Neural Networks and AI (2007) Has AI really advanced much?

 

It is widely held that the human brain is the most sophisticated object in the universe, but is it really? Early this century, guests on a radio show entered a dating contest. They had to chat with a random strangers on the phone and pick their favorite to be their dates. To their complete shock, all the guests had ended up picking someone named Alice. No, not the person, but Artificial Linguistic Internet Computer Entity. 

In the immediate aftermath, thousands of people on AOL Instant Messenger found out that their best friends were actually bots all along, much to their sadness. How can any of this be possible? The answer lies in understanding the structure of the human brain and using computers to imitate it. The brain has been found to be structurally very simple, being just a huge number of synaptic connections and their respective connection strengths. These are called neural networks because the brain is a network of neurons, while the computer stimulation of the brain is called artificial neural networks. The way they operate was best stated by Sirosh in 1949: “Neurons that fire together, wire together. As a result of this process, connections that are required are retained and others are eliminated." It is easy to implement this simple idea in a computer program, but the results were beyond expectations. 

The programs learned to speak English on their own from listening to nearby humans talk. At first they spoke simple short sentences with limited vocabulary, and sounded like a real human baby. And just like a real baby, they started seeing the patterns and following rigid rules incorrectly, to make up an example such as mispronouncing Carlisle as “CAR-LIE-SOLE”. Eventually after more listening sessions, they recorrected themselves to speak English like a real adult, and even beat people in chess, which they also learned on their own. Scientists did many similar experiments with artificial neural networks, and cracked open their “skulls” to witness the sophisticated connections that they formed, which to this day nobody understands. But these practical applications of neural networks pales in comparison to how its very existence can change the way we see the universe. It is this mirror image of our own human self that shatters our existing worldview, bringing all manner of questions about the nature of our identity, of knowledge, of space and time, of mind, matter, and soul. The existence of artificial neural networks encourage those in the know-how to envision the physical and metaphysical worlds as merely two sides of the same coin: as a construct of our mental associations.

Neural networks leads us to envision reality as made up of associations over individual properties, and these associations would recreate reality as a mental construct. This stands in direct contrast to Plato’s theory of forms, in which all objects have individual properties mysteriously attributed to them from a "Platonic heaven". Here an object exists in their own right, and are not a creation of our minds or cultures (cite lecture4). Plato uses the example of justice, which can be striped down into its “essence” (cite The Republic). However, in neural networks, all knowledge is stored in the form of synaptic connections, and nothing else. You see a banana, and it’s a banana only because it is associated via synaptic connections to the abstract idea of food, yellow peels, fruity, yummy, and etcetera, and in turn all these ideas to connected to yet other ideas. In fact these are ideas only because they are described by other ideas. It tells us that it would be impossible to understand what a banana is without associating it with everything else. There is no essence, only associations. Thus, knowledge would be understood to be made up of relations. In the same vein, people would be identified by the ideas associated with them, i.e. “friend”. This is in line with the Eastern emphasis on personal connections over individualism.

Discrete bits of ideas are emphasized to make up the whole just as in how in neural networks all big ideas are composed of smaller and smaller ideas. If we imagine a river, a flowing current of trillions of trillions of water molecules flowing by in front of us in a smooth continuous way. But there is only one river. It encourages us to look deeper into our ideas and categorize them into its individual components. For example, we can think of our country as an entity in and of itself, or as made up of its individuals. If the government one day orders everyone to sacrifice themselves as suicide bombers for the sake of the country, we would be encouraged to ponder what we’re fighting for if we’re the ones who are what creates the identity of the country. This path of thinking would be lead to believe that the government’s orders are rather shallow.

Neural network’s description of reality as made up of a myriad of connections is unfriendly toward the truthfulness of the physical world. The network can only process the data that enters from the senses. For the human brain, this would be taste, smell, touch, sight, hearing, and acceleration. Thus material objects do not exist to us unless we perceive them so that the experience would enter the network. Reality would be interpreted as a mere construct of our imaginations fueled by our senses. This has the underlining assumption that our total experience is derived solely from our physical brain at the receiving end. In contrast, René Descartes theorized substance dualism, stating that mind and body is separate due to the body being physical while the mind is beyond the physical realm. Here the mind would be the soul rather than the brain, but neural networks were not available during his time, so we can forgive him.

Neural network’s deterministic description of our experience of reality is unfriendly toward the truthfulness of the metaphysical world. Artificial neural networks runs via a strict set of mathematical equations, so all the processing is set on a predetermined path just like any other computer program. The logic is “if it works for computers, it can also work for us”. Thus, our experience of reality would be completely accounted for and explained from a cause and effect perspective; thus there would be no need for metaphysical forces to come into play. There is no need for a God or a “Platonic heaven” to keep things running right.

Neural networks lead us to believe that change is ever-present, that nothing can persist. Are we the same person we were yesterday? Most people would think so, but the existence of neural networks cast this into doubt. Consider that in an actual brain, the individual components of every cell would be replaced within a month, so we would literally have a different brain by that time. Neural networks take this identity crisis a step further, because every ounce of experience or thought changes the connection of the network. Even repeated thoughts would readjust the weights of the synaptic connections. Without a permanent body, our identity would solely be made up of our brain’s network connections, and thus our identities would change every second. This leads to the belief that since every moment we become a new person, we don’t have to behave according to what we believe ourselves to be, but can be whatever we want to be!

Free will is emphasized over determinism. Unlike traditional computer programs, neural networks learn by example and they cannot be programmed to perform a specific task. Traditional computer programs are 100% predictable since they follow step-by-step instructions, while neural networks are utterly unpredictable as to how it chooses to solve its problems. They can do anything a traditional program can do, while there are many things it can do that a traditional program cannot do. The traditional program would have a pre-determined result, whereas neural networks have a will of its own. Since our own brains have the same structure as artificial neural networks, it would follow that we would have the same freedom of will. The success of neural networks in solving problems would also give the view that learning from experience is superior to rote learning.

The existence of neural networks emphasizes the importance of internal mental events over external ones, since our physical and metaphysical realities are just a product of our mental constructs. We can envision life as a series of internal affairs, so it doesn’t matter what we do in the outside world so long as we’re happy inside. There’s no point in following a code of ethics, since we all die anyway.