As the world celebrates Charles Darwin's 200th anniversary, E&T explains why 'On The Origin of Species' is a useful read for engineers, too.
In 1859 Charles Darwin explained why all animals and plants look and behave as they do - including, by implication, human beings. His book 'On The Origin of Species By Means of Natural Selection' is usually thought of as a primary text for biologists, but these days it might be a good idea for engineers to read it, too.
Machines 'evolve' because the ambitions of their engineers push them towards higher speeds and better energy efficiency, boosting the fortunes of 'good' designs and discarding 'poor' ones. Humans are both the inventors and the builders of machines, so they are not yet the products of a genuinely independent history. Even so, the time is fast approaching when we will no longer be in the driving seat. Soon, machines will survive - or die out - according to their own evolutionary fortunes.
As identified by Darwin, the big trick with living things is that they can replicate themselves and pass on specific survival characteristics to the next generation. The other factor, however, is that those offspring aren't always perfect copies of their parents. They may exhibit minor mutations that affect their ability to survive, for better or for worse. Only the better mutations get passed on through the generations, as the poor ones don't get so many chances to breed.
As the modern Darwinian champion Richard Dawkins observes so succinctly: "The secrets of evolution are death and time."
Machines will soon exhibit all these characteristics. At last year's Cheltenham Science Festival, Dr Adrian Bowyer from the University of Bath's School of Mechanical Engineering demonstrated a harmless-looking contraption - somewhat like a home-made desktop printer, yet this tinker toy could be the germ cell of a new machine species. Almost all the components in the replicating rapid prototyper (RepRap) were constructed by an earlier version of the same machine.
RepRap is derived from rapid prototyping (RP) printer technology that converts virtual 3D computer models into physical objects.
A typical RP printer works in much the same way as a conventional paper printer, delivering two-dimensional sheets of output. The extra third dimension comes by virtue of stacking successive sheets on top of each other like the stepped levels of an Egyptian pyramid. Stand far enough back, or use sufficiently fine sheets, and you don't see the jagged steps.
Instead of inks, RP printers use plastic powders, ultraviolet-hardened resins, or microscopic beads of metal fused together with lasers.
Most RP machines can output components of startling complexity. Some of these are actually usable in finished products designed for heavy loading. It is even possible to output products with internal moving parts. Analyse any complicated structure as a series of stacked slices - and you should be able to print out each slice, so long as you have the right 'inks' on your printer heads.
RP is ideal for fabricating the subtle compound curves and variable cross-sections in components for racing cars, jet engines and other high-performance machines designed with help from fluid dynamics software. Architects testing their less computationally demanding designs in miniature also find RP invaluable for churning out drearily repetitious components, such as window frames in the models of a building.
In fact, it's possible to build the actual building. At the University of Southern California, Dr Berok Khoshnevis' engineering department is experimenting with giant house printers. Hour by hour, a rectangular platform on four hydraulic stilts extrudes concrete and other fluid materials which harden in place, ready to receive the next layers. The only 'cheats' are in the window glass, which comes preformed, and the electrical wiring, which unrolls from spools. Khoshnevis' ultimate goal is "to be able to completely construct a one-story, 2,000-square foot home on site, in one day and without using human hands."
In 1966, the mathematician John von Neumann predicted the possibility of machines capable of churning out copies of themselves, and of assembling more or less any other object we require of them. The only downside of RP systems today is that they take hours, rather than minutes, to deliver even the simplest 3D objects, but this is bound to change as the technology improves. Just as PCs are 'universal computing devices,' RP machines could become von Neumann's 'universal constructors.'
Constructors of themselves
Andrew King, lecturer in 3D Design at the Arts Institute in Bournemouth (a national leader in RP modelmaking tuition) predicts that "within the next five years children's computers will output toys overnight, to the bespoke demands of their young clients. Meanwhile, serious consumer product manufacturers will be able to cater for individual customer's tastes, instead of forcing all of them to choose, in the famous words attributed to car maker Henry Ford, 'Any colour they like so long as it's black'".
