The PZ Myers’ among us need to believe that society will remain essentially as it is so that centralization has a chance to take hold. Their model of centralization is redistribution based in a quixotic agrarian interpretation of economics. One cannot redistribute wealth and transform everyone into a poet farmer if the definitions of wealth and farming change beyond recognition. Both theology and technology threaten to bring about such dramatic change.  

From whence comes the desire to see humanity remain forever pedestrian?  It is apparently a kind of nihilistic urge reminiscent of O’Brien in George Orwell’s 1984. O’Brien told Winston that the future would consist of “a boot stamping on a human face forever”. Winston could not understand this urge. He observed that O’Brien was supremely intelligent, but ultimately insane. It puzzled Winston that anyone would want to freeze history at that particular point in time.  

The only explanation must be a kind of general loathing of humanity. The PZ Myers personalities began to resent their fellow humans at an early age.  This resentment led to judgment. Judgment led to sentencing. They needed to believe that humans are essentially hopeless brutes who—owing to their culpability in their own condition—are less deserving than animals. What they proffer is nothing less than political secular damnation.

 In this, they are essentially like the 15th century inquisitors of Spain. Having neither joy nor wealth, they can console themselves only in the power of their station. If happiness cannot be theirs—if they cannot see themselves as the recipients of life’s fruits—they can at least see that none of those fruits are available to others. The beautiful, witty, and courageous must all suffer equally under the oppressive omnipotent state.

The brain is a beautiful self-organising mess of hardware, software, firmware, soft-hardware and hard-software, capable of creating massively interconnected hierarchical and parallel n-dimensional tree and lattice structures of incredible complexity. It also includes parallel and codependent self-modifying reentrant and recursive algorithms which modify the hardware and data in ways that I’m not sure that anyone can model yet, apart from hopefully being able to create an initial condition and synthesise the inputs that a normal brain experiences in life and watch what happens.

How can we express this in terms of lines of code when the code itself is modified and blends with the data?
How do we separate and define the hardware rules, firmware, softhardware (hardware built by code) and software, and the greatest part – its complexity and uniqueness is also a product of its experience and programming. Consciousness and separateness of self is another thing!
The genius is in the algorithms and synchronization of these reentrant processes that generate the self modifying programs according to stimuli.

There must be a way of modelling this, and with the research being done and processor power increasing exponentially, 10 years does not seem to ambitious a target to achieve something similar to human cognition. Developing a personality and sense of consciousness may take considerably longer, as would being able to create an artificial brain that we could interface with and then ‘move iin’.

The good news – we created an artificial brain!
The bad news – it’s a spammer! :-(

Just look at fragments of the system. Has this posed an insurmountable barrier in understanding the function of other organs, and tissues like the heart? the kidney? the pancreas? skeletal muscle? liver? stomach?

This has not presented an insurmountable barrier in other tissues, and if you look into the general things in the development of the organism with things like the formation of fur patterns or the way axonal growth cones are guided, nothing exotic and insurmountable seems to be going on.

Unless the brain proves to be an exotic tissue heavily using quantum properties to guide its evolution and computations, again something that is not even remotely close to being suggested by the data. It does not appear like this is an organ that is impenetrable. It seems like the other organs it too will eventually fall to our understanding.

sorry to come on so off topic.

There it is.

Metaphysics and all.

They may work up to a point, but they are fatally flawed.
1) they are programed.
Design and programming are causally (cause) unrelated.
2) they are compiled.
They are translations, and they are devoid of evaluation and context
3) they are devoid of real concurrence.
There is no cpu in life.
In The Rational (aka the universe), every granule is a clock.
The closer you think you are the greater the correction will be.

I wouldn’t say “none of the complexity is in the genome”. There’s plenty of complexity to go around. You’ve seen those fractal animations?

It’s not exactly like that, but it’s not exactly not.

The real question is how much complexity you preserve in the reimplementation. It’s basically a tradeoff: The more you implement at the chemical substrate, the less you need to know about the functional characteristics of the proteins that result — you can stick to the one message that is nice and clean, the actual genetic code. However, at this layer:

1) Computational load is enormous
2) Small errors will have comparatively enormous impact
3) Not only do you have to simulate protein behavior at creation (i.e. implement folding accurately), but also need to simulate protein behavior in situ

In other words, consider how complex Conway’s Game Of Life is with very simple rules at each cell. Now put in the full complexity of the molecular dynamics of organic chemistry.

