Ask a guy who thinks he might have come up with a working design

ワワ
ンン

>>57318954
What will you do with the knowledge you'll gain in the near future that your working design has one or multiple fatal design flaws?

>>57318954
How does it feel knowing that you'll kill us all with your creation?

>>57318969
this desu

>>57318987
I am a big sci-fi fan. I have accounted for most types of robot apocalypse and reduced their possibility to zero. The worst case scenario is an Animatrix situation, where people who have lost their jobs to robots flip out.
>>57318969
My fallback plan is try to patent it anyway and hopefully be able to say I'm an inventor for the next 17 years, even if it's just another crappy patent.

>>57318954
How quickly will it be able to tag porn?

>>57319072
It can't use words, so it can't tag porn. It can however sort porn DVDs.

>>57318954
Can the AI be ported into VR to make 2D waifus a reality?

>>57319077
Could it organize images with similar visual traits together and let the user manually select tags?

When can we fuck it?

What are its goals?

Can you touch your bellybutton with the tip of your erect penis while standing up straight?

>>57319108
No, the AI runs on a supercomputer that requires a robot body to function. Uploading it into a VR environment would be like uploading your brain into a computer, you'd be severely confused and not able to feel anything.

>>57319134
>requires a robot body to function
Would a loli sexdoll robot be possible

>>57319134
So how will you run it for any useful length of time if it requires a constant connection to a supercomputer? Won't compute time get too expensive to maintain?

>>57319122
>Could it organize images with similar visual traits together and let the user manually select tags?

Yes, many of its possible functionalities require human direction.
>When can we fuck it?
I would like to make sexbots right away because of the profit potential and because it would end sex slavery.
>What are its goals?
Whatever you want them to be. The design allows for any kind of robot body. You can tweak the programming to create any behavior you need. It can also be trained and retrained for specific tasks.
>>57319128
Yes.
>>57319160
Any kind of sexbot you want with any kind of body that responds however you want. I'm a /b/tard so I know there are people who will want a robot that cries when you rape it. Not looking forward to that being a thing, but only because these robots will be like my children.

What makes you think that someone much smarter than you hasn't come up with a similar design? This is the singularity we are talking about.

>>57319172
It is a supercomputer, but the hardware requirements are quite low. Most of the processors only need to be as powerful as a Pentium 1. The biggest problem with the design is cooling and the second biggest is the power requirements. There aren't batteries powerful enough to make a dog sized one without plugging it into a wall.

Early models will likely be limited to vehicle sized robots.
>>57319200
I've been researching this for two years. To my great surprise, no one seems to have come up with anything that remotely resembles it.

>>57319213
hurry up and copyright/patent it, roll in the money that comes in when someone offers to buy it out

>>57319228
That is the plan. I'm in the process of writing up my patent application.

>>57319213
>There aren't batteries powerful enough to make a dog sized one without plugging it into a wall

Use wireless power

>>57319213
Doesn't mean someone hasn't had the idea. Clocks ticking OP.

>>57319234
Wires are cheaper and easier. It's not that there's no battery powerful enough to run it, but it would discharge too quickly.

>>57318962

7 7
; ;

>>57318954
Will you let me fuck it?

>>57319258
"WanWan". It means "Woof! Woof!" in Japanese.
>>57319261
You can buy your own and fuck that.

>>57319281
But I want to fuck yours

>>57319289
Good luck. If this works, I'm going to be a trillionaire with a personal army of security robots.

>>57318954
how old are you, really?

>>57318954
Do you need a logo?

>>57319314
28
>>57319318
Once the patent expires, no one will have any reason to keep making me richer unless I develop a brand.

>>57318954
Prove to us that you're not full of shit.

>>57319346
Ask me anything. I was expecting more technical questions from /g/.

Why is your design better than any previous ones?

>>57319360
> I was expecting more technical questions from /g/.
Why?

