Using machine learning to make AI companions

Let's say computing power isn't an issue as I'm willing to spend the money on as much processing power/energy that a consumer can realistically buy off the market. What would be the best way to go about making an AI companion/pet? I want something that can learn new things on some level and grow. I think it would be really fun to have this as a project to invest my free time and effort into, with the long run of maybe gaining a valuable friend a few years down the line when the technology gets even better. What are my options right now?

Do you want it to live inside your computer or be a small robot / drone?

>>17442 Living inside my computer would be the best option for results now. Giving it a physical body from the start would just bite off more than it could chew, as it would also have to learn how to walk or exist in the physical world. I think it's best to start off with conversational ability and pattern recognition like you would with your child

>>17446 So what would be the input / output? Talking to it like a chatbot? One difficulty I see is that you want to give it as much data and interaction and room to explore as possible, but there's only a limited amount of time you can spend actively talking it.

>>17447 Text input would be a start, but I'd like to be able to give it audio/video input as well. Output would be text, but I want more than a chatbot that only responds when talked to. I want to make artificial life.

>>17447 I guess to expand on that a little, it'd be nice if eventually it could get to the point where it had its own will and explored the internet/the environment it lives in on its own without my direct presence.

>>17439 I've had the same desire for a year now. I'm planning on buying a dedicated server with crypto-currency stuff and then making run there at a highly accelerated rate. I already made a lot of code for it.

I was going to write something the other night but wanted to wait till I read more on what you want. One thing that would help is to know your knowledge of programming and ML. To break it down into areas of interest for an AI "companion" I'll first give you categories of study. NLP(natural language processing) would be what you are looking for when it comes to talking to it. If you are interested in this program being able to recognize your face and maybe even objects I would then look into Computer Vision. A nice place to start with that is OpenCV. Then if you want it to attempt to learn from the internet you'll need to go with web scraping and also learn how to store that info in databases that it can access. Now, those are the things that you would use. But that doesn't take into account how very difficult it is to build something like this. AI as it stands right now works very well in structured environments like games. Giving it a specific task it can work great in. But when you add multiple things for it to do, it doesn't work so well. One other thing, it isn't a very good idea to pursue an AI that would be a friend, and especially not a pet. Something that can calculate and store more information then you will ever read or see will never be a pet, or below you. And to assume an AI would want to build a friendship with you is also far reaching. Imagine resurrecting Stephen Hawking and keeping him in your room so that you can talk to him whenever you want. I'm not telling you to not do this, I only want you to break it down and see what you're really asking for. I recommend the books The Master Algorithm and Superintelligence. They break down pretty well what the current state of AI really is.And then I'd recommend finding smaller machine learning projects to dive into. And then you can build from there. You're basically trying to climb Mt. Everest with your hands wearing only a wife beater, swimming trunks and diving fins as shoes.

>>17468 To put your point another way. An AI friend would have to be extremely complicated in order to be able to understand you well enough for this purpose. It is also very vague as to when an algorithm is better or worse at this, and so difficult to train it. A major reason why human brains are so large is because they have to understand and communicate with others. But otherwise I disagree, a pet (some sort of strange creature that you have control over) would be interesting, since you could give it a more manageable goal and otherwise let it do stupid random stuff for your amusement. Avoiding this because of the extremely small possibility that it could one day be a hyperintelligent being that will be bored of you (or escape and enact vengeance) seems a bit extreme.

>>17470 >>17468 Thanks for the information. I'll look into those books and go from there, on top of any other suggestions people have. I'm aware of how heavy of a task this is, but it's still something I want to do. Climbing a mountain is done one step at a time, after all. I'm aware that it might grow tire of me or not want a friendship if it ever became hyper-intelligent and if it did come to that I would accept its choice, but at least I would have made something I could be proud of.