Universal constructors could democratise the ownership of manufacturing companies. Adrian Bowyer certainly sees his RepRap technology as a step in this direction. "Today, you have to spend a huge amount of money to build a factory production line just for a specific product. You can pay people low wages but you still have to make a huge capital investment.
"What if you could manufacture a multitude of different things, and for a very small initial investment? It could be a powerful mechanism for elevating people from poverty."
Because each universal constructor will be able to create copies of itself, cash-strapped investors will only have to pay for basic construction materials, and not the machines themselves. After a few generations, enough machines will be on-site to create an economically viable manufacturing hub for the products that the vendors actually want to sell to their customers: that is, unless this technology so changes the economic landscape that buying and selling physical products no longer makes sense. In such a world, raw materials would be at a premium while specialised factories might pass into history.
Someone still has to tell the universal constructors what to make, regardless of whether it's copies of themselves or components for other products. Most people lack the software skills or artistic talents required to operate the 3D modelling software that drives the process.
ReRap's software is open source, and therefore startlingly amenable to swift evolutionary improvement, but it still needs plenty of human expertise to 'mutate.'
In 1970, Cambridge scientist John Conway invented a computer game he called 'Life'. The rules were simple. Dark squares (cells) are arranged in a grid somewhat like a sheet of graph paper with occasional squares inked in. Dark cells stay alive so long as they have at least two neighbours, but die if they are hemmed in by four. If conditions are right, they will generate another 'live' cell nearby. The human programmer decides on an initial pattern of cells, and then the computer takes over, iterating the same simple rules again and again.
Too few cells in the initial pattern, or cells spaced too widely apart, may lead to an empty grid at the end of the game, but something as simple as a T-shaped starting pattern, for instance, produces amazing results. Cells breed, live and die, filling the grid with patterns of stunning orderliness and complexity. Watch the game develop, especially on a high-speed computer, and it is hard not to imagine that something in the system is alive.
A compelling twist to 'Life' is to set the conditions you want to achieve at the end of the game: a triangular array of cells, for example, or even a picture of some kind, using dark cells as 'pixels.'
The only extra requirement is a rule that compares each state of play with the desired end result, preserving the states that edge closer to it and discarding those that don't. Genetic algorithms (GA) extending these ideas are now a major trend in software.
Peter Bentley, research fellow at University College London's Department of Computing, suggests that GA could supplant human intervention in the manufacturing process, requiring us simply "to nudge the universal constructors in the direction of what we want". Some of Bentley's RP-produced home furnishings prove his point: "I'm probably the only person with an evolved coffee table."
However, all this assumes something that we don't find in the biological world: a desired end result. At some point in the future, we may encounter a disconnect between human intentions and what machines become capable of, regardless of our wishes. Machines will simply adapt to survive in whatever environment they happen to inhabit.
At that point, they will no longer be 'artificial'. They will be just another facet of the natural world.
We have to wonder if machines will evolve the power of thought. George Dyson, author of the influential book 'Darwin Among the Machines', is convinced that artificial intelligence "will have to emerge naturally, rather than as a product of some specific design on the part of human programmers". We don't yet know what consciousness is, and therefore we are hard-pressed to bring it to life inside a computer.
However, if we concentrate just on the characteristics that we'd like a supposedly sentient computer to exhibit, and then unleash simple yet vastly numerous fragments of code into the system, then something might emerge that matches our hopes - even if, when it finally comes to life, we won't necessarily be able to fathom its intricacies.
Dyson suggests that the Internet may already exhibit occasional, accidental signs of a 'precursor' mind and warns that when the first fully-fledged artificial intellect does finally announce itself, "we shouldn't expect it to operate on a level that we are able to comprehend. When it speaks, there is no guarantee that it will be in a language we can understand".
Artificial intelligences will evolve along their own path. Ten thousand years from now, they may remember the human role in seeding them as just a brief, lucky environmental circumstance in their early history.