Love a good discussion, it's why I joined gaf after all these years lol
Ahh ok I understand your thinking, that to operate on its own, free will if you like, even AI would need some kind of emotional under-pinning to its reasoning/logic? that's true. But we are set with biological parameters that we will very much influence or decisions from birth (eat, sleep, air etc). It's the same with machines they will always be "born" with a set of parameters that will influence it's decision. The difference is the emotional aspect you refer to and I think when people start talking in terms of good, evil, etc we loose sight of what AI is. AI is and never going to do something good or bad it will just do so yeah I guess i do agree with parts of what you're saying. To answer your point specifically about non-prescribed decision I think your looking at AI wrong, even the most advanced AI we will ever create will always technically be a prescribed decision but when you have 100 million possibilities/outcomes from one scenario it would feel a lot less predefined and a lot more like free will. But no AI will never be able to make it’s own non prescribed decisions. Think of AI as point with nodes that connect to another point with more nodes each node representing a distinction/option that will lead to a different outcome. Because AI as it stands, is somewhat limited by the amount of options it can choose from, AI choices therefore seem predictable and still very artificial, games are a very good example of this limited node structure (e.g. you as an NPC character a question you may get three or four different anwers but are limited to just that).
Here is an example of Netflix and their recommendation engine in graph form, all those lines represent different decisions which connect to another node (date base) where the algorithm can then make another choice and arrive at another node so and so on.
So to go back on point of AI seeming more realistic, life like and seemingly having free will is not quite yet there because we simply don’t have the infrastructure to support that level of AI yet. The choices are limited for two reasons first being the amount of data it would need to make a choice is cannot be comprehended with the way our infrastructure is set up (I mean where would you store all of those nodes/data bases to represent or give the illusion of what would seem almost an infinite amount of choices) and processing power, to run even simple k means clustering algorithm on what would be hundreds and hundreds of zettabytes of data would probably take an incredibly long time and we are not just connected enough as a world. So yeah processing power and data is what has previously been holding us back really making any real strides with AI until recently.
I cant remember the stat exactly but 60% of electronic data has been produced in the past 5 years, as we become more and more connected with phones, ipads, laptops we are producing an incredible amount of data (a lot through social media) and currently entering phase 2, SMART TV’s, SMART homes, SMART Cars essientially we a moving to the age of Internet of Things, where even every day normal appliances/objects will become or have the ability to be connected to the internet with each device representing it own little node. The amount of data we are going to produce in the next five years is shocking and where there is data know that there will be algorithms tracking your behaviour, learning about your habits, do you wash at 60 or 40, what route do you take work, how much heat do you use in your house etc. everything we will do is constantly being tracked, clustered, analyses and mined. Another interesting example is a company I don’t want to mention by name but they are one of the companies behind the biggest rival to Siri. Did you know when you ask your phone something and it doesn’t have an answer to the question is stored, if enough people ask this question its flagged by the algorithm to production team who then develop a response to said question (not that clever but more just to show you that everything is recorded). But where things get interesting when (apparently in 10 years or so) we move into phase 3 of the cyber age you can read more about it here https://www.cisco.com/web/about/ac79/docs/innov/IoE.pdf where EVERYTHING is connected to the internet (including us humans), this is when I think you will get the level of intelligence you speak of, literally the outcome of any given scenario would be in the billions nothing would seem predetermined although, because technically it’s still subscribing to a grid set up.
In terms of processing power it would take to run algorithms on this scale, it is why GPU Compute is being made such a big deal issue of. The amount of processing power you can achieve with GPU compute is what allow algorithms on this scale to work. Again I use Netflix as an example here with what they are doing with an incredibly modest set up (four GTX680’s):
http://www.enterprisetech.com/2014/02/11/netflix-speeds-machine-learning-amazon-gpus/
As you can see I think they have many or similar problems to what I’m sure game developers are experiencing (speed not being the issue but more copying/syncing data and latancy in communicating) but there slowly learning how to manage that so the speed of the GPU compute offsets the balance. It’s also why I laugh when people I people tell me that the ps4 is an under powered and GPU Compute wont make that much of difference (sorry for that little fan boy shot, couldn’t help it lol)
Sorry a very long winded post and I may have totally detracted from the point but I hope that gives you alittle more insight.
In my opinion if you want to see really see how advance AI has become look at what a firm called Palentir are doing, you may have to do some digging around
Edit: I just want to add, alot of what is theoretical and there are sooo many issues we will have to deal with before we ever get to the state of AI before someone comes in spouting "its not possible because x y z" what I posted is my opinion based on facts
