Hacker News Reader

Comment by bnewbold | Hacker News Reader

"Scientific models are much simpler than most software"

I've been noodling with some similar ideas around collaborative editing of interconnected declarative models, and I don't think the above is necessarily true. A diagram of the Linux Kernel might be even more complex than a diagram of some wild systems biology[0], but surely there is ever more complexity to discover. The tools for massively distributed digital collaboration seem much more polished in software than scientific fields today. I hope the author's work is a step in the right direction!

For similar Scheme-y computer algebra stuff, see Sussman and Wisdom's work in "Structure and Interpretation of Classic Mechanics" and follow-ons.

[0]: https://s-media-cache-ak0.pinimg.com/originals/35/98/92/3598...

View on HN

>A diagram of the Linux Kernel might be even more complex than a diagram of some wild systems biology...

The only reason it could be more complicated is if we only include a small portion of the systems biology diagram. Since we do not understand every reaction pathway in any biological system it will necessarily be a partial picture on the biological side. On the software side we can produce a full diagram.

It will be a long time before we are able to produce a full systems biology diagram.

You can have a scientific model of photosynthesis and explain the concept and how it works to students based on that model. They wouldn't be able to somehow implement that process afterwards nor would anyone expect them to.

Software isn't like that. Software has to describe a process unambiguously and exhaustively. There is no option to say "and now you do X" while leaving X undefined.

Software can't be simpler than the process it describes, it can only provide a simpler interface.

You can also easily explain to students how a piece of complex software works, at the same level of detail. The point is biology is vastly more complex than any software humans have ever written. Evolution selects for spaghetti code.

Evolution seems to select for highly redundant and holographic code, where holographic means non-sparsely encoded in this context. Height, for example, is encoded in a wide variety of genes which can be turned on and off by even more genes.

Why would you use the term "holographic" instead of "non-sparsely encoded," "robust," "redundant," or any of the other better-understood and seemingly more descriptive terms one could use?

Because holographic encoding is a particular type of non-sparse encoding and redundancy. "Holographic" really is the most descriptive way I can think to describe it without using many more words.

Redundancy can be achieved through multiple independent systems, e.g. repeating the encoded text in multiple locations; such an approach is not holographic, though. Parity is a more holographic encoding scheme in that redundancy comes from the interaction of multiple pieces of encoded text.

The term is even more applicable when talking about convolutional neural networks in which every neural node significantly contributes to every encoding. Here the similarity to optical holograms becomes quite apparent.

The comparison is between a model and a software implementation, not with a model of the software. The point being that there are tools for helping with the implementation but not with the model.

Also, evolution selects for robust code.

Yes, but the software may be based on symbolic reasoning, in which case the computer can 'use' a description of process symbolically and unexhaustively just as a human would.

For example, these systems solve indefinite integrals without having to apply numerical methods with an infinite number of steps: http://www.integral-calculator.com/ http://maxima.sourceforge.net/ https://www.wolfram.com/mathematica/

>There is no option to say "and now you do X" while leaving X undefined.

Maybe not at runtime, but you seem to be describing lambda functions.

Reply on news.ycombinator.com