Tech moves fast! Stay ahead of the curve with Techopedia!
Join nearly 200,000 subscribers who receive actionable tech insights from Techopedia.
The value learning problem is a specific fundamental issue in the development of machine learning and artificial intelligence technologies that addresses the difference between humans and computers, and the ways that they think.
In a nutshell, the value learning problem is based on how hard it is for computers to figure out what to "value" (in terms of both data and policy) and how to act in a machine learning network, and how programmers can optimize how the program acts to match their original intentions when they created it.
Key to the value learning problem is that it's extremely important for programmers to be able to make machine learning programs that carry out the intended values. However, the catch-22 is that the values can't be explicitly stated in ways that hinder the program learning itself.
People sometimes talk about the ‘convergence’ of machine learning technologies as the successful focusing on value data, but the value learning problem is in some ways a little different. It's the idea that there has to be some core way to show the machine learning program what's desired, rather than just spelling it out, which is a deterministic way of running ML.
For example, take this paper on the value learning problem that suggests machine learning programs could have a storage set of inputs showing positive human responses to stimuli. In reading these types of addresses to the value learning problem, it becomes clear that there's a major gap in machine learning that's not easy to fix – essentially – how do people create machines that can really think like people? Another way to explain this is that the value learning problem goes to the heart of how we think as humans, and how our thoughts are not always based on rote input.
For computers to model our intuition, our instinct, our social inclinations and our deepest ethical values is a tall order, even when computers can learn to play chess in a human way, or outpace us in solving difficult math problems. Professionals can expect the value learning program to continue to be central in the development of machine learning technologies.