I want to build my own AI. Where should I start?

Hello, World!

>>8576970

I know python. Is this a good language to star writing my AI?

File: 11713640484_3096608d9f_b.jpg (208KB, 683x1024px) Image search: [Google]

208KB, 683x1024px

>>8576973
if you have to write AI then is it really AI?

>>8576979
So we grow it from an automatically self-improving seed until it realizes that it's just AI?

>>8576969
Adopt a kid.

You can use Tensorflow to train an AI chatbot with machine learning, and TF is in python, but it is somewhat difficult. Look at the Seq2Seq tutorial, neural nets, etc. and see what you can make of it. TF is CPU intensive-my ancient desktop was maxed out for a few hours training on just a few hundred mbs of prompt response pairs, and it is not entry level stuff but it is better than trying to reinvent the wheel from scratch if ML is your idea of AI.

>>8577407
Best answer. Also try Keras

File: nsa_eye.jpg (46KB, 500x350px) Image search: [Google]

46KB, 500x350px

>>8576969

Maybe build a bot-net that simulates human consciousness/conversations that eventually wakes up.

>>8576973
use python and google to make an irc chatbot. there, you have an AI

going to go against the grain and suggest that you do not study neural networks, at least not for a while. they may be very popular, but there are many unanswered questions about them. you would probably end up learning more by studying other ML techniques first, and then moving on to neural networks after you have a better foundation.

File: CeePlusPlus.jpg (105KB, 603x324px) Image search: [Google]

105KB, 603x324px

>>8576969
Im doing this right now, start with Calculus and move to Linear Algebra, then consider Python crash course, after that get for couple months into Data Science (yeah i know... ) this will lay the foundation to Neural Networks.

After that, look into stuff like this.
https://iamtrask.github.io/2015/07/12/basic-python-network/

And then search about Convolution Neural Networks.

This will get you going in no time, if after 2-3 months of work you can create something this simple by heart and even improve it or optimize it, and you are good with the math, you are ready to move on.

If you want your AI to do image search and CV general, consider OpenCV with Python or C++ (in sense of speed, the python act as wrapper to the C++ libs and code, so its mostly preference).

Otherwise, if you want to do other stuff (web scraping, ie Google Search etc.) this shit is even easier, you can get there in no time (you will never be able to make it reasonably efficient as google, but if you focus its meaning in single or several related fields it might actually be better)

Anyway, get to work, its not that hard and its not magic at all.

>tl;dr
>Get good at Linear Algebra (at least)
>Get good at data mining
>Get good at processing and plotting data
>Get good at the very basic logic of Neural Networks and ML in general
>Try to scale it or improve it
> ???
> Write yourself the desired "AI"

>>8576969
Find a woman, and pump her full of seed while her body is equipped to make use of it. The rest will take care of itself.

>>8576969
Pansy AI like neural networks or real AI?

File: 7.png (9KB, 120x182px) Image search: [Google]

9KB, 120x182px

>>8576969
www.youtube.com/playlist?list=PLiaHhY2iBX9hdHaRr6b7XevZtgZRa1PoU

This is the only good crash course on neural networks that I've been able to find. He explains not only qualitatively but also how you might code it using Python.

If you do the hand-holding thing when he starts to code, I'd suggest you try porting it over to another programming language afterward to make sure that you understand.

>>8577407
this

>>8576969
Prolog, then move to a rope, a small bench and a suicide letter.

File: prolog-tree.png (66KB, 775x687px) Image search: [Google]

66KB, 775x687px

>>8577756

>>8576969
Learn Python and how to use TensorFlow. You should get good results really quickly by doing so.

https://www.tensorflow.org/get_started/

How long do you need aprox. to learn the basics of Prolog? Anyone with experience?

>>8577885
the basics are incredibly simple. maybe like 2 or 3 hours.

From then it's learning how to write algorithms in it.

>>8576969
Georgia Tech OMSCS

>>8576969
LISP
tree networks
deductive abductive inductive reasoning
wave judgement

>>8577885
Prolog is just another language. Actually, it is the previous wave of AI approach. What Prolog represents is an implementation of SLD-resolution proof system.

So what you can accomplish in Prolog is automated proof finding.

In reality what happens is you create large proof trees and then Prolog does a DFS on them to return instantiated variables to your query.

Not really AI, but can be used to implement AI, and a very interesting paradigm of programming when you learn the theory behind it (which is really inseperable from practice).

>>8577989
Note, the current wage of AI is neural networks and machine learning.

AI itself is neither. Read the table of contents for Russel & Norvig to get a grasp of what AI is and is not. Maybe also read the first chapter.

>>8576969

what's your goal though

do you want it to be a chatbot? a car driver? something else?

you'll find that most of these problems can be solved with an algorithm that isn't really what you think of as "AI"

>>8576969
>AI
AI is really generalized term.

Did you mean Artificial Human Intelligence?