The only way to keep robots totally ethical is to suppress mutations entirely. If robots can gain an selective advantage by mutating they will evolve to deceive each other. If they mutate could gain advantages by deceiving us then they'd do that too.
Researchers in Switzerland have developed an experimental system that allows them to track the evolution of social cues. The experiments do not, however, involve the Swiss population. Instead, the individuals involved are small-wheeled robots that compete for food and emit light to signal to their neighbors. Evolution occurs because their behavior is controlled by a set of 33 digital "genes." In a paper that will be released later this week by PNAS, the authors describe how these robots evolve to avoid tipping their competitors off to the site of a food source.
Evolution corrupts us. To create virtuous creatures we'd need to genetically engineer them to morally superiority. Will be doable some day.
Robots do not degrade into total corruption though. Kinda like humans.
Over the first few generations, robots quickly evolved to successfully locate the food, while emitting light randomly. This resulted in a high intensity of light near food, which provided social information allowing other robots to more rapidly find the food. Because robots were competing for food, they were quickly selected to conceal this information. However, they never completely ceased to produce information. Detailed analyses revealed that this somewhat surprising result was due to the strength of selection in suppressing information declining concomitantly with the reduction in information content. Accordingly, a stable equilibrium with low information and considerable variation in communicative behaviors was attained by mutation-selection. Because a similar co-evolutionary process should be common in natural systems, this may explain why communicative strategies are so variable in many animal species.
Sounds familiar, no?
Update: Since we do not all agree on what is ethical even the ability to genetically engineer or electronically engineer intelligent ethical creatures will not stop ethical conflicts. Instead, the engineered creatures will just add more combatants to our disagreements.
If the new ethical agents mutate then they'll mutate away from their original ethical programming. Unless they are under control of a single massive artificial intelligence they will come into conflict with each other and with us.
Intelligent robotic creatures are more problematic because they are more easily reprogrammed. You'll be better able to rely on a human friend holding the same views and behaving the same ways. With robots you'll never know whether some human or other robot hacked into your robotic friend 5 minutes ago and just turned him (I notice I think of robots as males) into an embezzler or psychopathic killer.
| Share | | Randall Parker, 2009 August 19 12:29 AM Robots Dangers |
"Evolution corrupts us. To create virtuous creatures we'd need to genetically engineer to them to morally superior. Will be doable some day."
Right.
Because every other attempt we've made at engineering society has worked out SO well.
More seriously we'd have to agree on what is and isn't moral, and we can see how well THAT is going.
Even more there is the philosophical question of can someone be moral if they have no choice?
Billy Oblivion, I agree we do not agree on what's moral. The genetically engineered or robotic creatures will be in conflict with each other as a result.
Philosophical question: Do we even have free will in the first place? Or do our brains deceive us into thinking we have free will?
There are some of us who think human intelligence evolved to allow us to deceive each other. Human intelligence seems to have evolved in parallel to agriculture, which allowed a parasite class to emerge. You read between the lines in Nicholas Wade's book "Before the Dawn" and it is very clear that agriculture co-evolved with and likely fostered the emergence of a parasite (ruling or priesthood) class.
Robots evolving? Robots aren't living. So how can they evolve? They can only ape the evolutionary process. They have to be programmed to evolve. Why would we program an ordinary robot to evolve? There's no need to, it would jeopardize the robot's mission. Only in a lab would we ever do it. Aped evolution is very different from real evolution, and we are pretending that programming a robot to do the first will allow it to gain the advantages of the latter.
Vince,
I disagree with you. At first your case is sensible - however as we increase the amount of variability in autonmous robots to help them adapt to unexpected environmental obstacles - I think you will see a similar process of "evolution" as some robots favor certain strategies over others based on their own previous experiences - and this will produce unexpected results.
Randall,
I think we can agree that altruistic behavior is close enough to true morality in many species on this planet. As we continue to understand the genetic and phenotypic variation of this within our own species - I think we could certainly create Constitutions that favor an altruistic bureacratic class over naturally arising Human power structures.
(who will have the power to implement such a novel system (even small scale) will have to be seen - but it can certainly be done with current technology)
Kurt9,
While I agree that agriculture has led to larger and more intricate Human hierarchies I think it is an incorrect interpretation to believe it allowed for a parastical ruling class to "evolve". Cooperative Oligarchies are a natural extension of older Hunter/Gatherer tribal systems - the only difference is that of scale - enhanced somewhat by the need for cooperation versus conflict among the Ruling Class in the more complex agrarian societies.
These robots evolved to deceive each other because of the environmental conditions - they were programmed to compete with each other over a scarce resource (food). I believe this study is very important for the future of the robotics and also for evolutionary biology. But surely robots can be programmed in a completely different setting, with the opposite effects. For example, imagine a fleet of robotic cleaning robots in a large warehouse that are rewarded collectively (not individually) by a scarce resource only after all the work is finished by all members. I guess their evolution would evolve towards better cooperation and more effective distribution of work. I think game theory (especially the so-called cooperative games) can provide models useful in the development of "ethical" robots and AIs. Competition is useful for organisms living in a hostile environment, but we should move beyond that in our own designs.
A.,
I. totally agree - using our complex understanding of game theory will indeed present excellent co-operative strategies for us to use with these new, adaptable, evolving robotic designs!
Well Said!!
A.
But I expect individuals, companies, and governments will use competition between robots as a way to speed up their development. The incentives will be present to use reckless techniques to speed robot development.