Naked Science Forum
Non Life Sciences => Technology => Topic started by: Airthumbs on 28/10/2012 21:13:03
-
If a computer was given all the variables, materials and technologies currently available and then programmed to provide the most efficient way of getting into space, would it provide solutions?
-
Yes & No
While the early space pioneers undoubtedly were slide rule wizards, a computer can help optimize parameters, and compare scenarios. Perhaps one would use the computers to optimize shapes of structural components based on predicted acceleration and stress.
What a computer can't do well is "think outside the box".
Would a computer have been able to determine that the paint on a space shuttle tank weighs about 600 lbs, and could be left off?
One should be able to probe various scenarios with a computer.... for example asking the fuel and weight required to achieve 100 MPH vertical velocity. Or, perhaps the cost in wind resistance to achieve 25 miles altitude. Or, the fuel required to reach Mach 3 horizontal velocity at 80,000 ft (24 km). Then one can look for alternative methods to achieve those parameters.
It is my belief that we could reduce fuel cost significantly by building a supersonic super-transporter launch platform. But, the cost of engineering such a launch platform may in fact be prohibitively expensive.
A mountainside mag launcher? If only there were mountains in Florida.
We did some brainstorming of various launch methods here:
http://www.thenakedscientists.com/forum/index.php?topic=45669.0
Computers should augment the comparisons between different launch methods... but perhaps not think up new alternatives. Could the computer estimate the cost associated with de-novo development of a new technology?
-
Very interesting Clifford K, check out this link to WATSON http://en.wikipedia.org/wiki/Watson_(computer) (http://en.wikipedia.org/wiki/Watson_(computer))
And this http://en.wikipedia.org/wiki/Blue_Gene
Mod: fixed broken link
-
A computer is only as good as its program.
Watson was optimized to use web searches for answering trivia questions. In a sense it is based on hypothesis generating and testing which is similar to a scientific search. However, its whole internal hypothesis testing is based on an internet search and evaluation.
There are, of course, a lot of untested ideas on the internet. So, perhaps one could simply write a program to hunt down ideas on the internet, and then evaluate and test the ideas for feasibility. Or, have a "suggestion box" where people can write in ideas, then the computer can do number crunching to evaluate the ideas.
As mentioned above, simply not painting the space shuttle's fuel tank saved 600 lbs or so. A simple idea, but a big weight savings. Exotic lithium metals also saved some weight. These are kind of big leaps for a computer to come up with on its own.
Again, I think the biggest issues are not the lack of innovation, but rather the cost of implementation, and perhaps other factors.
Consider there are no major mountains in Florida.
But, Wheeler Peak (http://en.wikipedia.org/wiki/Wheeler_Peak_%28New_Mexico%29) in New Mexico is at 13,161 ft (4,011 m). Not that high, but launching from that peak (or putting on a magnetic launcher on the mountainside) might have significant fuel savings (or corresponding payload gains). But, the downside is that a launch path would take the rocket across 6 or 7 states. A computer would be able to help calculate the impact from different launch latitudes.
Perhaps one should consider Ecuador or Peru. For example Huascarán Mountain (http://en.wikipedia.org/wiki/Huascar%C3%A1n) in Peru, or Chimborazo Mountain (http://en.wikipedia.org/wiki/Chimborazo_%28volcano%29) in Ecuador as potential launch points. However, again this has the problem of launching over land. And, any pure "number crunching" would ignore the political, social, and economic effects of moving equipment and launches to South America. And, of course, there are seismic issues, as well as issues with snow.
As mentioned, supersonic launches would save fuel, but potentially at other costs such as the cost of developing a supersonic launch platform.
Anyway, there are a lot of decisions that go beyond pure number crunching of a per-launch cost. Some things could be added to the evaluation algorithm. Some things might be difficult to add.
I suppose one could always have a computer designed to "brainstorm", then allow an investor to choose the ideas they feel most suited to their needs. Or, better yet, allow humans to brainstorm, and make the computer crunch the numbers.