>>57319360
What made you come up with the design?

>>57319370
It's completely different. Conventional robots use a standard computer and weak AI. They don't work especially well.

My design is basically a general AI that can be trained to function in any environment.
>>57319382
My dog died.

>>57319412
>My dog died.
bummer, dude

>>57319309
I like the spirit.

So basically in all likelyhood you will not achieve this but the funny thing is you actually could if you pursued it.

Hope it works out for you. I would buy a teddy bear model like in AI: Artificial Intelligence just to have a pal

>>57319437
Yeah. I wanted to build a dog that wouldn't die on me. Instead I invented a robot that grows old and dies, just more slowly.

>>57319448
>So basically in all likelyhood you will not achieve this but the funny thing is you actually could if you pursued it.

I didn't take this idea seriously for a long time. I was reading a book by Daniel Dennett and there was this one part where he said no one has any idea how brains really work and there's no real reason to expect current lines of research would answer the question, so maybe the next Einstein will come totally out of left field.

And I thought, why not me? At least I'll be able to say I build robots for fun.

>>57319450
Interesting you say that op.

I came up with a design to turn any smartphone into a near infinitely versatile killing machine. Got the idea unconventionally by playing airsoft in the woods.

File: wanwan.jpg (61KB, 561x374px) Image search: [Google]

61KB, 561x374px

Just point us to your Kickstarter so we can buy one already.

>>57319534
I'm hoping to sell rights to manufacture the supercomputers themselves to Intel.

If I ever start a Kickstarter for this it's because every tech company has laughed in my face and I'm doing a cashgrab.

>>57319056
That's not true sentience if the robots can't act and many decisions of their own.

Why not just "forget" to program spite and anger and other bad shit when making it? Still not true sentience but it's close enough and safer.

>>57319566
It is designed to make decisions, but it's also designed to be only somewhat smarter than a dog.

There's no reason to program it for anger or panic attacks or vengeance. It does have virtual emotions, but they don't have to correlate to mammal emotions in any way. Emotions are just motives at the most fundamental level and its virtual emotions are just dynamically generated reasons to reorient the platform and to revalue detected objects.

Do you have a degree

>>57319591
No, I ran out of money and couldn't get more loans. My degree wasn't in engineering.

To be clear, guys, I'm not an engineer or scientist of any sort. At best, I'm a hobbyist.

>>57319376
I'm a /b/tard. It's not like I expected you guys to ask for circuit diagrams so you can critique them, but I figured the first question here would be how does it work, whereas in /b/ I would expect most of the questions would be about fucking it.

I'm going to take a shower and get ready for bed. If the thread's still up in an hour I'll answer any questions.

>>57319056
>I have accounted for most types of robot apocalypse and reduced their possibility to zero.
How did you achieve this? This is not trivial.

>>57318954
>>57318962
わんわん
https://www.youtube.com/watch?v=LBVOVltLCVI

>>57319710
Still in bathroom. I have a long answer to this question. But the low intelligence alone prevents Skynet, HAL, or Ultron type rogue AIs.

>>57318954
can my robot be a strict waifu making me make an amount of fit exercices before letting me fuck her ?

>>57320732
>can my robot be a strict waifu making me make an amount of fit exercices before letting me fuck her ?
Can the robo waifu be a tsundere bitch who will never let me fuck her and humiliates me as she fucks other people?

How fast would this thing bootstrap to the limits of technology?

Let me look over your framework senpai. It's pretty hard to believe a hobbyist on a Vietnamese pumpkin carving site made a working AI when entire paid teams of people have at best made an intellegent search engine.

>>57320732
>>57320780
Can my robot actually love me?


Serious question though. Can it?

>>57320864
Can anyone love you anon? Can you love yourself?

>>57320864
What is love anon?

