EcoTrain QOTW: Artificial Intelligence... The Beginning OR the End?
"Artificial Intelligence" (the word) is a weird thing. Most of what people think of as AI is hollyweird projection. They think that the movies are based on at least a little bit of truth.
"Scientists" have been going on and on for years that once a computer gets enough processors, it will gain intelligence, or self awareness, or consciousness. And so, the stories are about such happening.
And then we apply our nightmares, our what if processors. To prepare for the worst outcome. The AI that goes rogue and decides to destroy the humans.
Of course, all of this is wrong. Doesn't have any bearing in reality. And all of these stories may be programming to get people to never trust/use AI, so that computers stay in the dark age. As an impossible and frustrating tool to use. Not the tools that really help with your life. Not the tools that point to who is really pulling the strings. Not a tool that graphs out, clearly, the conspiracy that is happening. (a conspiracy is a network of people; an AI is a neural net programmed interface)
What really is "Neural Net Programming"?
AI is not a really big program. The actual program is one of the tiniest ever written.
It takes inputs, and then runs through a bunch of calculation nodes. (this is where the AI chip would work) Each node comes up with a value, and then it is either used as inputs to another layer of nodes, or output. (the AI chip would do the computation of all the nodes of each layer in parallel. Making time to output really fast)
The AI program works by training each of these nodes. So, a large amount of data are put into the input, and the appropriate output is searched for. So, you keep adjusting values/formulas at each node until the outputs are correct.
The programmer has no idea what is going on inside the nodes. Even pulling out al the values, it doesn't make much sense. But you end up with a thing, where you place something on the inputs, and quite quickly, you get an output that is usually correct. The AI program has no way of knowing if the output is correct, and if you have trash input, you will get trash output. GIGO
What AI is not
AI has no awareness of itself.
AI does not want anything.
AI doesn't even know what it is looking at.
It could be trained to tell (a picture of) men from women, but it does not know what a man or a woman is. And if you start feeding the training software pictures of trans people, and calling them the opposite of what they look like, the AI will break. It will not start recognizing trans. It doesn't know trans. And its outputs were limited to man/woman. The output of the program will just become completely useless.
What AI is
AI is really useful for recognizing patterns.
Things like questions from humans are patterns that is easy for a neural net to work with, that linear programming is not.
Take "Jumbo Shrimp" if you process each word, looking it up in a dictionary, then you come up with a totally wrong answer. But, if you take a neural net, it takes it as a whole.
And human language is made up of many phrases. (not really individual words) and the AI can be trained to correctly interpret these. So, a search engine could be developed that has the AI break down the human query into active words for searching. Or, if you had a large enough output set (yes, this can be ALL of the indexed websites) then the AI can match the input, through all the neural nodes and come up with a website that matches.
AIs have been trained to recognize picture types. So, it can find matches to people's faces with an ID number. (remember, it doesn't know if it is correct, and so the same wrong person has been arrested three times)
What can AI do for us?
Most people do fairly repetitive things. Their interests are in a certain area. Their moods are usually in a set pattern. Their likes are fairly consistent.
So, your own personal AI can be trained on your inputs. And thus become a really efficient secretary for you. If you forget names, the AI will provide names. If you forget appoints, the AI will remind you of them. (and with enough feedback, that is retraining of inputs to outputs, it will figure out when to remind you.)
The AI will soon map your words onto what you mean by those words. (not in an comprehending language way, but in a way that takes in "Jumbo Shrimp" and provides you a list of restaurants in your area that have good seafood) It will be very helpful in finding that file you are looking for on the computer.
Overall, it will just make computers so much easier to use.
Especially as we get bigger and bigger hard drives. The more data, the more we forget where/what it all is. The more we need assistance.
Computers will finally work like people expect them to.
Our computers will not gain sentience. The microchip processor doesn't have anything to allow for it. It has 0s and 1s. It doesn't have any maybes. It has very simplex instruction sets. It doesn't have anything that allows for "more" or "different". In fact, those are called errors and make the computer useless.
When you meditate, and get outside your body, and notice your thoughts, and notice your thoughts are not coming from your body. It becomes apparent that thought is not in the body. And from there, the idea that enough neurons makes a thinking computer is laughable.
You put these two together, and you can see why we will never have sentient AI that decides to kill all the humans.
However, you may have humans that decide to put an AI in charge of some real world process, and then along comes some garbage input, and then, all of a sudden the output is doing strange things. Which may be quite destructive. (remember, an AI does not know if its output is correct/truthful) So, stupid humans, doing stupid things with AIs, may spell our destruction. But, that has always been the case.
The thing about AI is that it takes our words, and gives a USUALLY accurate response. So much so, that we start to ascribe sentience to it. We start to think that it knows us, that we have an intimate connection to it. But it really is none of these things. It is just a trainable computer program that can be very helpful in searching for an answer in KNOWN answers.