The next step in engineering - AI algorithms

[Juddog]

Reading some recent articles, it seems that at the high end, AI is being used to solve engineering problems via brute-forcing.

That is to say, if a human were trying to go about solving a problem, they may not be able to solve all of the various angles of an issue unless they are some kind of once-in-a-lifetime supergenius, versus a computer AI / algorithm which can do the same thing every time.

These two articles are what made me think of this:

http://www.mit.edu/newsoffice/2013/a-faster-internet-designed-by-computers-0719.html

http://idesign.ucsc.edu/projects/evo_antenna.html

So what we're basically looking at here is the end of engineering design by humans, when they reach the limit of something, and letting AI generate a better solution.

What are your thoughts on this occurring? On the downside, engineers in the traditional sense might be out of a job - if you can for example build a bridge in the best possible way via a computer simulation which generates a custom design for that area and gives the maximal strength in terms of usage of the building materials, versus having a human engineer try and generate the design in auto CAD and take 4 times as long and have the bridge not be as good.

On a positive note though, everything built would be a lot better - super efficient cars that ran better on less fuel that were also safer, better medicine customized to an individual user's genetics, better operating procedures for a hospital, better rockets, satellites, etc.. Basically almost everything a human could design potentially could be designed better / more efficient.

What would be the upsides / downsides to this future occurring?

 

[Ancalagon44]

In some instances, such things have been used for years. Sometimes they are worse than work that humans could produce, but they save massive amounts of time.

An example is a compiler. Humans can write assembler if they want to, and when they do, they typically do it much better than a compiler can. However, a compiler can do the same thing much, much faster than a human, leading that human being more productive because he is free to work on other things.

The other thing to note is that algorithms that produce solutions to engineering problems must have their constraints defined explicitly by humans, as well as assumptions upon which the model works. For instance, we could tell an AI to design a bridge, but we'd have to tell it what the prevailing windspeed is, what the maximum windspeed is, what the tensile strength of the materials that it can use, the elevation of the bridge etc etc. My point is that, such algorithms will only be as good as the input into the system.

Also its worth noting that sometimes such algorithms tend to produce local maxima instead of global maxima. That is, a suboptimal solution which just compares better than solutions that the AI tested, but not necessarily the best solution overall.

 

[Hayabusa Rider]

Eventually humans will be obsolete living a pointless existence as some machine can do everything better. What joy.

 

[jackstar7]

Until an AI can imagine, we retain a function.

 

[sm625]

What a bunch of nonsense. Intelligence is not a clustering of algorithms. Algorithms are tools, no different than screwdrivers and vise grips. They will not be a conduit for any future sentience, self-awareness, or consciousness for some hypothetical sci-fi construct.

They will, however, be put to great use by such a beast when it does obtain sentience.

 

[Juddog]

In some instances, such things have been used for years. Sometimes they are worse than work that humans could produce, but they save massive amounts of time.

An example is a compiler. Humans can write assembler if they want to, and when they do, they typically do it much better than a compiler can. However, a compiler can do the same thing much, much faster than a human, leading that human being more productive because he is free to work on other things.

The other thing to note is that algorithms that produce solutions to engineering problems must have their constraints defined explicitly by humans, as well as assumptions upon which the model works. For instance, we could tell an AI to design a bridge, but we'd have to tell it what the prevailing windspeed is, what the maximum windspeed is, what the tensile strength of the materials that it can use, the elevation of the bridge etc etc. My point is that, such algorithms will only be as good as the input into the system.

Also its worth noting that sometimes such algorithms tend to produce local maxima instead of global maxima. That is, a suboptimal solution which just compares better than solutions that the AI tested, but not necessarily the best solution overall.

In regards to the bridge example, one would expect to input as much as possible for the program to produce the best outcome. It might be able to provide the best solution for a specific area of the planet versus a global solution which worked best everywhere.

We've had compilers for years, but they are getting better and better, and we haven't had something advanced to the level of the TCP traffic shaping until very recently. Given the jump in complexity that the AI was able to handle, if it continues to progress as it has, then more and more engineering problems should become available for it to solve.

 

[Juddog]

Eventually humans will be obsolete living a pointless existence as some machine can do everything better. What joy.

When it gets to that point, what will happen to society?

