Wow, what a hectic week full of changes and concern. First of all, I received Justin's feedback for my essay and it caused major concerns to both him and myself. My paper became more of an info dump rather than supporting my thesis and when I did compare the two, it rarely was on the same topic.
While I did expect some major comments and feedback, the extent of the comments were definitely unexpected. I made my paper an immediate priority for the entire week, just to make changes and adjust the structure accordingly. I went through numerous drafts and structures just to find one that seemed like it resolved the comments I had received. I cut down on unnecessary info to avoid my paper being an info dump, and structured it in a way to first inform the reader, and then compare the features of each. In doing this, I had to sacrifice some technical parts of the paper in order to get a better understanding of each in light of the other. I made sure to tie all my paragraphs back to the thesis and the performance of each. I added a comparison paragraph after each section to tie both systems back to the thesis. I also made sure to include the importance of certain statements and define terms when necessary. I included headers to make sure it was clear what was being discussed, which will hopefully help with understanding the structure of the paper.
The ultimate goal of researchers in the field of Artificial Intelligence is to create intelligent machines that behave like humans. This entails learning, problem-solving, speech recognition, among many others. By understanding how artificial machines or models perform different sets of tasks, researchers begin to understand how the human brain works. Chris Eliasmith and a group of neuroscientists at the University of Waterloo claim to have built the world’s most complex large-scale model of the human brain. As Eliasmith states, understanding the brain is important because it “lets us understand how the brain, the biological substrate, and behavior related. That’s important for all sorts of health applications. In testing [his model], he has ‘killed’ synthetic neurons and watched performance degrade, which could provide an interesting insight into natural aging and degenerative disorders.” Eliasmith’s model, called Spaun, consists of millions of simulated neurons and performs eight different tasks. Its goal is to mimic human behavior and the physiology that underlies the brain. Much similar to a human, Spaun receives visual input, processes the input, and outputs a result through a specific type of learning, known as Reinforcement Learning. This type of learning is similar to the learning that goes on in the brain as well as other artificial models. One such model is DeepMind’s AI, which was created by Google to play Atari 2600 games.
The comparison between the two is interesting because results from studies show that Spaun does not outperform the human brain while DeepMind’s AI does. This raises the question as to why Spaun, a model that is meant to mimic human behavior, simulate the human brain, and provide future applications to the field of medicine, underperforms the human brain and DeepMind’s AI. Using various pieces of literature and research papers, I will first present background information on both Spaun and relate it to its underperformance. Then, I will provide background information on DeepMind’s AI to familiarize the reader with its performance. Lastly, I will explore the limitations of Spaun and compare them to similar components of DeepMind’s AI. In this paper, I argue the following: Spaun underperforms compared to the human brain and DeepMind’s AI because of numerous constraints caused by its great goal of mimicking human behavior: lack of neurons and limited semantics, the variability of neural spiking, and no effective training methods. Most constraints are caused by the insufficient computation power to support the addition of more neurons. The lack of more neurons hinders the accuracy and performance of Spaun. On the other hand, DeepMind’s goal is less demanding with fewer constraints, allowing it to overcome each obstacle and simply solve for a function in the best way possible.
- Unknown Type