-
There is only one thing stopping computers from being able to match (and surpass) human abilities in this regard, and that's that the software running in them is not yet at the level where it can tackle any problem in the way that people do. Within the next few years, the missing capabilities will be added and we will sidelined by machines which can do all the same kinds of thinking much faster, to enormously greater depth, and without errors. Even so, it would still take time for it to evolve its thoughts towards finding the most efficient way to get things into orbit, and much will depend on new scientific knowledge which will keep changing the known possibilities. Intelligent machines will of course be able to steer science in the right directions to try to acquire the most useful knowledge first, and a lot of the experiments will be carried out directly by machines, so the rate of progress will only be limited by the environmental costs of supporting this research and development.
-
Old joke alert.
A home-accident survey showed that 90 percent of accidents on staircases involved either the top or the bottom stair. This information was fed back into the computer to analyze how accidents could be reduced. The computer’s answer: ‘Remove the top and bottom stairs.'
Seriously, computers don't understand things so they can make errors like that.
-
Program the computer to evaluate various scenarios.
But a lot of care needs to go into providing good data.
Some equations will be easy, such as reducing the weight of the rocket by 1000 lbs will directly correspond to fuel savings and cargo increases.
It shouldn't be too hard to calculate benefits of more equatorial launches, or increasing a static launch altitude (other than the cost of building a new facility, and potential political ramifications).
Other options such as using magnetic ground launch, or using jet engines/winged lift for the initial stage would have far more factors that would need to be estimated and supplied.
Again, a good program, and a good data in may lead to useful data out. A poorly written program, or bad data in will likely lead to useless data out.
Keep the scenario design, and ultimate evaluation of the system in the hands of people.
-
I think i'm missing something here, to me it seems like there is quite a simple solution to a program that would have a primary goal of finding these solutions.
Specifically in regard to finding something that has been overlooked such as 600lbs of paint!
I see this program starting as a basic simulation. This simulation will have all the necessary information to calculate the velocity needed to send a simple spherical object into low earth orbit.
Each component of the program could be modular so that every single detail about every part of the object would be entered, it's mass and composition for example. This is where things like paint would not be overlooked.
So far I think most of what I suggest can be done relatively simply on a calculator although would take a lot longer!
The next step is to introduce variables, and keep introducing them until you have a huge selection of different data sets; materials, accelerators, power consumption, energy potential, geographical restrictions, cost, and finally given a strict set of rules to find the most efficient way of launching an object into space.
This set of rules would consist of parameters such as at what temperature something melts at or what kind of acceleration can a material resist without damage to its structure.
What I like about this kind of simulation is that it would constantly be evolving with more data and information.
I suspect if it was that simple someone would have done it by now though. :(
-
Old joke alert.
A home-accident survey showed that 90 percent of accidents on staircases involved either the top or the bottom stair. This information was fed back into the computer to analyze how accidents could be reduced. The computer’s answer: ‘Remove the top and bottom stairs.'
Seriously, computers don't understand things so they can make errors like that.
Tis a poor one :) Firstly if you remove the top and bottom step, you still have a top and bottom step. The actual answer would be to "remove stairs" and that's pure genius I think.
-
I think i'm missing something here, to me it seems like there is quite a simple solution to a program that would have a primary goal of finding these solutions.
Specifically in regard to finding something that has been overlooked such as 600lbs of paint!
I see this program starting as a basic simulation. This simulation will have all the necessary information to calculate the velocity needed to send a simple spherical object into low earth orbit.
Each component of the program could be modular so that every single detail about every part of the object would be entered, it's mass and composition for example. This is where things like paint would not be overlooked.
So far I think most of what I suggest can be done relatively simply on a calculator although would take a lot longer!
The next step is to introduce variables, and keep introducing them until you have a huge selection of different data sets; materials, accelerators, power consumption, energy potential, geographical restrictions, cost, and finally given a strict set of rules to find the most efficient way of launching an object into space.
This set of rules would consist of parameters such as at what temperature something melts at or what kind of acceleration can a material resist without damage to its structure.
What I like about this kind of simulation is that it would constantly be evolving with more data and information.
I suspect if it was that simple someone would have done it by now though. :(
With all the resources needed in organising and processing the data, not to mention the initial programming etc, surely just using all the people and resources to come up with a solution, rather than to come up with a computer program for the solution would be more efficient?
-
I think i'm missing something here, to me it seems like there is quite a simple solution to a program that would have a primary goal of finding these solutions.
Specifically in regard to finding something that has been overlooked such as 600lbs of paint!
