With the recent explosive growth of computers and robotics, the question as to whether or not a robot will ever be capable of replacing a human being is becoming more and more of a topic of discussion. One of the biggest arguments used by those who claim the two will never be equal is the matter of imagination, creativity, and inspiration. These things, without any plausible science, lead to a primarily philosophical discussion. This paper will address some of the other aspects that people use when comparing the viability of robots as human replacements. The three following topics are rooted in comparisons between the human brain and a microprocessor. This is, at its most basic level, the robot verses human argument. We, as humans, are powered by our brain while they, as robots, are powered by their microprocessor. Three main criteria used to evaluate both parties are: capacity for memory, speed of processing, and the ability to learn.
It is very clearly understood that machines are almost infinitely superior to human beings in regards to memory. A robots memory is limited only by the amount of memory space given to it, there is no decay over time, and recalling something from anywhere in its memory banks is instantaneous. This clearly gives robots the edge over humans in the memory department.
What about the rate at which a robot can process information relative to how fast a human brain can do so? While there is no exact method to measure how fast the human brain is, it is commonly agreed upon that the rough estimate is 100 million MIPS (million instructions per second). This means that the human brain can process roughly 100000000000000 instructions per second. Look at that number for a second and attempt to count the zero's. I'll save you the headache, there are 14 of them. That's how many commands or instructions the human brain can process in a single second. Impressive, isn't it? Translating that into computer speeds yields roughly 1,640 GHz. The fastest production computer processor at the moment, the Core i7-975, clocks in at 3.33 GHz. 1,640 GHz versus 3.33 GHz. While the human brain may be lacking in memory, its victory over computers is crystal clear when it comes to processing speed.
The remaining topic from the list is the ability to learn. Learning is thought of by many to be one of the biggest differences between people and robots. It is seen as one of the most difficult things that robotics researchers and developers will have to overcome if they want to create a robot that can be considered as a rival to a human being (on a functionality/capability basis only, lets not dive in to a human versus robot war just yet). However, before we can delve in to whether or not robots can compete with humans in the realm of learning, we have to first agree upon what exactly learning is. This can be a very hard concept to pin down to a single definition, so for the sake of discussion we'll use the Mirriam Webster definition of learning as "gaining knowledge or understanding of or skill in by study, instruction, or experience". Now that we've established what learning is, do robots exhibit these characteristics?
Science fiction work, for the most part, affirms the belief that robots are capable of learning. One example of such science fiction is the movie Stealth. Stealth involves the creation of a completely autonomous and artificially intelligent aircraft and its integration into an elite squadron of human pilots. EDI, as the newly developed aircraft is called, is sent out to fly with the squadron he is to join so that he can learn advanced air combat maneuvers. This is a prime example of a machine or a robot learning things on the fly (pun intended) rather than relying on pre-programmed information. Air combat maneuvers, especially those performed at very high speeds and in close chronological proximity with each other, are some of the most mentally taxing things that a human being is capable of doing and they require extreme amounts of focus, skill, precision, and very quick thinking.
In the film, EDI is sent up to merely observe the other (human) pilots as they fly. The learning doesn't only take place from watching though. EDI learns about how his body (the aircraft) maneuvers through trying things and seeing or feeling the reaction. So EDI is learning on two levels, he is learning about tactical flying formation, how to be a wingman, and observing advanced combat maneuvers all by watching the human pilots, but at the same time, he is "feeling" how the aircraft he is a part of reacts to things such as aileron, elevator, and rudder movement. The learning done by EDI is not purely on a professional basis however. His learning capabilities extend to the behavior of the squadron commander. As the movie progresses, EDI becomes increasingly cocky, brash, and has a strong desire to take care of the majority of the missions himself even when told not to. These are characteristics very clearly portrayed by the commander of the squadron.
Much like the science fiction community, many scientists have wholly accepted the idea that robots can, in some capacity, learn. One of the common methods of a machine learning is making use of what is called a "neural network". The neural network has been around since the 40's, and the concept was derived from the infrastructure of the human brain. The basic idea behind a neural network is that it is shown or given a set of data to train (learn) on. These data sets are given to the network WITH the solution that came from that data set. The network takes the given data and weights each individual data input such that it gets the intended solution. This is done multiple times, the amount of learning cycles being very dependent on the complexity of the problem and the amount of input variables. After the network has reached a point where it is considered sufficiently trained, it is then tested. It is given data sets without a solution, it then outputs a solution, and the solution generated by the network is compared to the actual solution. It is not uncommon for a network to be trained to within 5% error.
There are other types of learning present in todays robotics and computers. There are even cases of four legged robots teaching themselves to walk. This creature is equipped with sensors that can determine the momentum of its body as well as each of its limbs. In essence, it can feel what its body and legs are doing. It then takes all of the data it gets from these sensors runs them through a simulation that outputs what is called a walking gait. The more the robot walks, the more data its sensors gather, and the more refined the walking gait becomes. It can clearly be seen in the video that at first the four legged robot (that looks all too eerily like a replicator) is very unstable and unsure of its footing, then very shortly afterward, it is efficiently charging towards the camera in a seemingly very stable fashion.
Another existing prototype learning robot is made by Barrett Technology and uses their advanced robotic WAM arm. In the experiment shown here, the robot arm first goes through "kinesthetic teaching" where a person flips the pancake in the pan while the robot is holding it. The robot perceives these movements and uses them as a basis to start its own attempts at flipping the pancake. As can be seen in the video, it takes the robot a fair amount of time (50 attempts) to master this rather basic motion. Regardless of how long it takes though, this is a clear portrayal of a robot learning a task.
There are a myraid of other examples of robots or computers learning in some form, we've even reached the point where robots have learned to more efficiently create themselves. This shouldn't come as a surprise because by the definition of learning we noted earlier, learning is merely "gaining knowledge or understanding of or skill in by study, instruction, or experience". Studying is the act of gleaning information from some medium, whether it is reading a book or observing a situation, and machines are already capable of this. Instruction is what computers and machines were built around, they only were created because we wanted to instruct them to do things, so they are capable of learning by instruction. Learning by experience, while having come along more recently than the other two methods of learning, is also something that todays machines and robots are capable of. The examples of the walking replicator and the pancake flipping robot show this perfectly. Both of these robots learn from their experiences of attempting the task they are trying to learn, and use that to learn how to more effectively perform the task.
With all of the time, money, and research going in to making robots capable of learning and people posing the question "Will robots ever be able to learn like people do?", Plato brings up an idea that counters that question with another question: "Do humans truly learn?". The quest that these robotics researchers have set out on is to raise the capability of computer brains to the level of human beings (in regards to learning, imagination is a different realm of discussion entirely). Many people think there are certian nuances about how a human being learns that cannot be replicated by a microprocessor. However, if Plato's quote rings true and human beings really don't learn, and are only beings of recollection, why can't this be replicated by a computer? A computer is essentially a recollection machine and it was stated earlier that the computer's capacity for memory far exceeds that of a human brain. So, with this in mind, is attemping to create a "learning" robot really that lofty of a goal? If you strip human learning of its supposed nuances, it becomes a task that is completely attainable by computers. As technology progresses we will see an increasingly smaller gap between the learning capabilities of robots/computers, and we may possibly even see them overtake us. The list of things people can do that robots could never do is growing smaller by the decade...