Open-source software for human brain simulation [closed]_问答_开发者

Closed. This question does not meet Stack Overflow guidelines. It is not currently accepting answers.

开发者_开发知识库

We don’t allow questions seeking recommendations for books, tools, software libraries, and more. You can edit the question so it can be answered with facts and citations.

Closed 7 years ago.

Improve this question

Is there any open-source software that is trying to implement and emulate the human brain (e.g. intelligence and feelings)?

Jeff Hawkins (the author of the book "On Intelligence") founded a company called Numenta. He came up with a theory on how a human brain works: the hierarchical temporal memory (HTM). Numenta has developed a software library called NuPIC (which is now open-source), which implements several algorithms related to this theory.

We don't know enough about how the brain works to attempt to do what you're saying in a principled way. (I.e., anything of the sort is "guessing wildly".) So this isn't really a software question--if we had any idea of what to write, perhaps it would be, but right now we don't.

However, you may be interested in the Blue Brain Project for a more biological approach, or in any of a number of machine learning projects like the DARPA Autonomous Vehicle Grand Challenge. A less useful but more conversational approach might be found in ALICE, but I wouldn't recommend that for anything useful.

Having used a brain for over 50 years, it's the last thing I'd choose to model an AI on. Brains are notoriously unreliable and arbitrary, and have hidden biases that could take a shrink years to sort out.

Most of the AI lectures I took in school were by professors who had been chasing the dream of "strong AI" for years, and had finally realized that if they could barely understand how a human brain and mind function (and the theories behind these functions sometimes change almost daily), how could they ever hope to simulate it artificially? Most of them were resigned to AI in niches where the problem is more clearly defined: pathfinding, applications of SAT-solving, image processing, chess-winning, conversation, etc... but they'd given up on the true, general-purpose "thinking machine".

My advice would be to look into a specific problem are that you are interested in (such as pathfinding; applications of SAT solvers, such as diagnosis systems; etc...) and see what AI approaches have been taken to solve them. Maybe the problem you are interested in doesn't have much in terms of AI solutions. In that case, you could get started on a new one! ;)

...But you will probably have to narrow it down to a specific class of problem if you don't want to be overwhelmed - at least at first.

There are algorithms which model the human brain. They're called Artificial Neural Networks (ANN). They basically model the synapses and attempt to model the way in which our synapses can accept signals and, if the combined signal input is strong enough, fire their own signals along dendrites to other synapses.

The thing is, building ANNs as a method of attempting to simulate the real thing, is a lot like using a nuke to simulate the sun; Sure, it'll give you some valuable data, but, in terms of its ability to approximate that which it's modelling, it falls WAYYY short.

I'm not 100% positive on the relative scales here, but to give a decent idea, consider the following (this is definitely going to be off by a few orders of magnitude... but it's close enough to get an idea of why ANNs aren't running the world for us):

If you took every single computer on the planet and had them using every single available resource to create the largest ANNs they could, and then connected all those different ANNs to each other (thus creating an even larger ANN) you MIGHT start to get close to the number of connections present in the human brain.

You could have a look at Cyc:

Cyc is an artificial intelligence project that attempts to assemble a comprehensive ontology and knowledge base of everyday common sense knowledge, with the goal of enabling AI applications to perform human-like reasoning. The project was started in 1984 by Douglas Lenat at MCC and is developed by company Cycorp. Parts of the project are released as OpenCyc, which provides an API, RDF endpoint, and data dump under an open source license.

Not precisely a brain, but an important component of an articifial intelligence.

The field you're looking for is Machine Learning. Specifically evolutionary algorithms like Genetic Algorithms or Genetic Programming. One algorithm that I know that specifically set out to mimic the human brain is Hierarchical Temporal Memory which I read about here. But this is a very difficult problem and we are still YEARS away from mimicing the human brain in any meaningful way.

There is a field of computer science known as Organic Computing http://en.wikipedia.org/wiki/Organic_computing Some of the goals of this effort are to have the following.

self-organization
self-configuration (auto-configuration)
self-optimisation (automated optimization)
self-healing
self-protection (automated computer security)
self-explaining
context-awareness

The closest possible thing that I know to this would be the Watchmaker framework. While not related to the human brain, it does seem to strive towards an AI-type framework.

http://watchmaker.uncommons.org/

The Watchmaker Framework is an extensible, high-performance, object-oriented framework for implementing platform-independent evolutionary/genetic algorithms in Java