Above is a video that takes artificial intelligence into a whole new realm. Up until now, machines have been programmed only to do a specific set of tasks. But with advent of Hod Lipson's models, it is now possible to create machines that, following trial and error, can learn about themselves learn to do a specific set of tasks to meet a specified goal. Here, I shall discuss the how far this technology can simulate the living world, what are it's limits, and how it could change the industry's views on the division of processes that can be mechanized and otherwise.
Basic necessities for simulating a bot that learns and evolves:-
These robots work on a 'Reward' system. It is equipped with some basic motor/output functions, along with some sensors/input for feedback. The bot feels rewarded when it achieves a certain task, which it perceives via it's sensors. Beyond that it is not told what it looks like or what each motor function does. Besides, a specific probability is coded into the bot, which decides how often the bot would try out a new set of moves.
To understand this better lets take the example of Hod Lipson's "spider". It has eight motors and two tilt sensors to start with. It feels rewarded if it moves forward. The greater the speed, more the reward. The "spider" would, based on the probability coded into it, would either do a specific set of movements (say 'A'), that it knows would yield the best reward (tried and tested), or try out a random set of movements. If the new set of movements were more rewarding (say 'B'), then 'B' would be the new 'A'. And this cycle would continue till it is switched off.
Hod Lipson has also observed that when there is no specified reward that has been coded for, then the intrinsic reward of the bot is to self-replicate, when in a population.
Analogies with the living world:-
Just as human beings, these bots are equipped with sensory and motor functions, and are not programmed to do any one particular task. And just as human beings, these bots try random tasks to achieve the ultimate reward. To understand that human beings also work on the basis of a reward system, one can simply refer to the reward of the AI bots as an "Artificial orgasm". If we look at the big picture, it is easy to see that all our actions are pointed towards attaining the ultimate reward. It is also possible to interpret that we are built with no specific reward encoded in us. But then, mechanists can draw analogies with Lipson's squares (refer the video) where in no specific reward is encoded, and that the intrinsic reward is to substantiate their population as a whole. This makes reproduction, multiplying and substantiating the populace as the intrinsic reward for all living organisms, including humans. Thus, these AI bots can simulate any living organism.
One can also argue that these bots simulate the ultimate designer, hired by mother nature herself - Evolution. Evolution is the process by which traits in organisms get honed as they go down the generations. This is done by the process of mutation, where in a slight variation is induced into one of the organisms of a species. If this variation were beneficial, it would endure and the trait would be passed down the generations; else it would disappear. This same is imbibed into these AI bots; wherein the probability encoded deciding how often the bot reverts from it's normal course of action to a random set of movements, simulates mutation in living organisms.
Limitations of AI bots:-
It would be difficult to simulate an organism to it's fullest as the sensory and motor modules of living organisms are very vast. Consider human beings. Each cell of the retina (part of eye), each taste bud, each recipticle in the nose, inner ear and skin represents a separate sensor. Imbibing this in an artificial organism is very difficult. Besides, even if we do create such an organism, the processing power of the driving chip of the bot would be much slower than the rate at which a living organism evolves. That would mean, if it took years for a human being to evolve and groom itself, it could take millenia for the same kind of evolution in artificial human beings.
The statement - "Lipson's model of AI bots simulates evolution in itself", is not entirely correct. The code of a living organism (it's DNA) also includes the description of the sensory and motor modules and their functions. Mutation and evolution has caused living organisms to sprout new sensory and motor functions. But artificial machines cannot alter it's sensory and motor modules.
How could it change the industry's ability to differentiate the mechanizable from the not:-
First, we must understand why automation is used in the industry to replace human labor where ever possible. The human brain receives a lot of data from it's senses to comprehend and do tasks at hand quickly. But in automaton, only the data necessary to do the task at hand is fed in, and the machine can do these tasks much faster. These evolutionary bots have the scope to raise the bar even higher, given time and effort. "Management is an uncertain event and is thus non-mechanizable", would be a statement of the past with the advent of these AI bots. But industry has no room for mistakes, which are the basis for the bot's intelligence.
