Top-down vs Bottom-up Design
The Top-Down approach
There a several approaches to achieve AI, even classifications on what particular kind of AI it is. Artificial intelligence can be broken down into several different types of systems such as : Analytical AI (Narrow AI), Artificial General Intelligence (AGI), & Super Intelligence just to name a few. With that being said there're several different approaches one may take to achieve a particular type of AI. Ideally a Super Intelligence AI system will comprise the characteristics of a AGI and Narrow AI. This would mean that a super intelligent AI system will contain cognitive, emotional, social intelligence and also be self-aware. Well I guess the only question now is where do we start?
In essence the top down approach to developing AI consists of a plethora of preset methods and goals that they want the automated intelligence to achieve. Essentially the top down approach will develop a system that has pre programmed definitions on cognitive responses analytical intelligence. The system of the approach would be to go from the top down and essentially add detail into every layer of abstraction. This is proving to be difficult for researchers being that there is no direct path to creating human consciousness how can we program understanding, love, pain, sadness? These are the struggles of trying to create an artificially intelligent being that has a million pre programmed methods to stimulate its cognition. This gave birth to a new thought of the Bottom-up design.
Bottom-up approach
Essentially the bottom up approach consisting of the construction of neural networks is the complete opposite of the top-down model. Naturally there'll be program builds that stimulate learning and “commonsense”. The bottom up approach will consist of simple methods and systems that will grow and slowly become more complex. Most of the simple systems are linked to a larger subsystem that encompasses all learned knowledge. The main issue with the bottom up approach is that it will take a long time for the computers to learn appropriate responses to any given situation. We're so desperate to improve the overall quality of life so the bottom up approach doesn't seem plausible if we want to find out the secret to AI within our lifetime.
They both win
By using the program methods the from the top-down approach and incorporating learned knowledge from the bottom up approach you can successfully build an artificially intelligent system that can operate on a cognitive level comparative to that of a human being. Essentially the perfect artificially intelligent beings will have to use the top down and the bottom up approach in tandem. Keep in mind there are certain things you can't teach a top down AI system such as the feeling of loss. Whereas at the bottom up approach you can teach the system to react in anyway that's appropriate for them based on their experiences.
Pop Culture
In Westworld..
In HBO’s Westworld we made it, we figured it out, the secret to artificial intelligence and what do we do with it? We create a place where people can go and “enjoy” everything in life has to offer. Westworld is occupied by inhabitants known as hosts. These hosts are artificial beings constructed by the top down approach. There are bunch of preset methods and responses that they are preprogrammed to respond to and anything that isn't programmed is improvised through other pre-made methods. The essential goal of the creator of Westworld is to have these beings find their way to consciousness. I guess you could say it's a top-down approach taking some insights from the bottom up approach mainly because the heavy influence learning has in the development — or rather in the process of achieving consciousness. Everything the host experiences in Westworld is saved into something called a reverie these reveries store data , instances or situations the host may encounter. In order to stay true to the top-down approach the reveries are the actual layer of detail added into their cognitive abstractions.
In Sword Art Online..
Sword Art Online (SAO) is an anime series in which a virtual reality game called SAO is developed and players are able to control their avatars with their bodies. Within the first season there're many artificially intelligent beings stationed in the game referred to as NPC’s. NPC's would give players quest, items, or give them information about the area in town that they were traveling through. But in some special cases the NPCs would give players quests based on what they're wearing or what they were feeling at the time this would mean that the artificially intelligent being was capable of “reading the room” and making a executive decision based on its own perception. However that's not all The creator the game Akihiko Kayaba had the intention all along to do research on his players. Keeping track of how players would react in any given situation created an ever improving cognitive response function within the AIs. His ideals even transcend his death when another company takes over.
In comes Alice, an artificial human living in a virtual world. A company called Rath is trying to develop AI soldiers for the Japanese army. The CEO is a firm believer in the bottom up approach, he insists that the bottom up approach is the most plausible way to actually achieve perfect AI. The only issue is the time that it would take to learn all this knowledge. By integrating his AI into the system that Akihiko Kayaba created. The CEO of Rath was able to place his AI in a virtual world. In this world they will progress, grow and learn just like humans do in the real world. In order to make better use of the bottom up approach time is accelerated in this world. One day in the real world can easily be over a few years and the virtual world, to the inhabitants of the virtual world times seems to be flowing normally. However just like with a top-down approach the bottom up approach has encountered issues. The artificial beings living in the virtual world have created a system of rules which they're unwilling to break. But in our natural life can we think of instances where it's okay to break a the law for the sake of doing something good? How are we able to teach artificially intelligent beings about this phenomena unless they've been experiencing it since birth. Which is why Rath has created clones of actual newborn babies as the base framework for the AI beings in his virtual world. I know this sounds like something drafted out of a comic book but the ideas here are pretty cool in my opinion. With a framework like this there are no real limitations to what we can or can't do the only thing that we would have to do is just wait.
Conclusion
I'm interested to see how the development of AI will progresses over the next decade or so. I love fiction because it opens the door for new ways of thinking. When I'm presented with a new idea in a television series or movie I often correlate it to what I know about the real world, questioning my own ideas and beliefs. Although these movies and shows are merely for entertainment there’s so much we don't know about AI ,the dangers they might bring, the prosperity in the quality of life of those living on earth. I will continue to delve deeper into my understanding of artificial intelligence. I firmly believe that understanding and creating more efficient AI can improve the overall quality of life on earth rather than be a detriment to it. I hope you found this entertaining I was just spitting out ideas. Hopefully they provoke thought or lead to a breakthrough.
Ciao!