Will we have utopia, or some dystopian future where society degenerates?

 

[v-600]

To go slightly off topic, the little anarchistic part of me is eagerly awaiting this. Bridges are not all that important in the grand scheme of things, so moving away from your bridge example to more basic examples before we come back to it.

I would say that people need food, water, shelter to survive (saying nothing about thriving at the moment). I have seen GPS controlled tractors and ploughs on TV. I dare say water treatment plants rely heavily on computers and I can easily imagine construction robots building prefabbed houses. Likewise I can imagine automated mining machinery producing raw materials for all this.

I look forward to the day I don't need to pay for food, water or housing because it was made for me by a machine that didn't need paying, nor indeed itself cost anything as it was made by another machine that didn't need paying etc etc etc. This puts workers out of paid jobs, but its not as bad as it sounds as they can get food, water, shelter etc for nothing from a machine that didn't need paying etc etc etc.

People then work not because they need to, but because they want to and because they enjoy it. It doesn't matter that somebody can do something better than you, that already happens and people get one fine with it. I enjoy running with my friends, I don't enjoy it any less because other people can run further or faster than me. I dug a pond last weekend and it doesn't devalue it to me that a landscape designer could have made a nicer one, or that a machine could have made a bigger one more quickly. I enjoyed the work and the result.

In part we already live in a culture where many people have things they don't need purely to show other people they can have them. Imagine tomorrow that everyone can have an iphone for free if they want one. It doesn't suddenly become less good, it can do the same things as before but some people would want one less because it loses that "Look at me and what I can have that you can't" appeal. Some people don't just have a pond/pool/car/phone/whatever because they like to look at it, but because someone else can't have one.

Unfortunately, all the above is still constrained by energy. The best computers/machines/people in the world are useless without energy (electric/hydraulic/food). In the long view until we crack the problem of producing more power than we need, the above won't happen.

 

[Juddog]

To go slightly off topic, the little anarchistic part of me is eagerly awaiting this. Bridges are not all that important in the grand scheme of things, so moving away from your bridge example to more basic examples before we come back to it.

I would say that people need food, water, shelter to survive (saying nothing about thriving at the moment). I have seen GPS controlled tractors and ploughs on TV. I dare say water treatment plants rely heavily on computers and I can easily imagine construction robots building prefabbed houses. Likewise I can imagine automated mining machinery producing raw materials for all this.

I look forward to the day I don't need to pay for food, water or housing because it was made for me by a machine that didn't need paying, nor indeed itself cost anything as it was made by another machine that didn't need paying etc etc etc. This puts workers out of paid jobs, but its not as bad as it sounds as they can get food, water, shelter etc for nothing from a machine that didn't need paying etc etc etc.

People then work not because they need to, but because they want to and because they enjoy it. It doesn't matter that somebody can do something better than you, that already happens and people get one fine with it. I enjoy running with my friends, I don't enjoy it any less because other people can run further or faster than me. I dug a pond last weekend and it doesn't devalue it to me that a landscape designer could have made a nicer one, or that a machine could have made a bigger one more quickly. I enjoyed the work and the result.

In part we already live in a culture where many people have things they don't need purely to show other people they can have them. Imagine tomorrow that everyone can have an iphone for free if they want one. It doesn't suddenly become less good, it can do the same things as before but some people would want one less because it loses that "Look at me and what I can have that you can't" appeal. Some people don't just have a pond/pool/car/phone/whatever because they like to look at it, but because someone else can't have one.

Unfortunately, all the above is still constrained by energy. The best computers/machines/people in the world are useless without energy (electric/hydraulic/food). In the long view until we crack the problem of producing more power than we need, the above won't happen.

That itself may be something that we need an engineering AI program to solve. We know that fusion exists, and the mechanism for how it works, but have been unable to create a fusion reactor yet.

The day may arrive where we are able to create a safe reactor through the use of an AI creating a way to do so.

 

[Moonbeam]

When it gets to that point, what will happen to society?

Will we have utopia, or some dystopian future where society degenerates?

Obviously we will either have to set the machines the task of designing people with happier DNA so that pessimistic and gloomy folk like Hay become anachronistic or design mechanical psychotherapist which will know how to cure his symptoms. There have to be reasons for how he sees things and how v-600 does.