--Amit M. Warrier
EE09B004
References:-
1) The Video shown above
2)Defecating Duck, by Jessica Riskin (for the understanding of the mechanist and realist views on automation)
Basic necessities for simulating a bot that learns and evolves:-
These robots work on a 'Reward' system. It is equipped with some basic motor/output functions, along with some sensors/input for feedback. The bot feels rewarded when it achieves a certain task, which it perceives via it's sensors. Beyond that it is not told what it looks like or what each motor function does. Besides, a specific probability is coded into the bot, which decides how often the bot would try out a new set of moves.
To understand this better lets take the example of Hod Lipson's "spider". It has eight motors and two tilt sensors to start with. It feels rewarded if it moves forward. The greater the speed, more the reward. The "spider" would, based on the probability coded into it, would either do a specific set of movements (say 'A'), that it knows would yield the best reward (tried and tested), or try out a random set of movements. If the new set of movements were more rewarding (say 'B'), then 'B' would be the new 'A'. And this cycle would continue till it is switched off.
Hod Lipson has also observed that when there is no specified reward that has been coded for, then the intrinsic reward of the bot is to self-replicate, when in a population.
Analogies with the living world:-
Just as human beings, these bots are equipped with sensory and motor functions, and are not programmed to do any one particular task. And just as human beings, these bots try random tasks to achieve the ultimate reward. To understand that human beings also work on the basis of a reward system, one can simply refer to the reward of the AI bots as an "Artificial orgasm". If we look at the big picture, it is easy to see that all our actions are pointed towards attaining the ultimate reward. It is also possible to interpret that we are built with no specific reward encoded in us. But then, mechanists can draw analogies with Lipson's squares (refer the video) where in no specific reward is encoded, and that the intrinsic reward is to substantiate their population as a whole. This makes reproduction, multiplying and substantiating the populace as the intrinsic reward for all living organisms, including humans. Thus, these AI bots can simulate any living organism.
One can also argue that these bots simulate the ultimate designer, hired by mother nature herself - Evolution. Evolution is the process by which traits in organisms get honed as they go down the generations. This is done by the process of mutation, where in a slight variation is induced into one of the organisms of a species. If this variation were beneficial, it would endure and the trait would be passed down the generations; else it would disappear. This same is imbibed into these AI bots; wherein the probability encoded deciding how often the bot reverts from it's normal course of action to a random set of movements, simulates mutation in living organisms.
Limitations of AI bots:-
It would be difficult to simulate an organism to it's fullest as the sensory and motor modules of living organisms are very vast. Consider human beings. Each cell of the retina (part of eye), each taste bud, each recipticle in the nose, inner ear and skin represents a separate sensor. Imbibing this in an artificial organism is very difficult. Besides, even if we do create such an organism, the processing power of the driving chip of the bot would be much slower than the rate at which a living organism evolves. That would mean, if it took years for a human being to evolve and groom itself, it could take millenia for the same kind of evolution in artificial human beings.
The statement - "Lipson's model of AI bots simulates evolution in itself", is not entirely correct. The code of a living organism (it's DNA) also includes the description of the sensory and motor modules and their functions. Mutation and evolution has caused living organisms to sprout new sensory and motor functions. But artificial machines cannot alter it's sensory and motor modules.
How could it change the industry's ability to differentiate the mechanizable from the not:-
First, we must understand why automation is used in the industry to replace human labor where ever possible. The human brain receives a lot of data from it's senses to comprehend and do tasks at hand quickly. But in automaton, only the data necessary to do the task at hand is fed in, and the machine can do these tasks much faster. These evolutionary bots have the scope to raise the bar even higher, given time and effort. "Management is an uncertain event and is thus non-mechanizable", would be a statement of the past with the advent of these AI bots. But industry has no room for mistakes, which are the basis for the bot's intelligence.
--Amit M. Warrier
EE09B004
References:-
1) The Video shown above
2)Defecating Duck, by Jessica Riskin (for the understanding of the mechanist and realist views on automation)
No comments:
Post a Comment