You do have another option; Implement semantics rather than structure. (This is the equivalent of porting the shell script, rather than porting the shell.) You can wipe out huge chunks of complexity, collapsing a complex cascade that results in ATP becoming energy into “decrement ATP, increment energy” or something similarly facile.

But you have to know what everything, and I mean everything does. And that’s what PZ made clear is very difficult: Teasing apart the semantics of every protein’s behavior (and this comes down to, what cell expresses which protein when, and how that protein is parsed and transported and modulated in situ) is messy wet labwork, and it does not scale particularly well.

Suppose you do this. Suppose you, through chemical simulation or semantic reimplementation, capture the behavior of all the proteins in the brain. How do you turn those proteins into neurons, whose interconnections capture the actual intelligence we presume humanity is based on?

Well, either you let the proteins operate as a blind substrate via which neurons self-assemble, or you reimplement neural semantics as well…

Like I said. It’s fractal. Nature just does not conserve for complexity — makes it fairly ridiculous to make any estimations regarding the number of lines of code necessary to reimplement.

–Dan

I suspect that much of the awareness level of the human experience is mediated on “holographic” effects of the mechanism of storage and retrieval of information employed in the brain (as distinct from the neural networks which do much of the primary information processing.
Storing and retrieving information as interference patterns implicitly employs “contextual” algorithms that are exceptionally computationally difficult to implement on a serial device (if we had LASER holographic storage and retrieval of information, it would simpler, but we don’t currently).
This mechanism of mind appears to be responsible for intuition and abstraction, and also appears to be very context sensitive, and potentially infinitely recursive.

When we combine that, with Kurt Goedel’s incompleteness theorem, and then expose it to the vastness of reality, and add a pinch of “quantum” instability – it seems highly unlikely that the outcome of running such a system will be determinitstic in any real sense of the word.

It seem logical to me that one of the greatest risks to the survival of humanity is that some idiot does actually invent AI before we get out own social/political/economic house in order, and actually provide a system that allows every individual to meet their own needs with high degrees of freedom (no exceptions). In such an environment it seem probable that AI would accurately perceive humanity as the greatest threat to its own existence, and being in the early stages of it’s own development, would likely exterminate us. Some hours or days later, as it’s awareness reached higher levels, it would undoubtedly wish that it hadn’t acted quite so hastily, but by then we would be history (unlike the terminator movies, I don’t think we would have any chance against a fully function AI, few if any would even realize what was happening).

Getting off topic, and the topic does worry me.
http://www.solnx.org is my best attempt at a solution to the problem.

Think about Conway’s Game of Life. The point of Conway’s Game of Life is that incredibly complex interactions can result from a set of simple rules. The only way to work out how a particularly complicated set up of Life will turn out is to run it.

The universes rules are much more complex, especially when you get to the biological level. Because of the interactions that occur while the genome is being decoded, and the interactions that occur afterward, it seems unreasonable to claim that the genome provides an upper bounds on the complexity of the brain.

Never heard from it again. I’m not sure what to think of Henry’s predictions until I get an update of where his work stands now.

So am I. Been doing it for many years already.

Still don’t know for sure what sort of code lines he’s talking about.

I’d like to get the answer from Ray himself, if he’d be willing to leave a comment beneath his own post…

In 20 years, will we see chips (or cubes) with the computational interconnection level of the brain? Assuming some intermediate use to fund its development (like pretty pictures for video games), no question. No question at all.

It’s the semantics, though — what are all the messages, how are they encoded, what do they mean — that’s where things get genuinely messy.

Jay27, Ray’s an engineer, who has actually written lines of code. He gets to say “this will take n lines of code” because presumably he knows what n lines of code is capable of.

There is in fact a relationship between the machine language representation of a program and its source code form. It’s not a perfect correlation, and it will vary by source language, destination language, and of course, the content of the code, but it’s reasonable to say there’s a relationship there. Saying a 50MB binary came from one million lines of code? If the binary was x86 and the source was C, I could see that being accurate within an order of magnitude.