 

[Hayabusa Rider]

Obviously we will either have to set the machines the task of designing people with happier DNA so that pessimistic and gloomy folk like Hay become anachronistic or design mechanical psychotherapist which will know how to cure his symptoms. There have to be reasons for how he sees things and how v-600 does.

Heavens M I've been an anachronism from the day I first learned to see. The world moves, the circle turns and my sense of things will pass. The new will come with no such complications. Things will be simpler then.

 

[Sheep221]

The only advantage of this is time, it won't design anything better way than regular workers, yet it will do it faster and can make multiple attempts way faster to get best results.

 

[Paul98]

http://www.dailymotion.com/video/x1...man-evolution-hdtv-****-afg_tech#.Ue7Lp403veG

 

[Hayabusa Rider]

The only advantage of this is time, it won't design anything better way than regular workers, yet it will do it faster and can make multiple attempts way faster to get best results.

But what about those "regular workers"? What of human psyche which wants to add value, to provide and care for those he or she loves? What happens when a child is born in a world where all his contributions are inferior? When aspiration is pointless? When even art and music is usurped?

What happens when there is no need to be better? When the core of what we are, our dreams and wishes irrelevant, being fed like cattle from a trough?

Even of there were no Morelocks I'd not trade my fears and dreams and hopes for the life iof an Eloi.

I am a dinosaur.

 

[Sheep221]

But what about those "regular workers"? What of human psyche which wants to add value, to provide and care for those he or she loves? What happens when a child is born in a world where all his contributions are inferior? When aspiration is pointless? When even art and music is usurped?

What happens when there is no need to be better? When the core of what we are, our dreams and wishes irrelevant, being fed like cattle from a trough?

Even of there were no Morelocks I'd not trade my fears and dreams and hopes for the life iof an Eloi.

I am a dinosaur.

I don't want this to happen either, yet it may once become reality.

 

[v-600]

Your contributions will not be worthless, they will just be voluntary. If you make something of worth/quality, it does not suddenly become pointless just because someone else can also do it. However you no longer have to work merely to survive, but can work to further your dreams and wishes. E.g. I wish I could climb mount everest one day but a permit costs thousands, and the gear is so expensive and I've got a family to feed vs, I wish I could climb mt everest and I can because I don't have to worry about the necessities.

 

[v-600]

PS. I just re-read that and I appear to be coming across as a benevolent dictator/propagandist. Come live in my utopia

 

[Juddog]

The only advantage of this is time, it won't design anything better way than regular workers, yet it will do it faster and can make multiple attempts way faster to get best results.

If you check this link, you'd find otherwise:
http://idesign.ucsc.edu/projects/evo_antenna.html

The whole point of the original post is that we're at the stage now where we can have AI design better things than humans can. How many years would it have taken humans to find that oddball shape for the antenna as being the best?

 

[Juddog]

The way I see it is this. Traditional engineering designs something, then builds a model replica, then puts it through a simulation to see if it works, then goes to build the real thing. You simply don't have enough man hours to have a huge staff of engineers all working on the same thing at the same time to get the best of the batch.

With an advanced enough computer AI, you can essentially throw some serious processing power at something and come up with the equivalent of an entire team of engineers all trying their different way to design something, then taking the best design, then testing it, etc., but all from a single program. Each generation the AI gets better, the computers get faster, and it is able to get even better than what it did before at brute forcing every possible design that would fit something in particular the best way. At some point it will get to be so far beyond human level engineering that anything created by a person would appear to be inferior.

 

[v-600]

I think it will be a long time before simulation begins to approximate real world designers. I read this article a few years ago now and it kind of stuck in my mind (especially after a lecture on neural networks in computing at uni).

http://www.damninteresting.com/on-the-origin-of-circuits/

Computers now are able to perform trial and error testing simulation far faster than people could do a thought experiment. However actually testing these in the real world will still be important.

 

[Pulsar]

If you check this link, you'd find otherwise:
http://idesign.ucsc.edu/projects/evo_antenna.html

The whole point of the original post is that we're at the stage now where we can have AI design better things than humans can. How many years would it have taken humans to find that oddball shape for the antenna as being the best?

You're confusing two things in your thoughts.

On one side, engineers are very required. They have ingenuity, and understand how to apply formulas to problem solve.