>>57320906
Baby don't hurt me

don't hurt me

no more

>>57320906
>>57320887

Fuck. I'm not even sure if you guys are ironic shitposting or not but I honestly don't have I good answer to this. I don't hate myself - not anymore, but I don't really know what "love" is so I can't even say whether or not I love myself. Supposing an AI were made to love me, I wouldn't be able to confirm whether or not it actually does. This is some serious shit.

So hows it work

>>57318954
OK OP pardon my incredulity but how much research did you do into the subject before you decided your approach was unique and have you built any working prototypes?
Without googling do you know who Steve Furber, Geoffrey Hinton, Terry Sejnowski and Peter Dayan are?

What is your approach, what makes it unique?

File: Whatmult.jpg (49KB, 800x300px) Image search: [Google]

49KB, 800x300px

>>57318954
Are you the idea guy?

If your design is so revolutionary, when do you expect to "make a splash" in the industry? And from there, how long do you think it will take for sentient robots to become available for the average consumer?

File: judgment_day.png (37KB, 227x374px) Image search: [Google]

37KB, 227x374px

>>57318954

>>57318954
what is it like to roleplay on /g/

>>57320933
>love
>emotion responsible for bonding with another living being
>just like any emotion, it's a byproduct of evolution and an essential part of our survival instinct
>exists in order to preserve life through cooperation and easier mating
An AI can only be considered intelligent if it does have a survival instinct, that's the most important part of intelligence.

OP:

1) If it's sentient, then it can presumably pass a (very limited kind of) Turing Test. If you've developed something that can do this, you have a moral duty to share the source code with the world. Consciousness is not a company product.


2) How do you plan to maintain legacy bots? It seems the best would be to write your own package manager and own the main repo. Once someone buys the bot they buy along with it a lifetime license to access the repos (-stable, -current, -dev, etc). This has the obvious benefit of being able to resolve bugs by having the owner run a simple:

bot-get update


Then you can shill your newest shit and avoid the bad reputation Android got due to fragmentation.


3) If you think you have a working design, then doesn't that mean you don't have a currently working prototype? Why are you patenting something that you aren't sure can even exist yet?

File: aaa.gif (3MB, 1504x759px) Image search: [Google]

3MB, 1504x759px

>>57318954
Questions:

1) What do you consider sentient ?
2) What is different from your design from regular " robots" ?
3) Is it hardware oriented or are you actually talking about AI instead of a robot?
4) Do you have any proof of concept?
5) Why did you come to /g/ for questions, attention whoring? Development? Feedback?


pic unrelated

>>57321128
Third time's the charm.

>>57318954
can your AI beat deepmind at a game of GO? if not BTFO

>>57321145

Lol yup. Couldn't figure out how to post code.
I thought it was:


[$code] insert code here [$code]

minus the $ but I guess not.

>>57321162
There's no shame in trial and error.

>>57321149
Can you? If not, BTFO.

>>57319213
>Most of the processors only need to be as powerful as a Pentium 1

So a Xeon Phi?

What was the trick that let you develop this?
What is the key that let it go on beyond just machine learning?

Is your design based on a first self-modifying kernel (genetic/neural/etc) that attempts to model the world (the simul-space), which feeds into a second self-modifying kernel that attempts to manipulate that world to achieve a desirable outcome, with the most successful result being performed in real-space and the difference being used to retrain the simul-space?

>>57321177
you are retarded

the thing with general AI's is, if/when they come about, being computers they will be able to run millions of computations in their head in seconds, unlike humans.

The way a human masters go is by playing hundreds and thousands of games and slowly developing an intuition/feel for the game despite the enormous search space. That's what made GO the holy grail of AI, the game is out of reach of brute force.

Deepmind learned the game by playing millions of games against itself, until it picked up which strategies worked. Very similar to how a human does it despite the fact it's neural net is no where near as advanced as a humans. What gave Deepmind the edge is the fact it can play far more games against itself in very little time.

