Re: Artificial Intelligence: How far is it? Quote:
Originally Posted by ashokrajagopal My post above was not about limiting learning or limiting experiences. It is only about limiting the objectives. In fact the machine has no objective to learn unless you provide one. It is not "free" to learn. It is programmed to learn, so it behaves as if it is free to learn.
The same way as it is programmed to learn, it can also be programmed not to harm. Or it is not programmed to harm.
There is so much of science fantasy fiction out there that separates the software and hardware and expands on that premise. This is also one reason why I call it machine all the time. There are some weird fictions where somebody writes a program and the program begins to gain a physical form. That is just religious mythology manifesting in science fiction garb. I guess Ultron came about like that.
Software has no existence on itself. |
Take the example of machine learning - Predictive text input software that we have in our smartphones. It not only learns the usual words that I type, but also suggests the next word and thereby the full sentence, by learning the usual patterns of words that I write. For example, I ping my wife everyday when I am leaving office for home. it goes like this - Leaving ofc, ping me what to buy home. As soon as I write "lea", it suggests the whole sentence progressively. And that is a basic predictive software, not an AI.
Agree, that Learning as a standalone objective cannot be the sole objective. It has to be specified further - what to learn? That alone feels like a good fail-safe bet against rouge AI. But we need to understand one more thing - softwares are rarely personalized, they are generic. Android software is same across billions of smartphones, it learns each individual's usage pattern though.
Google's database would know much more about us than our own near ones. That database is a big information repository, it known when my credit card payment is due, it knows when I should leave home to catch my train whose tickets I had booked last month, it knows (and auto-fills) my passwords. We rely on these software as it makes our life easy. We don't need to remember things, google will remember for us and notify us when it is due.
Have you ever encountered a case when you want your software to recall or furnish some information and it says "the software is out dated, please update now to use it further"? How frantically we go on to update it then...
AI doomsday scenarios are much more related to defence (and there, it does pose a serious issue in future). For example, there are softwares that scan private information, phone calls, emails of individuals for words/patterns that suggests that the person may be a national security threat; there are softwares that scan the skies for areal threats, there are softwares embedded in missiles enabling it to independently identify, target and fire towards an enemy vessel/aircraft/missile/individuals.
The objective of different softwares here is different - identify potential threats, identify location, lock the target and fire to kill. If all these softwares one day start interacting with each other... |