6

Nature has proved with the brain that it can create complex computers with very little energy consumed and released, extremely low energy compared to a common computer.

However, I noticed the design is completely different, why hasn't anyone tried to or succeeded to create a similar one?

For example:

1) Computers use memory as the main building block. After that, a processor is used to manipulate it.

2) However, biological computers do not use memory as the main building block. This is a by-product. The main building block appears to be a combination of static geometry and design of the computer (DNA) which is combined with the energy generators (mitochondria, components of both DNA and energy transfer).

Suresh Venkat
  • 32,071
  • 4
  • 95
  • 271
j riv
  • 161
  • 1
  • 2
    There's plenty of work on biologically-inspired models of computer architecture and computation, one only needs to know how to find it. There was quite a lot done in the 80s wrt hardware, and I believe that there's a bit of a resurgence these days. Here's a starting point: http://en.wikipedia.org/wiki/Bio-inspired_computing – Dave Clarke Jan 28 '11 at 07:45
  • 4
    Isn't this out of scope ? this is after all for THEORETICAL computer science. – Suresh Venkat Jan 28 '11 at 08:16
  • 2
    I'd say that is on the border (models of computation?). Hopefully the link I posted will help satisfy the asker's curiosity. – Dave Clarke Jan 28 '11 at 08:44
  • 1
    When biologists figure out how our brain works, I am sure engineers will copy it. That is, if we can live with inherent mistakes and -- in the worst case -- our slaves becoming sentient. Uh oh. – Raphael Jan 28 '11 at 09:02
  • 3
    @Raphael: that's not a particularly helpful answer, especially considering that artifacts such as neural nets have existed for quite a long time, and DNA computers have actually been build and perform useful, albeit slow, computation. Sure the models can be refined as our understanding improves, but inspiration can be drawn from biology even at our present level of understanding. – Dave Clarke Jan 28 '11 at 09:16
  • 2
    That's why it is a comment, not an answer. ;) It is true that there are Turing complete models using DNA building blocks, but they have problems wrt reading the output, obviously. But that is beside the point: the question aims at copying "biological computers" (brains, imho, not DNA and the other stated notions), and those have yet to be fully understood. Neural nets are merely a model for a tiny bit we do understand, and it shows. On larger scales, biology has been understood and copied better. – Raphael Jan 28 '11 at 09:32
  • 7
    @Lela Dax: Historically, perhaps the primary obstacle was the failure to understand how complex are the problems solved by biological systems. Computer scientists in the 1950s were confident that they could quickly produce robots that could see and walk like humans, because those functionalities were easy, but true creativity (like playing world class chess) might forever be out of reach. Exactly the opposite has been shown to be true: playing chess is easier than walking across the room. It's almost hippie-dippy: a realizaton that, hey, nature has something to teach us. – Aaron Sterling Jan 28 '11 at 15:40
  • 11
    Because computers aren't brains, and brain's aren't computers? We use computers as a metaphor for brain/mind much in the same way that earlier thinkers used telephone networks, hydraulic systems, steam engines, and clocks. (Ever "blown off steam" or "blown a gasket"? Felt "under pressure" or "wound up"?) Computer memory is nothing like natural memory. Neural nets are not built of actual neurons. It's just a metaphor. – Jeffε Jan 28 '11 at 17:57
  • 2
    I think you would probably get more useful answers if you phrase your question positively, something like: are there (theoretical) models of computation occurring in nature which are closer to biological computation in the following ways ... What are the problems with having such models? Why people prefer current models? (and I would also get my up vote for your question :) – Kaveh Jan 28 '11 at 22:29
  • 7
    DNA are the blueprints; mitochondria are the energy suppliers. I don't believe biologists think either has anything to do with how the brain actually computes (which biologists believe is accomplished primarily with electrochemical impulses produced by neurons). Since we don't actually know that many details about how the brain works, it's really hard to mimic brains in silicon. – Peter Shor Jan 28 '11 at 22:38
  • I believe Jeff Hawkins of Numenta has made inroads as to how the brain operates, you might want to check out his company and some of his lectures on YouTube. –  Jan 28 '11 at 20:23

5 Answers5

7

This might be more suitable as a comment than an answer.

From computability point of view, any machine which uses finite memory (but not necessarily constant) at any point in time and modifies it according in a local way is equivalent to Turing machines (this is an informal statement of a result by Robin Gandy from 1980 IIRC). On the other hand Turing Machine models are simpler to deal with in many respects and therefore are easier as a object of study. Natural models of computation are usually more complicated to study.