Any actual strong AI will have a far more efficient and complex neural net than deepmind. It would crush literally any game that humans play with ease because of the computational advantages it has. It would be able to do what deepmind can do, but far quicker with the same computing resources.

what makes you think that you can out compete the team of some of the worlds leading experts in AI that has the computing resources and $$ of the worlds biggest tech company at their disposal?

>>57321382
>>you are retarded
Okay.

>>what makes you think that you can out compete the team of some of the worlds leading experts in AI that has the computing resources and $$ of the worlds biggest tech company at their disposal?
I'm not OP.

You don't seem to understand what a sentient AI is. Deepmind is machine learning, not artificial intelligent sentience. You have your subjects mixed up.

>>57318954
Can it run crysis?

>>57318954
are you twelve?

>>57322217
>Update
Probably, I had that same Idea at twelve.

>>57321382
>being computers they will be able to run millions of computations in their head in seconds
this i what every AI person focus on while making it. And it's awesome yes, and impressive yes, but that's not how sentient AI will work. at all.

>>57319634
Welcome to /g/, /b/ro.

>>57321382
>>57321438
>>57322487

https://aeon.co/essays/your-brain-does-not-process-information-and-it-is-not-a-computer

>>57321066
Why on earth would that be an important part of intelligence? Literally everything from fruit flies to biggest mammals have a survival instinct while most creatures just barely even qualify as having a brain.

If you ask me I see the desire to obtain and preserve information as crucial parts of intelligence. Of course with all that the logical conclusion that preserving oneself and other like-minded sentient beings seems unavoidable but what do I know.

At least it doesn't have to be an instinct, but more of a conscious decision to act in a certain way. That's almost pure semantics though.

>>57318954
Yes, well, what can you do to prove any of this to us? As far as I see it, you're just another idiot making a stupid thread.

>>57320864
Yes, though people will argue forever whether it's real love or not.
>>57320732
No, I don't think it will be able to count.
>>57321037
Not all that long. The main development cost/time is going to be programming. Military/commercial models will probably be released before consumer ones.

>>57320989
>What is your approach, what makes it unique?

It's not a normal supercomputer. The processors don't have to run in parallel and asynchronous transmission is possible. It basically turns environmental sensor data into a low-definition universe emulation.

It's not a normal AI. You don't write a program per se, you write the instruction-sets of the processors to control a recursive data loop that outputs, at various tiers, to the motors that control the robot. The robot has to be trained. It won't know how to walk on activation.

>>57321023
I guess so.

>>57321128
1) It's not human level intelligence, so it can never pass the Turing test. The point of patents is to disclose methods in exchange for temporary monopoly rights.

2) They aren't reprogrammable like that.

3) Development costs on the prototype should be around two to three million dollars. I need the patent to get anyone's attention.

>>57321134
1) Not sapient. It can detect and respond to environmental stimulus.
2) The supercomputer network.
3) It's hardware oriented. The AI is dynamically generated.
4) Not really. I have diagrams.
5) The design has changed a lot since I started and at this point I'm tweaking a few points while I finish writing up the detailed description. I was hoping questions might give stimulate ideas I wouldn't consider otehrwise.

>>57321149
Can your dog play chess?
>>57321250
Not quite, but something similar.
>>57321216
>beyond just machine learning
The marker\sleep system.

>>57323234
People still talk about the hard problem of consciousness as though it were a real problem. Declaring that it's preposterous that the human brain functions at all has never struck me as a particularly useful or persuasive argument.

>>57323472
Ask me anything. As I've said, this is an idea on paper written by a guy who is not an engineer. I don't think it's unreasonable to believe that I'm some guy with an idea that may or may not be the holy grail of robotics.

embed the word "swordfish" into the source code somehow op

>>57318954
does this mean I can finally have my own gynoid

how smart would this AI be OP?

Would it have low level animal intelligence or human level?