If you tell the computer the components that you need to build a rocket...
Metal,
Supports
Paint,
Engine,
Controls,
etc...
The computer will religiously follow those guidelines. It really takes the engineer to go back and say... well maybe we don't need the paint for a one-time use object (yet, I still see many beautifully painted one-time use rockets, apparently the lesson was only applied to the fuel tanks).
Deep space missions are designed with a critical concern for watts and milliwatts as everything is related from the weight to longevity of the mission. Will the computer be able to choose the correct computer/cpu for the mission needs?
What about the use of gold vs aluminum wiring? Gold... expensive & heavy. Aluminum... light & cheap. The answer should be trivial... right?
-
I totally agree, telling a computer how to build a rocket would not provide a more efficient rocket.
Lets say we tell a computer a + b = c
a = the rocket.
b = every available alternative.
c = the most efficient way of getting into space.
?
Are you sure your not just scared of what would happen if a computer can? :P
-
I would have to say "Not yet".
Current computers can calculate specific numeric values to enormous accuracy at high speed, and can read and write data from the environment at high speed.
Some large computers even achieve reasonable levels of parallelism (100,000x) - but nowhere near what we see in the brain of a whale or even in the human retina.
But they are not so good at complex, open-ended optimisation problems with interlinked dependencies, which are on the edge of feasibility with today's materials.
Current computers lack:
- The knowledge base of humans - but Google has been searching this, and Watson clearly excels in this area.
- Power efficiency - the human brain achieves its results with 20-25W of power consumption, compared to perhaps 100W for a moderately powerful CPU, and >100kW for a highly parallel computer.
- Physical experience of the real world - but perhaps increasing applications of computers in robotics will start to give computers experience of physics, chemistry, materials science and biology
- Wants, Needs, Imagination, Inquisitiveness, Aspirations & Goals - most current computers only carry out the specific goals that are preprogrammed into them; computer goal-setting has been a long-term goal of artificial intelligence research.
- An ability to write & rewrite their own initial programming - effectively to set new Goals
- Ethics - the ability to know whether those Goals are beneficial for the overall good, or not
- Consciousness - we don't really know what this is, but the above factors would certainly contribute!
- Experimental abilities to test results beyond the current knowledge base - but recent progress in genetics research shows a compute-intensive research environment where the computers do the experiments and produce potential hypotheses within the very structured domain of DNA (often ignoring the complexities of real-world chemistry, physics and biology!)
- Sanity-Checking of results - but then humans aren't very good at critiquing their own ideas, either!
- Communication and Collaboration - the ability to work together with humans and other computers to produce, share and coordinate results that no individual computer could. You could call this "Conferencing" or "Society".
Of course, if we do end up with computers that can do all these, we may end up with a "Skynet"!
For now, the computer often is the solution to getting into space: It's too hard to send a person to Mars (with food, water & oxygen), but we can send a nuclear-powered computer... [and ongoing research is looking at making it self-driving, so it doesn't need to roll a few meters, then wait 6-45 minutes while someone on Earth says it is OK to roll another few meters.]
-
Some genetic algorithms can think far, far outside the box, but the big problem is explaining the realistic physics and practical constraints of real world materials to the computer; otherwise the genetic algorithms basically kinda 'cheat' and do things you don't expect.
There's also the issue that the more exact and realistic the models are, the slower they run. So it could take a very, very, very long time to calculate a good answer.
-
Using a Genetic algorithm means that the coder doesn't need to optimise the results - the genetic algorithm does that.
However, the coder does need to provide:
- An accurate evaluation algorithm, which will evaluate the fitness of each generated solution in the (simulated) real world
- A clever way to code every possible solution in a standardised form in which it can be "cross-bred" with any other potential solutions
- Genetic algorithms are naturally suited to parallel evaluation
-
I think accidents on staircases could be reduced by a simple modification ie remove the overlap on the top steps where you turn onto the landing so that you do not catch your toe.
PS I wrote in a school essay in 1942 that race cars could be designed on a purely mathematical basis, I believe HRT tried it but it did not work out.
-
Have you all heard of the latest model? The Super-Turing Machine? It sounds to me a little like what I suggested earlier in the topic and is the kind of computer system that I am sure could do it.