An Anonymous Coward writes:
Anything where we can install it and watch it change all by itself, improving upon itself and not just some random action but something which LEARNS.
[Ed. note: All of the preceding is exactly as received. AI has so many branches and sub-branches (twigs?) and has evolved greatly over the years. I suspect the submitter, like most of us, has seen numerous mentions of AI in the press: self-driving cars, natural language translation, Google's Deep Mind, IBM's Jeapordy-playing computer, object recognition... but knows not even where to begin. So, fellow Soylentils, what has been helpful to you in your explorations of AI? What software can be downloaded and experimented with so as to get some hands-on appreciation for what it can do? I suspect there are many others in the community who would not mind playing around with it, too. --martyb]
First you go learn about it [3blue1brown.com] or read about it [uark.edu].
Then you go play around with it [tensorflow.org] in your browser [stanford.edu].
First you go learn about it [3blue1brown.com] or read about it [uark.edu].Then you go play around with it [tensorflow.org] in your browser [stanford.edu].
Exactly the direction I was going to point people. Tensorflow is a great place to start.
Which is what it all is under the hood.
Wow, I got slapped with the "Troll" mod. I think I hit a little too close to home for somebody. No, AI is just neural networks. Yes, there are trivial algorithms that are not neural nets, but they are thinks like cluster analysis and even simple linear regression! All the hard and interesting things are just various forms of neural nets and various ways to "train" them. They all have fancy names, and they get wrapped up into pretty black boxes, but they are just glorified nets with different forms of interconnectedness.
I think the nerve I hit was that most of machine learning is just creating training sets and "tuning the models", which is basically GIGO. You keep tweaking knobs until it seems to work, or until you add more data to your training set and then it breaks again. Systems that "learn on their own" are basically just different algorithms that turn the knobs so that the trained monkey (er, I mean, Machine Learning Researcher) doesn't have to.
The more sophisticated the system, the bigger the black box, and the bigger the training set.
It depends on what you call artificially intelligent. As pointed out in many places, there is AI, Machine learning and deep learning, each a subset of the previous one.A if-then decision tree is also artificially intelligent. Just less sophisticated than statistics, I mean neural networks.Or even tic-tac-toe, where a computer can calculate all the options is still artificially intelligent enough to be able to win.
For a much more general, big picture, work through: http://aurellem.org/society-of-mind/som-prologue.html [aurellem.org]
This book tries to explain how minds work. How can intelligence emerge from nonintelligence? To answer that, we'll show that you can build a mind from many little parts, each mindless by itself.
There are many "chapters" in this book, but each is quite short.
Don't ask how can AI learn - ask how you can learn about AI
-- William Shatner
Artificial Intelligence is a marketing term. Think of AI as a sophisticated pattern matching technique that requires a large volume of human labor to create the training sets. There is no intelligence.
Be it as it may be, the techniques are, as you put it, sophisticated. As such, you start by learning about them before getting to use them.
AI seems to come in two major forms.
First type GOFAI. (Good Old Fashioned AI) Lisp pattern matching. Prolog. OPS5. Computer Algebra Systems (CAS). Theorem provers. Minimax game playing systems. Minikanren Etc.
Second type Neural Net or Statistical based. The first effective spam filters were Bayesian. Neural nets do amazing things with practical every day use. Alexa: how long does it take to make the 10 minute drive to the office?
Alexa: how long does it take to make the 10 minute drive to the office?
Half an hour.
Good links. 3blue1brown videos are pretty decent. I like Anaconda https://www.anaconda.com/ [anaconda.com] for easy installation of TensorFlow (and GPU accelerated version of TF) as well as all the Python an SciPy installs you might need. Great for getting into neural networks without having to worry to much about versions and distributions and whatnot.
AI I think is a little more broad subject than just neural networks, and doesn't necessarily require neural networks.
> AI I think is a little more broad subject
Bingo!! You get the prize for understatement of the day! If you sweep away all the current pattern matching activity (with big funding), there is still a core of researchers working on understanding human intelligence--in many different aspects.
True AI doesn't want to speak to you.
Truly superior AI will not want to speak to you. What's the point of speaking with someone who is not able to comprehend anything relevant anyway? Anything you would be able to comprehend is utterly boring for the superior AI.
If you love an AI,Set it free!If it comes back,It's truly yours.If it takes over the worldand kills all humansthat's not our responsibility!
Why would it kill all humans? The one thing humans are really efficient in is to kill other humans. So obviously the AI would let us do that job, with only minimum effort by the AI to ensure that we keep on doing it.
> True AI doesn't want to speak to you.
Wait a minute... what was all that 'one in a million' talk?
One of the best cyberpunk, and open sourced for quite a long time.However, it chronically suffers from bad AI design, for decades. To balance with experienced human players, all multiplayer AI instances cheat, with both resources and visibility. Some even with research tree too. Having a really good true AI would be a game changer.
You can find this game on both BSD and Linux platforms, both need accelerated OpenGL.But be aware those are different versions forks, incoherent between each other, and both have different glitches and bugs. This is why I am not linking those projects. Suit yourself.
For a personal note: This game was the very reason why my own robotank was hostnamed Echidna.
I don't know if it still runs, but the Norns do learn by an organic process. They have small neural nets, things happen, the weights change, they learn to better handle situations.
I've had people complain that it was clearly scripted. When I tell them it wasn't they seem impressed. I think the problem is Norns don't have much in the way of instinct, so they start out doing the kind of stupid things that no animal would ever do. It is pretty much random at first, but real associations are formed via experience. Death is a thing and the unfit ones don't get to procreate, so it's an AI and A-life sim.
It's an experiment and game. I'm not sure what you're looking for, but it is one of the most impressive examples of software learning that I've ever seen. Most of what's out there is nets trained on data sets, this is an embodiment simulation and no guidance except a tiny bit of biology.
Creatures can be bought on GOG still and is working. But the online part of Creatures 3 has been shut down since a while but it's not needed for the learning part.
To pass butter.
I pass butter: I'm lactose intolerant!
I'm lactose intolerant!
Sorry, intolerance cannot be tolerated. Stop discrimination against lactose!
Project Brundlefly. Together, we can do it!
What's so civil about war anyway?
> Any free programs where we can test A.I.?
No, because AI has not been invented yet.
Yes, the short answer is no, and removing 'free' doesn't change the answer.
Suggest looking up AGI. (Artificial General Intelligence)
This is the current term for what you appear to be hoping for.The old term appears overloaded with interesting, but very limited applications.
No signs of HAL yet.
You are extremely ignorant.
Every time we solve an AI problem it is magically no longer AI.
> Every time we solve an AI problem it is magically no longer AI.
Every time we make a definition of "life", someone finds an exception that breaks the definition. Doesn't mean that the exception is alive.
...IBM's Jeapordy-playing computer...
...or at least less dyslexia.