There’s something neat about error correction. Error correction is something everyone does, but I think the way computers do it is really neat. It’s all around us and we don’t think about it. It’s built into the way we use our phones, the way we watch TV, and even the way we shop. Error correction makes our modern life possible. It it weren’t for error correction, we wouldn’t be able to make phone calls, watch TV, or do pretty much anything online.
Error correction is the process that computers use when there isn’t enough information. There are a lot of forms of error correction. For example:
When you call someone using a landline phone, the buttons make sounds. There are two sounds per button, and a total of 7 different sounds that can be made (in a 3×4 matrix, that’s 12 different combinations.) Computers listen for those sounds. If they hear other things, they ignore the other things. Why? Because the rules say that only these 12 sound combinations are important.
This one’s pretty easy to understand. Humans do this all the time, at least young ones do. Let’s say I’m sending you the message “3+4.” I would send you “7” also, if all you heard was “3+sqqqkkk” you would be able to figure out that missing number is 4.
Another one that is easy to understand, because we do it. When we repeat ourselves, that’s redundancy error correction. Computers do this all the time and you just don’t realize it.
When you get out of the simple numerical realm into the real world, it takes a lot of really complex error corrections. Today, we talk about self-driving cars. These seem to be “just around the corner” and some cars already have lane sensors and radar-based cruise control. In many cases, they come pretty close to driving themselves… except when they don’t. Then they fail spectacularly.
The holy grail here would be an error correction system that actually tries to understand the world in which we all exist. It’s not enough to look at the lines on the road and stay between them… what about when someone else wants to be in our lane? What about when the lines are too hard to see? Humans have a tough time with these problems, and so far they’re still beyond what computers can do.
I’ve been thinking about this a lot. It seems like AI is all around us all of a sudden. Computers are doing jobs that no one thought they would ever be able to do, and they’re growing at a ridiculous rate. Just a year ago, it was news when AI created images that looked pretty close to reality. Today it’s something pretty much anyone can do with any number of apps. AI’s growth in such a short time is due to a form of error correction. AI learns by taking in a massive amount of information which is graded so that they know that this thing is right and this thing is wrong. As the AI tries to do something, it learns how to get progressively better at it and correct its errors. But will AI get to the point where it can truly take over critical tasks?
After all, in order to correct for really complex errors you need a system to constantly evaluate the world around them. That computer needs to look in real time, figure out what’s going on, and respond in an adaptive way. Rules are only going to go so far. Isn’t that kind of what consciousness is?
Backing up a little bit, we don’t really understand or agree on what consciousness is. We don’t have a really good test of how to measure it in ourselves, never mind in nonhumans. But isn’t it possible that consciousness is, at its heart, a form of error correction that’s advanced to the point where it forms our entire world? Is that the kind of thing we want to impart to computers?
Now that’s food for thought.
The post FOOD FOR THOUGHT: Error Correction appeared first on The Solid Signal Blog.
Continue reading...
What is error correction?
Error correction is the process that computers use when there isn’t enough information. There are a lot of forms of error correction. For example:
Rules-based error correction
When you call someone using a landline phone, the buttons make sounds. There are two sounds per button, and a total of 7 different sounds that can be made (in a 3×4 matrix, that’s 12 different combinations.) Computers listen for those sounds. If they hear other things, they ignore the other things. Why? Because the rules say that only these 12 sound combinations are important.
Checksum-based error correction
This one’s pretty easy to understand. Humans do this all the time, at least young ones do. Let’s say I’m sending you the message “3+4.” I would send you “7” also, if all you heard was “3+sqqqkkk” you would be able to figure out that missing number is 4.
Redundancy error correction
Another one that is easy to understand, because we do it. When we repeat ourselves, that’s redundancy error correction. Computers do this all the time and you just don’t realize it.
Some error correction is harder
When you get out of the simple numerical realm into the real world, it takes a lot of really complex error corrections. Today, we talk about self-driving cars. These seem to be “just around the corner” and some cars already have lane sensors and radar-based cruise control. In many cases, they come pretty close to driving themselves… except when they don’t. Then they fail spectacularly.
The holy grail here would be an error correction system that actually tries to understand the world in which we all exist. It’s not enough to look at the lines on the road and stay between them… what about when someone else wants to be in our lane? What about when the lines are too hard to see? Humans have a tough time with these problems, and so far they’re still beyond what computers can do.
When does error correction turn into…
I’ve been thinking about this a lot. It seems like AI is all around us all of a sudden. Computers are doing jobs that no one thought they would ever be able to do, and they’re growing at a ridiculous rate. Just a year ago, it was news when AI created images that looked pretty close to reality. Today it’s something pretty much anyone can do with any number of apps. AI’s growth in such a short time is due to a form of error correction. AI learns by taking in a massive amount of information which is graded so that they know that this thing is right and this thing is wrong. As the AI tries to do something, it learns how to get progressively better at it and correct its errors. But will AI get to the point where it can truly take over critical tasks?
After all, in order to correct for really complex errors you need a system to constantly evaluate the world around them. That computer needs to look in real time, figure out what’s going on, and respond in an adaptive way. Rules are only going to go so far. Isn’t that kind of what consciousness is?
Backing up a little bit, we don’t really understand or agree on what consciousness is. We don’t have a really good test of how to measure it in ourselves, never mind in nonhumans. But isn’t it possible that consciousness is, at its heart, a form of error correction that’s advanced to the point where it forms our entire world? Is that the kind of thing we want to impart to computers?
Now that’s food for thought.
The post FOOD FOR THOUGHT: Error Correction appeared first on The Solid Signal Blog.
Continue reading...