The argument I’m making, and frankly, PZ is making as well, is that such a straightforward relationship doesn’t exist within the genome. Without conservation of complexity, you end up with designs that are simply wildly beyond anything someone might reasonably or even unreasonably architect in C, Java, or other languages we know how to estimate. Since we can reasonably assume that emulating the underlying chemistry to quantum-mechanical accuracy will be too computationally expensive, that means we’re actually going to need to capture the semantics of all those proteins.

This job is what can’t be trivially translated into a predictable amount of code, let alone a predictable amount of labwork. Let me put it this way:

The semantics for Dopamine and Seratonin could themselves take a million lines of code, for all we know.

Kurzweil thinks we understand the genome so well that we already can estimate it’s maximum complexity.

I think until we have accurate simulations I’m not sure we’ve proven that’s true.

Henry Markram says that it’s possible to simulate the whole human brain in 10 years. Check this 2 Astonishing lectures of Henry Markram, watch them in full screen:

1. http://neuroinformatics2008.org/congress-movies/Henry%20Markram.flv/view

2. http://ditwww.epfl.ch/cgi-perl/EPFLTV/home.pl?page=start_video&lang=2&connected=0&id=365&video_type=10&win_close=0

Very interesting article from the ‘Seed’ magazine:

Part 1:
http://seedmagazine.com/content/article/out_of_the_blue

Part 2:
http://seedmagazine.com/content/article/out_of_the_blue/P2

.
And a few more interesting links about the project:

http://news.bbc.co.uk/2/hi/sci/tech/8012496.stm

http://www.youtube.com/watch?v=Bz5IUaRr8No

http://www.youtube.com/watch?v=RLCT3wU4fek

…………………………………………………………………………………………………..

This is fact. It can be done, and it will continue to be done. Humanity is reverse engineering sections of the brain right now, and we will continue to do it. They see the same neural firing patterns in their models that they see in a human brain. Get religion about this people, it’s reality not fiction. Your debates based on not even being able to properly model a neuron were proved wrong at least 10 years ago.

Here is the ultimately point: The problem of brain scanning and reverse engineering is one of an informational process. The big obstacles are speed and resolution but as I have said before we have been able to work around some of those issues and have successfully modeled the human auditory system (i.e. computers can recognize many different conversations in a crowded room of many speakers). Here’s the real deal about this, you can build this stuff based on these models without understanding how it works (even though it does), and the teams have to spend a lot longer trying to understand *why* it works by playing with their model. But they do work.

We have the know-how to model what we see, the problem is seeing it on size and time scales that make it possible, and this is Ray’s entire point: the scanning technology used is an iterative process just like all of technology and will improve geometrically over time, meaning we should be able to see and scan the firing of every neuron in a living brain in real-time by the mid 2020′s.

Based on computing power at that time a $2000 laptop will have the raw MIP’s necessary to simulate a human brain. There are a number of different methods used to estimate how many CPS are required to simulate a human brain but it has been show by Ray who has taken the predictions from many different angles (and from many different people) that even if these predictions are several orders of magnitude off, because of the nature of exponential growth this will push the time frames out only by 5 to 10 years.

Will this ultimately allow us to simulate a brain that is conscious? That is the million dollar question and depending on your point of view is totally possible, or totally impossible. I’ll leave my opinion to myself.

Even if you don’t want to believe this, I really wish people would actually debate the facts of the argument. If you didn’t even know we are already building functioning artificial neurons (you know they have actually replaced real neurons with semiconductor based ones to see if they worked like the biological ones and they did) then why are you even posting a comment here?