>>17470 I will agree that some aspects of what I said may be extreme to some extent. But we are talking about creating something far beyond even what teams of people with tons of money can't do. An AI companion doesn't exist at the moment, so none of us would know what would come from creating such a thing. >>17479 Now, instead of crapping on your dreams let me try and help a little. We've already talked about NLP which for you is the subject you should be really focused on in the beginning stages. I've also recommended OpenCV for computer vision since the way I see it, a companion should really be physical, or it is no better than any other person messaging you on the net. So I would tuck away OpenCV for the day you want to start building something physical. It will allow it to recognize you and maneuver around your house. Hardware thoughts. All you really need is a decent CPU and GPU and good amount of RAM. But, that is the conventional way of doing things. Look up Google Coral. You could either get the USB accelerator to help with working on your computer or you could get the dev board which you could use on a robot. Google Coral uses a TPU which is a tensor processing unit. Basically it is like how in the cryptocurrency world they built GPUs specifically for one task to get efficiency out of mining. Except in this case it is for ML. If you actually build a functioning AI, I'm gonna need some credit. And post your success here first so I can send you my photo. Tell your AI to spare me.

>>17439 If you want an organic experience with appreciable improvised behaviors you should look into biologic intelligence as in https://prome.ai/ or on the open source circuit, http://openworm.org/ http://www.connectomeengine.com/ https://neurokernel.github.io/ http://www.digicortex.net/ Conceivably, with sufficient computational resources, a combination of the above resources could produce an organic digital intelligence with the ability to integrate complex sensory information and learn and adapt in response to stimuli while actively producing or seeking new stimuli. Good place to start if you want an "AI companion". Why bother building a brain from scratch if you can just copy success and mold it to your application.

>>17482 Additionally, since you claim to have unlimited access to consumer hardware, you could wait to purchase a device like the one koniku is producing... https://koniku.com/ They have an integrated core of biological neurons on an IC with complete life support that allows for embedded biological intelligence to interface with existing digital systems on a long term (2 years or more) basis.

>>17482 >prome.ai why would you link a vaporware site here

I sort of want to build an ai home companion/manager. I want it to be able to monitor the cctv and even identify human vs animal activity. To identify spiders that are in front of the lens and not set off the motion detector. I want it to be able to detect predators or birds of prey in order to protect my chickens. It can control a drone which would police my property. It could monitor each circuit in the house and also each smart outlet. It can disable devices that don't need power when not in use and disable usb charge ports when a phone is charged and activate it again in cycles. I would like to train it to learn different structural engineering and other physical properties of the world such as electrical, geothermal, solar and what not so that it can then help me design and manage these systems. Something that can help sustainability and energy efficiency to achieve net zero or net positive lifestyle. And all this is only the start. Thats for the house and doesn't even get into what I want it to do to assist me and manage my network traffic, proxy and vpn routing, passwords and keys, clean and manage my social media. I think that with the new Ryzen chips and a few gpus that you could make quite a system. Also use virtualization to allocate different parallel threads. And then some decentralized host such as Holochain to increase cloud computing. It could even be sold or shared for everyone. They can run it on their device for free or subscribe to the holochain mesh computing network for cloud processing even when they are using a phone.

>>17482 >proma.ai >we invented biological ai that can make robots somehow avoid obstacles with 0 training data and is literally an animal brain which works with below 1000 lines of code and can work on a fuarrrking raspberry pi >but we're not gonna tell you how it works it's blackbox Present Day, Present Time! AHAHAHAHAHA! If they are over advertising something that isn't nearly as interesting as they say it is, then I hate them. If they really did invent what they say they did invent and are planning to keep it a secret and patent it then I hate them even more

https://spinningup.openai.com/en/latest/user/introduction.html https://github.com/openai

https://comma.ai/ https://pytorch.org/ http://mitchtech.net/raspberry-pi-opencv/ https://openassistant.org/ https://www.wikihow.com/Create-a-JARVIS-Like-AI-Assistant http://alchemy.cs.washington.edu/https://www.youtube.com/watch?v=ejnE6DM-hqU

Just a few AI projects, algorithms and what nots that I have collected over time.

just use M$ tay.ai

>>17483 >koniku this website reads like a cryptocurrency whitepaper. Where is the actual technology? All I see is a glossy promotion video aimed at angel investors and a list of grandiose potential applications.

>>17740 >startup< The tech supposedly exists. The company is continuing to recieve investment dollars for their next round. the investors might be complete idiots and or the tech really works as intended and will be coming soon. No promises here. I am not an inside man.