The applications that you are linking to are situations where engineers have set up the formula and written the programs to solve problems that are highly iterative. That means there is no single solution, and it requires solving the same equation billions and trillions of times while varying different inputs to find the best solution.

For instance, weather forcasting is the solution of multiple iterative equations.

This isn't a computer designing a better system. It's a computer iterating a formula to optimize settings. There's a big difference. We're still quite a ways away from having computers that make decisions of any sort.

Fluid dynamics, nuclear engineering, chemistry, DNA and Gene sequencing, and a whole slew of other applications can allow computers to iterate to a solution. Hell, they even use computers to iterate statistics for sports like baseball.

 

[Hayabusa Rider]

PS. I just re-read that and I appear to be coming across as a benevolent dictator/propagandist. Come live in my utopia

Na, not at all.

I'm not even saying that what passes for intelligence in machines is wrong in itself as long as the humans are what ultimately matter.

Here's the thing. We've seen people become redundant. If the truth be known our society is doing it's damndest to figure out to do with people who used to make things. Machines are cheaper or at least less trouble in the long run. Since those who control business select that which provides the less economic cost machines win. There is no reason whatsoever to believe that as the capabilities of machines increase that the trend will abate.

What you hope will happen is that this will eventually eliminate working for one's needs. Ask and ye shall receive, manna for the masses by machine. Perhaps that's true.

That ignores some basic facts about humans. We are a social species, and as such we have hierarchies of dominance. We insist on it. Most of us cannot imagine a world where we do not have people who lead us. There are also people who are compelled to lead, or more likely rule in some way. The difference is that the former can be done by example, but the latter allows coercion. That's how governments enforce laws, and it's not inherently a bad thing. It just is.

I expect that the majority of people if given enough material goods will accept a great deal. Those who's nature is to control though aren't likely to surrender it. Far different than goods and money and things which can be bought is power. The control of others by means good or ill.

Now who would control this magic box? This stuff? You might say that no one has to and you would be right, however neither corporate nor government mentalities will accept that. There seems to be a majority view that corporations own government because politicians are greedy. That may be, but consider that the majority of people at the level of national leadership don't need more. They have enough material wealth. Why then? Power. Power to do good or bad or benefit oneself or others in a beneficent way, but it's still control.

That isn't going away.

Still, let's assume that at some point the machines are benign and that even the most alpha human is surpassed. Plenty for all. What then?

Let's go back to human nature and look at your example of climbing a mountain. I never will be at the top of Mt. Everest, but I have been to the summit of a mountain which has some of the most hellacious weather on the planet. You can also drive to the top.

In terms of absolute objective reward, driving and climbing ought to be identical, but I can assure you that for me and most people I know it's not. There is a satisfaction associated with the accomplishment of such a feat which cannot be replaced with ease of access. Take away the challenge, take away the reward. It's how we're wired.

That brings us to another quality of humanity, the desire to make a difference in a unique way. To strive to go where no one has before, if it's a new land or a new discovery. The thrill of the chase, of being unique. The very best.

What happens when the best that can be done is not within human ability? What happens when fathers have to tell their sons that their best isn't necessary. Some machine will do it better. Soon the sons won't be saying anything to their children. It will be accepted that all of mankind is redundant. No frontiers in reality. Oh we can pretend it's not so, but inside we'd all know.

And so we become pets. We can pretend we have the ability to create something unique and meaningful, but the machine has eclipsed us. What happens when the master sees it's beloved pet unhappy? In this case what Moonbeam suggests is likely. Breed out those obsolete notions. Happy creatures all the time without care or desire or a spirit to overcome. Placid beasts.

The day of the Eloi will have arrived, and there will be nothing but contentment. May I and those I love die long before that time.

If there was ever a time where our collective fate hangs by our choices it is now. I suggest a lot of careful thought on how to proceed.

 

[Moonbeam]

Na, not at all.

I'm not even saying that what passes for intelligence in machines is wrong in itself as long as the humans are what ultimately matter.

Here's the thing. We've seen people become redundant. If the truth be known our society is doing it's damndest to figure out to do with people who used to make things. Machines are cheaper or at least less trouble in the long run. Since those who control business select that which provides the less economic cost machines win. There is no reason whatsoever to believe that as the capabilities of machines increase that the trend will abate.