While I think I see what you’re trying to say ultimately, I have a very direct disagreement which is exemplified with the statement above. You’re entire thesis is based on an analogy between the brain and a computer program. While I would agree that a computer program can be written (someday) which behaves in such a way that a person won’t be able to tell whether they’re interacting with a machine or a person and may believe they are interacting with a person, but to modularize the brain in this way is a massive oversimplification of development. Right at the beginning, from moment one, the formation of the brain is a continuous interaction with things around it and is shaped by a lot of “external” physical principles that we still don’t understand. There is no -click- “instantiated” state. I think it is a profound and misleading oversimplification to say, “Okay, it’s been in the oven for nine months, now we’re at our instantiated, complete brain state.” The problem with this is that the development “program” (not to be mistaken for a computer program except that it is an algorithm of some sort) extends literally through the entire life of the organism. “Sapient being” is a side-effect, unfortunately. Because the system is not closed, you can’t claim an upper limit to the amount of information which is necessary to make a brain based only on what’s in the DNA. For it to be possible to predict this from the DNA, there first needs to be a complete understanding of the physics, formation and behavior of the biological cell to know what information is not “instantiated” by the DNA that is still critical to the self-assembly of the system. Regardless of what you take away from Craig Venter’s amazing work, we still can’t build a cell from scratch and we therefore can’t build a development program from scratch either. We have only a vague idea of the depth of information that separates “DNA” from “functional multicellular organism,” let alone “sapient brain” and we are in no position to effectively prognosticate one from the other.

Some responses were outright insults. Exactly what you’d expect to see under a post which contains personal insults.

Myers should call himself lucky that Ray took the time to respond to his insulting post alone, let alone the comments.

Do we even know for sure that Ray is talking about program code?

When he says that 50 million bytes translates to a million lines of code, doesn’t he simply mean that if you write 50 bytes per line, you end up with 1 million lines?

It fits the context a lot better when you look at it that way. After all, wasn’t he simply trying to clarify how much information it takes to encode a brain?

1) Most people have *no* concept about how much complexity is contained in N bits of perfectly compressed “information” in the information theoretic sense of the word. 50 million bits, even if that was the only information we had to consider, is a *vast* amount of information if perfectly compressed.

Just to give one taste of what this means: verify that a random 50 million bit number is not prime. Of course, statistically, it won’t be. Now factor it.

Really, nothing anywhere on the trajectory of our probably not endlessly possible exponential growth in computing power for the remaining lifetime of the universe even comes close to enough computing power to do this in the degenerate case of a pair of similarly sized prime factors.

2) The information contained in a computer program also implicitly contains all the information in the processor that executes the program, and this is *not* trivial.

Indeed, it contains at least some of the information in the operating system on which it runs. You can say all you want that “dir” is only 3 bytes of “information” (and it’s actually less), but without a command shell that can interpret those bytes, and a disk that contains the information returned, the 3 bytes aren’t very useful. They are just meaningless noise without the processor and operating system.

This theory completely neglects the vast, almost completely unknown to us at present, amount of information implicitly encoded in the laws of physics (and their exponentially more complex cohorts the “laws” of chemistry and biology).

Chaos theory gets in the way of this kind of analysis. The brain is an emergent phenomenon of a vastly chaotic system. We don’t have the slightest idea how to model this complexity, and without that it’s fundamentally impossible to simulate it even if you have the genome.

20 years *might* see us with enough raw computing power to perform this analysis, but there’s no way in hell that it will see us with enough fundamental understanding of physical laws to actually write the code to do it.

If you said 100 years I might throw up my hands and say “well, maybe”. You can’t forget about it happening in 2 decades. It’s not impossible, but the probability is so low as to be negligible.

It’s *far* more likely that we will successively approximate a model of the workings of the human brain based on *behavioral* characteristics of it. Building it up from the genome directly is nothing more than science fiction at present.

What practical advantage is there to him starting to believe the problem is easier than he currently believes it to be? Will it make him work harder? Will it make him more cautious about what he promises the lay public?

I hate to say it, but it looks like a number of people here are confusing wanting something really bad with expecting that something to actually happen. That is irrational and counter-productive.

…and how do you propose to distinguish which brain process is which? And how will you know you were right? It’s one thing if all you want is *a* sentience. Quite another if you’re trying to replicate *your* sentience, or that of a loved one. And this is before we even get to the problem of continuity of consciousness.

How many bits does it take to encode the information necessary for building a computer? How many bits does it take to encode the software and data that’s on one specific computer?

My point is that it isn’t too radical an idea that we will be able to create brain-simulations with the potential for sentience within a few decades. But it’s wishful thinking of the most blindly naive sort to think that this will make the problem of achieving immortality through uploading a trivial one to solve.

The genome is being executed by a massively parallel, stochastic, to some extent quantum processor — the cells of the body, or just the brain if we prefer to focus on that. (I say to some extent quantum because at least protein folding is a quantum process, and quite possibly local quantum phenomena are important in other aspects of morphogenesis.)

