Hacker News new | past | comments | ask | show | jobs | submit login
Intelligent Machinery (1948) [pdf] (weightagnostic.github.io)
72 points by hardmaru 33 days ago | hide | past | web | favorite | 16 comments



Hi,

In addition to the link posted by dang about the previous discussion, here are some other articles I found interesting on Alan Turing's "Type-B Unorganised Machines"

Evolved Turing neural networks

http://compucology.net/evolved

Turing's Neural Networks of 1948

http://www.alanturing.net/turing_archive/pages/Reference%20A...

https://en.wikipedia.org/wiki/Unorganized_machine


A definite classic. What's unique about human intelligence? There doesn't seem to be anything in our carbon based systems that can't be theoretically replicated on silicon. Turing may not have been religious, but the amazement I think he felt about human intelligence was close to religious awe.


>What's unique about human intelligence? There doesn't seem to be anything in our carbon based systems that can't be theoretically replicated on silicon.

Organical systems convey much more, and predominantly analog signals, that don't have any sort of equivalent silicon basis. Wetware, a computer in every living cell by Dennis Bray goes into detail in how multidemnsional and complex the information processing in even individual cells is, down to single celled organisms executing complex behaviour.

You don't need to be religious to be sceptical and to see the difference between organic and non-organic information processing. In computer science circles there tends to be an overemphasis of neural connections over the actual matter and structure of neural cells.


If you are claiming that our current computers are way too slow/small to replicate our brains, well, I don't think anybody disagrees.

But the GP is saying that it's possible to create some computer that replicates it (even if we can't do it now). I don't see anything disagreeing with it in your post.


I'm sorry, but they kinda said nothing about speed or size.

>predominantly analog signals, that don't have any sort of equivalent silicon basis.

They said the computing units we use on silicon are, at a fundamental level, completely different. And I'm inclined to believe them. Yes, it is computable to simulate the sort of computational processes they allude to. And yes, it would be very slow to do that. But I'm just noting that that was not their point.

To your point, some people subscribe to a theory that the distributed nature and synchronicity of neural mechanisms makes then similar to n-body problems (then you're modeling these mechanics as ODEs). To that end, I don't think any amount of compute will solve them exactly at that level of granularity. So yeah, 'theoretically' replicable but many things are theoretically possible, and I don't think they'll be practically useful soon (like Dyson Spheres for an extreme example).

That being said, I don't think approximation is a bad approach at all given we are aware of the fact that we're approximating. Directly replicating... while philosophically interesting, is just not useful.


> some people subscribe to a theory that the distributed nature and synchronicity of neural mechanisms makes then similar to n-body problems

I'm sorry, but are you claiming that our brain is hard to simulate because it's chaotic?

That's not a useful idea. If that is true, than it only means that we have a lot of accidental complexity and it's much easier to make something more intelligent than us than it's to simulate ourselves. This would add no constraints on the creation of a general AI.


You can easily build analog circuits. It’s actually how most early computers where designed.

Further, the ability to withstand minor effects like the pressure changes from each heartbeat or the shock from individual steps represent a robust system. Worse the brain needs redundancy to deal with things like cell deaths.

We can argue that simulating you needs quantum level precision, but we can’t tell if someone is replaced with a fairly close copy because that’s literally happening all the time.


I'm sorry, did anyone mention quantum effects..?


I am not sure what else you mean by ‘at a fundamental level’.

Get enough neurotransmitters of the right type at the receptors of each neuron and it fires. At the level of chemistry that’s very much computable. It’s only by digging one step below that and saying you need to simulate quantum effects for some reason that doing so in a chip becomes an issue.


> Organical systems convey much more, and predominantly analog signals

Lots of evidence neurons encode for both digital and analog signals

https://en.wikipedia.org/wiki/Neuron#Neural_coding

https://www.zdnet.com/article/is-your-brain-analog-or-digita...


On the one hand, I've got a foot in both worlds, and I emphatically agree that people tend to naively intuit that "the brain is a really big neural network". This is simply false, as you say: each cell in the body has complex functionality, and brain cells communicate slow, analog, low-precision information by secreting and absorbing neurotransmitters, and there's fast, precise information communicated via pulsatile spikes.

On the other hand, the more I learn about the biological side of things, the more I begin to suspect that there are compositional principles underlying biological function, that some of the complexity is in fact mere degeneracy evolved into living systems for the sake of redundancy/survivability rather than performing entirely independent functions.


There are 2 questions to answer:

1 - Are unknown physical principles employed-by / affective-to the living sentient organisms. (Interesting follow up: are all physical processes and phenomena comprehensible.)

2 - Are all physical processes and phenomena subject to simulation (in practice, not principle).

If No (Yes) Yes, then I suppose there is nothing stopping us from building "a man in our own image".


Why are we trying to build a man of our image in the first place? Are there not enough men on this planet or do we want to build a better man, a man that has no faults, a man that is above all men, a man that can be a master and slave at the same time, a man we can control at our will? Why copy all the vestigial faults that come with man and not build god instead? I'm being sarcastic here and yes, I think we are trying to play god but all we have is a man to copy from, a man that we don't fully understand, a man capable of driving their whole species to extinction, a narcissistic man who wants everything replaced with their image. Had we concentrated on nature instead, we'd have long realized that everything we need is already here.


>Had we concentrated on nature instead, we'd have long realized that everything we need is already here.

What's that mean?


Intelligence may be seated in an immaterial soul and function as a halting oracle. Turing doesn't disprove such an idea.





Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact

Search: