Tom Craver]

Dan - thanks for presenting the other side of the argument.

I'm not 100% convinced this is right for CRN either - but I think it's worth brainstorming on, a bit.

Feasibility - I wouldn't use the current economic crisis as an example of unpredictability. It was predictable - several smart investors predicted it and made a killing. They didn't have to predict the whole economy - just that a major component of it was going to seize up.

Clearly it is very ambitious - which is why I'd like to think some more about how it could be modelled after wiki's.

Even if it were limited to a small group, I think something interesting could be done, though it'd be much less specific.

Frankly, there is such a confluence of major, civilization-impacting factors over the next few decades, that it feels like we're flying blind into a storm. We know these things are coming - we just haven't got any idea how they're going to interact. The stock market collapse and on-going recession may just be the first gentle turbulence, hinting at what is to come IF we don't start steering the plane better.

[Chris Phoenix]

Tom - lots of smart investors didn't predict the current crunch. For example, Stanford's endowment is down tens of percent. I think we have to call those who did predict it "lucky," or at least not attribute their prediction to their smarts.

Dan - even though you won't get "the right" prediction from a chaotic system, you can run many simulations with different starting parameters and get a range of predictions. That can be very useful.

I agree that it's not a core CRN mission, but if it makes the world a better place, helps CRN's mission, and doesn't require a lot of CRN resources, I'm not ruling it out.

As to the open source MM development project - that probably needs a lab, which we don't have. If anyone cares to suggest under-explored development pathways, I will be happy to evaluate them, publicly discuss my evaluation, and pass on the good ones to the people with labs.

Chris

[Tom Craver]

Maybe just list "major change" trends that look to have major impact on their own, and estimate their phases as if they weren't interacting? Simply charting those together might start to stimulate thought about interactions.

Banking/Credit train-wreck
Peak Oil
Peak Gas
solar power advances
wind power advances
nuclear power advances
Nuclear proliferation
Climate change
Red vs Blue cultural divide
religious reaction against secularism
unconventional warfare
super-empowered individuals
internet crime, worms
panopticon/transparency
New diseases
Species extinction
Baby Boom Retirement
population growth / shrink
immigration
refugees
water shortages
Moores Law
Robotics/AI improvements
Bio-medical advances (repairs, cures)
Bio-tech (synthetic life, tweaked organisms)
mind leverage effect of the Internet
nano-materials advances
micro and nano machines

[Tom Craver]

Chris:

Read this story about John Paulson (and others) - I don't think it'd be fair to say Paulson was just "lucky".

He may have been lucky to spot hints back in 2005 of how crazy the banks had become - but he also invested a lot of time and effort into figuring it just how crazy they were - and how to profit from their eventual collapse. (No one loves a vulture, but they do serve a purpose in the natural order :-) )

Perhaps if there'd been a simiki-Econ, the wwweb-mind would have spotted the same problem in time to spread the news and avoid the collapse.

[Chris Phoenix]

How many people are there, equally as smart as Paulson, who took different bets and lost? If there are 4 Paulsons in the world, and 3 of them lost money, then Paulson was lucky.

Paulson certainly had a plan to get his money. His plan worked. If he had had no plan, he wouldn't have gotten the money. That makes him look non-lucky. He was dedicated and gutsy.

But how many other short-sellers failed to make money by shorting the wrong things at the wrong time? And how many smart people whose job it was to see the collapse coming didn't see it at all?

Chris

[Loki]

I had a similar idea a few years ago. My idea was to create a graph of "claims" and "events". There could be contradictory claims.

A sample event is "Dow Jones goes below 5000". A sample claim is "If 1e15 ops/sec computing is available, a human brain can be simulated in real time within 5 years" or "if the Dow Jones goes below 5000, Moore's law will slow by 30-40%". Claims would have probabilities or confidence intervals.

All this would be entered in machine readable form for later execution.

You could then set up models and run scenarios by selecting a subset of events and claims. The scenarios would then be run programmatically.

You don't necessarily have to predict the future. You can look at scenarios and be prepared to react.

[Chris Phoenix]

That sounds kinda' like an expert system, where my impression is that most future modeling uses equations that are assumed to always be in effect, like modeling the weather.

I like your idea. It sounds good for generating a wider range of scenarios - makes explicit that we don't know what options will happen.

Chris

[Dan S]

Actually, many prediction systems for use neural networks or similar technology. Their results can sometimes be better than that from equation-based models, especially if system in question is poorly understood