/aig/ - Artificial Intelligence General

While neural networks can produce amazing results, they are obviously limited and not "intelligent". Isn't the more interesting question what software logic could create strong AI? Is it even possible?

>>58541498
>While neural networks can produce amazing results, they are obviously limited and not "intelligent".
I think that's entirely contingent on a person's definition of intelligent. I've found in my time that for most laypeople it's extremely lax so to them a computer is for all intents and purposes intelligent. It's really just who you ask. Most normies are impressed by Cortana and Alexia. My mother actively tries not to upset her Siri and feel like she's having meaningful conversation with the chatbots I made so you get the idea.

>Isn't the more interesting question what software logic could create strong AI? Is it even possible
Well if we had the answer to that question it would already exist. I do believe however that "true" intelligence actually could be evolved in a sufficiently large and compartmentalized ann, but that the design of such a system is simply out of our capabilities at the moment. In any case, time will tell.

File: 1484272227987.jpg (98KB, 409x409px) Image search: [Google]

98KB, 409x409px

>>58541479
>My mother actively tries not to upset her Siri
you made me kek

>>58541648
Yeah she's not that smart honestly... it kinda makes me sad because she's the living incarnation of Dunning Kruger effect. You want to tell her why she's wrong but you just know that it won't get through to her...

>>58541690
That isn't necessarily true. I can't think of any examples off the top of my head but there's no logical reason to believe that. I agree with you but we shouldn't make statements like that without any real proof.

>>58541753
I logically fukked ur mum.

keras.io is really nice. It uses TensorFlow or Theano but it is even easier to build a NN with it. I still do not know that much about Machine Learning but nevertheless is is really easy to put some layers together.

I just took the included VGG16 model, put some fully-connected layers on top and was ready for training. Unfortunately the loss converges really slow and I am too stupid to know the reason (learning rate seems not to be the problem). I think I have to go back and read some more theory. And maybe my model is too big for my problem, too. Training takes really long. (2 hours for 300,000 images in batches of 25)

>>58541666
your trips tell you to kill her