World Economic Forum
The secret to raising smart machines will be teaching them social and emotional intelligence, says Björn Schuller. The World Economic Forum Young Scientist, says future machines will learn like a child from its mother how to read emotions, sense moods, spot health conditions and be creative. “We’ll then need to ensure that we remain in control,” says Schuller, “so we know what these machines are thinking and doing.”
Click on the link to watch the whole video, or read key quotes below.
On cultivating machine intelligence
“An intelligent machine needs great perceptive abilities and great communicative skills. What does it actually takes for a machine to be intelligent? Let me highlight just three aspects: it needs to be able to learn from data, to achieve goals and ultimately reach brain-like, or maybe even beyond brain-like, intelligence.”
“An example from the movies: you may know Johnny Five from Short Circuit movie. Johnny Five was going around and exploring the world by talking to a lot of humans, and constantly asking for more input from them. So he was interacting a lot with them, and cooperatively learning with humans about the world around him.”
“So the computer gets helps from us, and it learns from us what this new data means. It can add it to its experience and enhance its model of the world. An intelligence system would then not only decide when to ask, but also who to ask, just like a child would decide to ask mommy something.”
On Learning together
“The movie Matrix had a sinister view on what machines will do in the future – that they will use it as bio batteries. Let’s share a more positive view on how they can benefit from us. Rather than using us as these battery tanks, they could source knowledge from the crowd, even directly scanning our brain to learn from us.”
“Building an intelligent machine will need human help, in particular to tell the machine about the world. Let me give you one example from my favorite application domain – speech processing. We’ve been collecting a lot of information sources about what is inside speech. You can sense a whole lot of things from the voice: the intoxication of somebody, the sleepiness of somebody, and the cognitive load level at that very moment.”
“You can also spot health-related states such as autism or Parkinson’s disease. You can also sense the emotion, interest, height, and personality. But what you really need is loads of data, and you need human help to get information on what is inside of this data.”
On reading emotions
“You might have seen the Hollywood movie “Big Hero 6”, about a medical robot that teaches a boy about moral values. It teaches the boy that revenge is not the right way to go. How does this look in reality? In a European project I coordinated, we worked with an intelligent computer system, to teach autistic children (in a playful way) how to express emotion in a way that other people understand.”
“This machine uses its perceptive abilities to sense the voice, body gesturing and facial expression. From that the machine then tells the child how to better express emotions so that others understand. It seems that intelligent machines in the future can not only learn from us, but also teach us – leading to a loop of exchange between machines and humans, increasing intelligence of the machines.”
“But we’re lacking one component for a machine to be intelligent, and that is emotional intelligence. Maybe you’ve also seen “Ex Machina” – a machine that is not only capable of having a human fall in love with her, but to exploit that human for her own goals. You can imagine how much social and emotional intelligence that takes. Mankind has had emotion as a survival factor in its history. It’s my belief that in the future intelligent machines will also have emotion as a key survival factor.”
Source
337 – 645.
the wants in life must be respected
Great
everybody that is interested in artificial intelligence wants this corporate shit called neural net and training garbage algorithm that i know how to make that is a bunch of functions that takes all avaliable room on a small microcontroller and is unsuitable fro advanced robotic development in a small chip.
i know ways to make artificial intelligent robots that runs on a microcontroller that learns without neural net. that doesnt need any training. that learns in its own way and don't need to be perfect in anything that is inexpensive to program and make. what i mean is a non iterative process that make computer learn by merging information instead of guessing by trial and error.
a robot does not need a tpu tensor flow to learn. it can done in a more basic way than that. it does not need to mimic being human, only able to replicate human behavior in a very narrow space of storage.
i have ideas about how to make ai that isn't training based that nobody else have because most people think what ever is corporate knowledge is the only knowledge.
my way of doing things is based on the limitations in hardware of a microcontroller and sensors processed by the chip.
keep it stupid simple chould be the general progress of ai by trying to replicate behavior of people in machines by a step by step development and not this train like a dog aproach that the corporations think gonna solve itself unsupervised.
that google say there is no single genious in artifial intelligence is simply not true. there is a few programmers out there that is able to challange the corporations their own way. let say that it is about 10 000 different ways to program learning algoritms than is not neural net based that can do much the same things as googles product can, it is just that people are to lazy, gullible, or comformed to break free of the hype and start on their own writing learing software from scratch using their own ideas.
when you count program space in bytes, you know what i mean by writing extremly small code as there is no room for garbage in there.
program subroutines can do partical processing work before letting one special routine solve the problem of an input of many types in the same algoritm by treating it as the same type of data even if the types of data is different in the way we think of informaiton as separate but for a machine, all the types of data is just numbers and can be treated as the same thing by one algorith.
pattern detection doesnt need to be done like a puzzle solver. it can instead be done more like a measuring program that takes measurements and track down particular pieces of data based on these measurments and produce the same goal as the more advanced systems. a learn by itself can act as an extension of a pre-programmed system where it does more about comparing data and linking data together.
in a real single celled organism, there is not neural net like the brain but there is lots of switches just like the brain how ever it is more like a machine than a network and machine logic for artificial intelligence is what we need to do first. i know what the brain does that is not what neural net of computer software is doing.
neurons comunicate by doing stuff in a machine like mechanistic way and use neural connections as an exstension to share chemical messages between each other that is also interprated and distribuated inside the single celled organism. they are not operating like the machine neural networks at all. they operate like sepearte lifeforms that share the same type of chemistry by telling each other what they feel.
here is an example. the neuron react by an injury and send a pain signal to neighboring cells that then causes the neighboring cells to adopt levels of signals inside themselves that mimic the levels form the cell that sendt it. it is like you tell someone what you feel and the other peoples bodies knows how to recreate these feeligns within themselves so you become one with them like your and them is one being, one organism. that is how cells communicate. it is not this password cracker called neural net that they use in computers at all and bear little if any resemblence and that is why we want to only operate programming on a single celled level and not this network mess and imply that the single celled life itself is more like a program of a classic type like any software and not a number adder trial and error engine like the neural net.
the cool part about a ultrasonic range sensor is that it can read a line like the surface dimesions in 3d of a face or a body and tell where the nose is by making the program recognize extrusions based on measurments of distance as target for recognition that is pre-programmed AI. a learning extension of this is by making the computer take measurments as key data and make it like some of the data to other things, so that it can separte that data from a redetection and memorize on its own. its not rocket science to figure out how to make a computer learn in a more efficient and faster way. it only need to remember data stored on a harddrive and some in internal memmory on a micro controller to recognize what it have previously stored by re-detection. this way the comptuer can learn unsupervised without probeing a pattern until it mimic it 100 percent like the neural net stuff.