Bridging the gap
Oct. 26th, 2009 05:33 pm![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
andrewduckerImagine, for a moment, that there are two kinds of thought:
1) The kind that animals are good at - looking at clusters of "stuff" and associating it together. So that "padding the sleeping human on the nose" = "breakfast" and "eating the food from the kitchen counter" = "having things thrown at you". Lots of simple associations between things based on similarity and connection. Everything is fuzzy, but you have the ability to create new associations as and when you need them.
2) The kind that computers are good at - taking a known "rule" and processing it incredibly fast to see where it leads. Given the rules by which things attract each other via gravity, they can calculate exactly when and where a rocket fired from Earth will hit Mars. Given a set of data and a set of logical operations to perform on it, computers are amazing.
The first kind is "induction", the second is "deduction".
People bridge these two worlds of thought - taking the simple associations of inductive thought and producing the kind of laws that can then be fed into computers and used to deduce new things. Animals can watch things fall and learn how to get out of the way. We can watch them fall and work out the laws that affect them, which we can then feed into computers, allowing us to perform feats that would be either staggeringly hard or impossible to perform by hand. We have a foot in both worlds, and this has turned out to be very, very useful.
This is game changing. It's not the first of its type - I've definitely read about similar things before. But it's a computer system that, when pointed at the real world, was able to deduce simple physical laws and use them to make predictions. It's the first step for computers out of deduction and into induction.
If they can extend this (and that's a very big if) it has the potential to change the world.
1) The kind that animals are good at - looking at clusters of "stuff" and associating it together. So that "padding the sleeping human on the nose" = "breakfast" and "eating the food from the kitchen counter" = "having things thrown at you". Lots of simple associations between things based on similarity and connection. Everything is fuzzy, but you have the ability to create new associations as and when you need them.
2) The kind that computers are good at - taking a known "rule" and processing it incredibly fast to see where it leads. Given the rules by which things attract each other via gravity, they can calculate exactly when and where a rocket fired from Earth will hit Mars. Given a set of data and a set of logical operations to perform on it, computers are amazing.
The first kind is "induction", the second is "deduction".
People bridge these two worlds of thought - taking the simple associations of inductive thought and producing the kind of laws that can then be fed into computers and used to deduce new things. Animals can watch things fall and learn how to get out of the way. We can watch them fall and work out the laws that affect them, which we can then feed into computers, allowing us to perform feats that would be either staggeringly hard or impossible to perform by hand. We have a foot in both worlds, and this has turned out to be very, very useful.
This is game changing. It's not the first of its type - I've definitely read about similar things before. But it's a computer system that, when pointed at the real world, was able to deduce simple physical laws and use them to make predictions. It's the first step for computers out of deduction and into induction.
If they can extend this (and that's a very big if) it has the potential to change the world.
![[identity profile]](https://www.dreamwidth.org/img/silk/identity/openid.png){kind=small}
no subject
Date: 2009-10-26 05:47 pm (UTC)no subject
Date: 2009-10-26 06:11 pm (UTC)no subject
Date: 2009-10-27 12:59 pm (UTC)no subject
Date: 2009-10-26 06:58 pm (UTC)no subject
Date: 2009-10-26 10:12 pm (UTC)[[
Deduction says that if you know A, and you know that A implies B, then you can infer B.
Induction says that if you know A, and you know B, then you can infer that A implies B.
Abduction says that if you know A implies B, and you know B, then you can infer A.
]]
There's also a long-standing problem associated with induction, namely that you can't establish the validity of inductive reasoning.
As for the system in the article, this sounds as though it's in the tradition of Lenat's Automated Mathematician.