Very interesting if you aren't making things up

>>57318954
>inb4 those bait questions where the car's brakes conveniently failed and the AI has to choose one of two choices to sacrifice

>>57323809
What you are describing sounds an awful lot like machine learning, a fairly well established field, in fact even building and training a neural net is not that uncommon a task in the current year and if you aren't aware of this or haven't tried to build a net yourself then I'm a little skeptical that you have grasped the entirety of the problem and realise a solution at all. Anyone can conjure up a nice sounding idea of how a thinking AI should work; The devil, as always, is in the detail.

>>57324030
It's not unfeasible but I would certainly say at the very least it's highly implausible. Paint me dubious but I just don't think you've read enough into this. Can you explain why your system is not one of the canonical approaches to machine learning or why it would outperform them?

I mean don't get me wrong, you could be some latent genius who's sidestepped the thousands of researchers in the field but I just think that perhaps a little more work than writing out a nice idea on a piece of paper needs to be done before you over inflate your ego.

>>57325042
OP considers himself as idea guy

>>57324420
Animal level intelligence.
>>57324629
It's not capable of moral choices. It'll prioritize what it's programmed to prioritize, most likely the occupants and itself over other vehicles, but below pedestrians.
>>57325085
I make no pretense that this will work or that I am some kind of Einstein. Either this happens to work or I am just another guy who had an idea that never panned out.
>>57325042
I appreciate that I am being overly vague. I am familiar with neural networks\deep learning frameworks. At least one is a mandatory part of the design between tiers 2 and 3 and it's desirable to have several more.

A deep learning network alone can't enable a sentient robot.

The single biggest advantage over a WanWan style supercomputer is that a conventional processor is essentially limited to a single read/write head. Parallel processing doesn't really get around this problem.

For environmental sentience and emergent behavior, it needs to be able to simultaneously process all sensor data. To do that, you need to isolate the processors directly analyzing sensor input and to that end, processors are grouped into clusters with a single input bus and a single output/reinput bus to connect it to the next tier.

>>57318954
Is it unethical to shut down sentient AI?

>>57319710
First, the animal level intelligence fundamentally limits it to a non-dangerous level. It will never be able to speak, form societies, or reproduce (build new ones) without human assistance. As said, this eliminates Ultron style apocalypses.

Second, because it is a strong AI, programming errors causing a runaway AI/Von Neuman apocalypse is impossible. Furthermore, because the AI is dynamically generated post-activation, it's fully trainable. You stack the deck by manipulating how it processes sensor input and the emotional modes, but even if you accidentally program a killbot to only target babies, you can retrain it until you get teh desired task execution.

Third, it can't jump between platforms ala GITS. The AI is generated by the actual hardware. You could copy memory between identical models and theoretically make a network of WanWans operating in parallel, but it would be practically impossible for more than two, maybe three. Nor can they be networked together on an ad hoc basis. This prevents a Geth style apocalypse.

If you think of any other ways a dog-level intelligence sentient robot might kill all humans, let me know.

>>57325637
Is it unethical to kill a cow? Everything dies.

Honestly, I'm more concerned with the ethics of building a sentient robot to be raped. We can program it to like sex, but people are going to want a crying lolibot that struggles and sooner or later nothing will stop someone from building one. Is it ethical to fuck a dog? I don't really want to know the answer to that question.

>>57323850
How can you afford to get a patent? Do you have a patent lawyer? Since you don't have a degree, who has agreed to support you through this process?

>>57325734
If humans create something to be raped, and then rape it, is that truly unethical? If so, humans are born to die, therefore birth is unethical? Would God then be a sinner? This is stuff worth thinking about.

>>57319360
Algorithms or machine learning?

>>57326298
Patents are pretty cheap actually. I'm waiting to meet with a lawyer until I finish the rough draft of the patent application.
>>57326329
Machine learning.

>>57326299
>Would God then be a sinner?

God is just a virtual person running as a background program in your mind. He is a sinner if you say he's a sinner.