There are also studies into natural models of computation. You may want to take a look at Aaron's answer here.

Community
  • 1
Kaveh
  • 21,577
  • 8
  • 82
  • 183
  • 1
    Springer just published (Nov 2010) the first attempt at a comprehensive resource for this area: the Handbook of Nature-Inspired and Innovative Computing. I haven't looked at it yet. – Aaron Sterling Jan 30 '11 at 16:04
6

Why? Because we don't know how to, yet.

Someone else already mentioned that Jeff Hawkins at Numenta has attempted to build an AI model based on hierarchical temporal memory - which is his theory that the human brain works by recognizing patterns, then patterns of patterns, then patterns of patterns of patterns..

Really its all very complicated and we're pretty far off, it would seem, from even modelling exactly how the brain works. So it'll be years still before machines are built based on the circuits that make decisions how the human brain makes them.

bobobobo
  • 391
  • 2
  • 7
4

1) An average human consumes 100 Watts of power, which is roughly the same order of magnitude as a desktop (mostly display). The brain consumes about 20% of this, which is about 20 Watts.

Now, your ipad consumes about 2 to 3 Watts on average, which is an order of magnitude better than the human brain. Therefore, I am not quite sure what you mean by "... extremely low energy compared to a common computer."

2) It is definitely true that our thoughts/engineering are inspired by nature, but we don't have to mirror them in-toto.

I doubt if you will call our aeroplanes flawed/lacking because they don't flap their wings to fly!

We abstract important properties from nature (e.g., as JeffE says above, about neurons and neural nets) but we do not have to mimic every aspect of nature.

V Vinay
  • 3,873
  • 1
  • 24
  • 30
  • 4
    Try recording from two HD cameras, normalizing the input, combining it with a handful other sensors' output, piping it through advanced filter, pattern recognition and priorisation algorithms and storing it to harddisk with 20W. Good luck. – Raphael Jan 29 '11 at 08:23
  • I said "2 to 3 Watts on average"; on average. The same is true about humans too ... try a 100 metre dash or a marathon at 100 W! Good luck :-) (Wow! This earns me my very first downvote as well. Thanks.) – V Vinay Jan 29 '11 at 10:40
  • 2
    I do not buy your argument 1 because you are comparing the energy consumption of a human and the energy consumption of a computer when they are doing different things. I cannot see any way to compare them fairly, and I have no idea what Lela Dax’s claim “[nature] can create complex computers with […] extremely low energy compared to a common computer” precisely means. – Tsuyoshi Ito Jan 29 '11 at 22:16
  • The human brain on average does significantly more work than an ipad... – MGZero Oct 03 '11 at 20:07
2

There has been research that been done on this... see http://netresearch.ics.uci.edu/bionet/

It is not difficult to imagine a future where billions of people regularly access applications running inside the global network as part of their daily lives. To make this future a reality, network services and applications must satisfy the following requirements:

* they must be able to scale to billions of nodes and users
* they must be able to adapt to diverse and dynamic conditions in the network
* they must be secure and highly available
* they should require minimal human configuration and management

We believe that large scale biological systems, such as the bee or ant colony, have already developed many of the mechanisms needed to satisfy these requirements. We have identified several key principles and mechanisms in these biological systems, and we are now applying them to the design of network services and applications.

Albert T. Wong
  • 161
  • 5
1

An answer from a biological perspective is that biological 'design' (in the evolutionary, not the ID sense...) is often fundamentally different because it uses different principles and different materials.

I'm really just repeating the perspective of an excellent book called "Cats' Paws and Catapults" by Steven Vogel which is that even if you make machines 'inspired' by biological design, it is often impossible to directly copy Nature. Maybe the author is a bit grumpily contrary about claims of biomimicry, but still.

So, in the specific case of the brain and computers; brains are warm lumps of wet tissue that are continually self-repairing and self-regulating. Computers chips are hot sand with electricity running through them. I wouldn't agree with Penrose that we have some kind of quantum microtubule core in the neuron, but cells are quite complex micro- and nano- tech devices even before they are assembled into tissues.

So yes, you can make 'analagous' systems - but it may be that you are better off using the strengths of the particular materials and principles you have to work with and not try and emulate those from another domain.

gilleain
  • 141
  • 4