The Future
As we progress technologically, there is an increased focus on machine learning. That is a machine that is not necessarily programmed with the algorithm to solve a problem, but one that has been programmed with the ability to take a given set of input, classify it, and generate an answer. This is achieved via supervised, unsupervised, or reinforced learning techniques. My background in this field is creating neural nets and I am more that a bit rusty so for more on topic of machine learning see wikipedia. I believe machine learning to be only half of the equation to achieve true AI.
Achieving True AI
Machine learning still requires a human to create the learning data set, expected results, and training algorithm. This is an incomplete view of what is necessary to achieve true artificial intelligence.
- We must teach machines how to learn a new general task using its current capabilities and apply that to future tasks.
- We must teach machines how to incorporate new capabilities and algorithms into themselves.
- We must teach machines a process to learn new tasks with little or no user input.
- Above all, we must teach machines how to be curious.
The Curiosity Engine
As living, breathing creatures it is our curiosity that drives us to learn new things. It is how we learn to do anything new from walk to drive to fill-in-the-blank. I believe that Machine Curiosity is critical to the future of AI. Once we figure it out, the possibilities are endless. So this raises a good question, how do we teach machines to be curious? I don’t think that anyone really knows the answer to that question. Here, however, are some possibilities.
- Mimicking – While mimicking is a learning method, maybe it can help with curiosity too. The machine could analyze behavior, identify behavior that it doesn’t know, then attempt to mimic the behavior. This could be considered a crude form of machine curiosity by itself.
- Chaos (random combinations of known abilities) – Another crude form of curiosity. The machine simply pairs capabilities in the attempt to do something new. This is brute force, and might take along time to create useful abilities.
- Observation – The machine selects observed capabilities and applies its current capabilities to it in order to learn about the expected results. The machine would need a way to create new analytical capabilities.
- … Many many more methods …
- A mix of the above – All of the above can be combined into one curiosity engine. This method has the potential to be the most efficient since it will make intelligent use of all other methods.