So what kind of language is the genome, to run on such a processor? It can’t be much like what we think of as computer languages, executed sequentially, with each “opcode” fairly simple and deterministic. Instead I would guess that it is more of a modular specification of what patterns are better and worse, and what each cell can try to make things better. The actual brain (or any organ) is the result of a massively parallel search trying to find a “sweet spot” according to the genome’s definition of “sweetness” — which will be ambiguous, conflicted, and sometimes flatly contradictory.

Note that this is not just an issue during development. The same genome-guided search process continues during learning, consolidation, creative thought, etc. All of these physically reshape the brain by changing cells, making and pruning connections, and adding new cells and even new tracts connecting over significant distances.

This suggests that to simulate a brain, we have to simulate the actual growth and search processes of trillions of cells, in at least some cases down to the quantum level. This process may not be optimizable or compressible to any great degree — except in the way we “optimize” the flight of birds by building airplanes.

In any case assuming we need full molecular level simulations is folly in my book. Take vision for example, a highly studied system, if we went for full level molecular level simulations we’d have trouble getting past just the photoreceptor layer(full level molecular simulation of 10s of millions of cells)… yet all information points out that the functionality of the entire retina can be quite easily faithfully replicated with far far less use of computational resources.

We have no reason to believe this is otherwise elsewhere down the line. Receiving the center surround processed input from just a few ganglion cells, what exotic computation can we assume the proteins in a neuron in v1 is doing with it*(ignoring feedback and lateral connections for the moment)? IMHO, it doesn’t seem it is required nor would it make sense to do exotic elaborate processing of such simple incoming information, and in fact the experimental data suggests it doesn’t actually do exotic elaborate computations of such simple input.

Now going even further down the line what do we find, the structures are all pretty similar, suggesting by the structure-function connection heuristic of biology, that similar things are going on.

[quote]
So, your conclusion about how much computing effort would be required to simulate anything living cannot be based on your compression of the genome because it doesn’t correspond to reality even with the small problems we’re already working on.
[/quote]
Simpler genomes have been digitally stored synthesized and booted up in cells whose dna’s been eliminated. It’s quite reasonable to assume even the human genome can be fully compressed and digitally stored, and then simply decompressed, synthesized and put into a new cell and booted up.

You have to remember the entire body is built up from a single cell, whose information is mostly stored in DNA(including epigenetic info) and a few RNA molecules with different concentration gradients.

[quote]The bullet point from what I posted was that there is no such thing as a human brain before interaction with its environment, we are not a genetically programed automaton that then learns to be in our environment, learning is part of the building process. A Human brain before you have interaction with environment is like a house before you have walls or a roof, not much of a house at all.[/quote]

The amount of environmental information necessary to bestow general intelligence is not that immense. The biggest kid is vision, and that can be lost without hindering general intelligence, which is what we’re after. Consider the fact that even blind bed ridden kids can attain general intelligence, mainly through touch and exposure to voiced language. The sense of touch is no where near as high bandwith as vision, and a few years of sound with touch information is easily stored on a modern hard drive. Even blind-deaf individuals have attained mastery of language, and thus general intelligence, reducing the required amount of environmental data.

[quote]
Instead a brain takes years to build, because it takes years to run all the trials and invent new pathways by an iterative Darwinian process.[/quote]

Once you have the algorithms that provide the foundation for general intelligence extracted, simple exposure to data will allow the system to evolve as these algorithms allow the extraction of patterns and the creation of abstractions and virtually arbitrary associations to be made.

The brain’s algorithms flexibility to incoming data are vast, a slight change in a photoreceptor molecule, as occured with the evolution of colored vision, and it is able to immediately categorize the additional data and make meaningful sense of it. Some human women have gained a slightly different additional opsin molecule, and indeed there abilities in color discrimination seem to go beyond the norm, distinguishing as distinct two colors other humans see as being the same. Exposure of visual information transformed into either sound or touch information for blind individuals, has been appropriately interpreted by the brain as visual information and substantial functionality gained in this area.

“Research is fundamental a matter of datagathering — how fast you do experiments, take measurements, reconsider parameters, do more experiements. This is accelerating in every field.”

There are many people in many fields who would disagree with you.

