Artificial intelligence is a buzzword that seems to appear in the news almost every day. We hear about new computers and software that can beat the world’s premier go players, or can make beautiful works of art. However, we are nowhere near what I consider to be the holy grail of AI, and even computing as a whole, general artificial intelligence. Today, we have thousands of programs that can all perform incredibly well at one task, but that's it. I want to solve the problem of general artificial intelligence, in order to create a computer that performs proficiently at every task. Such a system would be the absolute peak of human achievement, remarkably changing every field of human endeavor, from medicine to agriculture.
The question remains, is such a goal even possible? I believe that general artificial intelligence is not only possible, but that we will achieve it in our lifetime. However, it is not enough to simply put together thousands of tradition computer programs that excel in a single discipline, such as image recognition, or chess strategy. Creating such a system would result in an unscalable mess that could never compete with the heights of human reasoning. In order to realize the true potential of computing, I believe that we need to mimic God in his creation of man. Huge strides are being made in the field of neural networks, and training models that mimic the functionality of a human brain. Currently, the largest neural networks ever constructed can simulate roughly 16 million neurons, which is remarkable, but falls short of the human brain’s jaw dropping 80 billion neurons by a factor of 5000. Furthermore, these neural networks can only be trained to perform a singular task, a far cry being able to handle my goal of general artificial intelligence. So how do we overcome this challenge? Looking at the issue through the admittedly simple lens of Moore’s Law, we can expect our computing power to increase 5000x roughly in the next 25 years. However, the issue is not as simple as computing power, as Moore’s law begins to slow in the near future, even stopping when we reach the theoretical limit of transistor density, and heat dissipation on computer boards. If I were to venture an admittedly uneducated guess, I believe that the future of AI, and likewise my dream of general artificial intelligence, lies in our ability to implement neural networks in hardware, not software. Today’s largest neural networks are all simulated in software, on traditional supercomputers. While these computers do exceedingly well performing traditional, logical, computing tasks, their traditional digital architectures aren’t always best suited to the inherently analog processes of a true neural network. The optimum system, I believe, would be implemented in analog circuitry but would act in digital-analog symbiosis with a traditional system for training. This idea is not completely unfounded. Even today, there are a number of chips optimized specifically for neural network processing including some work at big name companies such as Facebook, Amazon, Google, and Baidu. We can expect to see more of these chips in the future. I believe such innovations are absolutely crucial to the development of general artificial intelligence, and likewise the full potential of computing.
So why is general artificial intelligence so damn important? I strongly believe that the creation of such a system would be the pinnacle of human achievement, affecting everyone and everything with its scope. Such a device, that could perform human reasoning and cognition, with the electronic speed and intelligence of a computer would bring about massive changes in almost every field, surely raising the standard of living worldwide. Just as God created man, so will we create general artificial intelligence.