What you hope will happen is that this will eventually eliminate working for one's needs. Ask and ye shall receive, manna for the masses by machine. Perhaps that's true.

That ignores some basic facts about humans. We are a social species, and as such we have hierarchies of dominance. We insist on it. Most of us cannot imagine a world where we do not have people who lead us. There are also people who are compelled to lead, or more likely rule in some way. The difference is that the former can be done by example, but the latter allows coercion. That's how governments enforce laws, and it's not inherently a bad thing. It just is.

I expect that the majority of people if given enough material goods will accept a great deal. Those who's nature is to control though aren't likely to surrender it. Far different than goods and money and things which can be bought is power. The control of others by means good or ill.

Now who would control this magic box? This stuff? You might say that no one has to and you would be right, however neither corporate nor government mentalities will accept that. There seems to be a majority view that corporations own government because politicians are greedy. That may be, but consider that the majority of people at the level of national leadership don't need more. They have enough material wealth. Why then? Power. Power to do good or bad or benefit oneself or others in a beneficent way, but it's still control.

That isn't going away.

Still, let's assume that at some point the machines are benign and that even the most alpha human is surpassed. Plenty for all. What then?

Let's go back to human nature and look at your example of climbing a mountain. I never will be at the top of Mt. Everest, but I have been to the summit of a mountain which has some of the most hellacious weather on the planet. You can also drive to the top.

In terms of absolute objective reward, driving and climbing ought to be identical, but I can assure you that for me and most people I know it's not. There is a satisfaction associated with the accomplishment of such a feat which cannot be replaced with ease of access. Take away the challenge, take away the reward. It's how we're wired.

That brings us to another quality of humanity, the desire to make a difference in a unique way. To strive to go where no one has before, if it's a new land or a new discovery. The thrill of the chase, of being unique. The very best.

What happens when the best that can be done is not within human ability? What happens when fathers have to tell their sons that their best isn't necessary. Some machine will do it better. Soon the sons won't be saying anything to their children. It will be accepted that all of mankind is redundant. No frontiers in reality. Oh we can pretend it's not so, but inside we'd all know.

And so we become pets. We can pretend we have the ability to create something unique and meaningful, but the machine has eclipsed us. What happens when the master sees it's beloved pet unhappy? In this case what Moonbeam suggests is likely. Breed out those obsolete notions. Happy creatures all the time without care or desire or a spirit to overcome. Placid beasts.

The day of the Eloi will have arrived, and there will be nothing but contentment. May I and those I love die long before that time.

If there was ever a time where our collective fate hangs by our choices it is now. I suggest a lot of careful thought on how to proceed.

The robots in The Day the Earth Stood Still had the power and last night I achieved my personal best on my Ipad playing solitaire. I jogged around the block screaming at the top of my lungs, I did it, but my score was maybe only 1/4 the best. Eh, I only play during commercials. And while you may die, but Eeyore is eternal and will always bring great joy when children play Pooh Sticks.

 

[smackababy]

What you're talking about really isn't an AI. More of just a calculator. They will never come up with new ideas. Just ones based on calculations. Computers have no imagination and can not (as of now) learn new ideas through experience. They are programmed to test based on knowledge we already know.

They might find the most efficient route for network traffic or the most efficient antenna design, but they will never invent an antenna. They will not discover new technology, just advance current technology, for the most part.

Even still, true AI fascinates me quite a lot. A lot of jobs can be eliminated by programming and robotics, but thankfully my job is software development so I am not really in danger of that happening.

 

[Hayabusa Rider]

The robots in The Day the Earth Stood Still had the power and last night I achieved my personal best on my Ipad playing solitaire. I jogged around the block screaming at the top of my lungs, I did it, but my score was maybe only 1/4 the best. Eh, I only play during commercials. And while you may die, but Eeyore is eternal and will always bring great joy when children play Pooh Sticks.

Ahh but we're not discussing the power of machines, or at least I'm not. For clarity I'm referring to the possibility of self aware and superior minds given control because of economics at least initially. The conclusions I reached is based on that as a possibility. Whether it comes to pass or not we have already cut out people from traditional jobs because we have decided that their dignity is secondary to productivity. I think this isn't a thing to be taken lightly.