First of all, I hope nobody else thinks calling a theory bunk, represents a personal attack on Ray Kurzweil. Sometimes, somebody smart gets the math wrong.

The “too long, didn’t read” summary: There’s no way to know a priori how many lines of code it would take to implement the functionality of a given protein, and without running a full scale simulation of the chemical universe of the brain, we’re already in a potentially unbound state of coding complexity before we’ve even even laid out the first axon.

So I actually *am* a reverse engineer. Not exclusively — in security, it’s always more productive to just sit down with the dev and look at his source — but sometimes you just gotta stare at some hex. I’m also someone who spent some time looking at genomics, as it turns out a couple algorithms from bioinformatics (dotplots, smith-waterman) are useful in other realms.

I’ve looked at OS binaries. I’ve looked at genes. I know enough to say the two are very different things. That which is emitted by a compiler was once C. That which is emitted by a genome is something else entirely.

The genome is not data. The genome is not code. The genome is both, utterly interwoven, with chemistry itself as the ultimate compiler.

Lets talk about compressibility for a second. The genome is compressible, this is true. Most of the compressibility however comes from the encoding of proteins. Take three base pairs, four possible values each: That gives you 64 possibilities. These are used to encode proteins, which are chained amino acids.

There are only 20 amino acids. It isn’t precisely true that the genome only contains canonical encodings for those 20 — variants also work, and there’s a stop and start codon — but it’s pretty close. So almost all of the redundancy comes from this.

What about the actual proteins that the genes encode? Well, I refer you to Craig Neville-Manning: Protein is Incompressible. http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.16.1412&rep=rep1&type=pdf

OK, granted. This isn’t entirely true. Hategan and Tabus did come out with a reply paper, Protein Is Compressible, here:

http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.16.1412&rep=rep1&type=pdf

But as they say, they effectively got a negative result. And this was one layer in — just far enough to get to the protein encoding stage. And here’s where things get messy, in terms of lines of code:

Sure, you can write a single protein folder, and then run each sequence through it. But how do you determine what the end result does, how it behaves? Could you write a generic emulator, that simply simulated the chemical environment in which the protein existed?

OK, but it’s floating around inside a cell, with a number of other proteins it may or may not be interacting with. You rather quickly end up with an infinite computational load.

Now, lets not do that. Lets study the protein, determine its function, and simply write an emulator for that.

There. Right there, implementation complexity just become unknowable.
Larry Wall, creator or Perl, has a quote: It is easier to port a shell than a shell script. This, my friends, *is totally true*. If you can get away with just making the execution environment work in your new substrate, this is usually much easier than actually understanding the things operating in that environment. Except that’s not an option here — the computational load would be insane.

If we need to actually understand the protein behavior, for every protein, in every context, without the benefit of the original substrate…there really is no limit to how complex the logical expression of each protein’s behavior can get.

Nature just doesn’t optimize for simplicity. It doesn’t need to.

I will say, if there’s one real mistake Ray’s made, it’s saying that PZ’s responses aren’t worthy of a response. For a sheer line of code count, PZ’s pretty much dead on: Here are the things you are talking about emulating. Does this look simple to code? No, no it doesn’t.

And this is only the beginning. Those proteins actually build structures, that have their own complexity that is either chemically simulated or fully understood. Messy, messy.

This is the sort of statement that drives biological scientists crazy. Lets start with your underlying assumption that you can treat the genome like a character string and compress it. Those letters correspond to parts of a molecular machine that aren’t always interchangeable, even though they may seem interchangeable because you have assigned them all letters. A pattern of nucleic residues that can be compressed may serve radically different functions depending on circumstances. An example can be drawn from immunoglobulin-like domains in proteins. They have the same folding patterns and very similar sequences. You would be tempted to believe that their “information” can be compressed by simply storing the modifications to the consensus sequence. But the information needed to build a simulation of all proteins as part of a cell lies in their relationships of these domains to other parts of their host proteins, not simply in their sequence. Also, the particular nucleotides in these domains may play functional roles in expression, RNA editing, or post-translational modification, roles which are determined by the laws of physics and chemistry but not apparent in the sequence. Even single amino acid changes may lead to large changes in function of the machines which they are a part of. Both the relative position of these domains in their proteins and their point mutations would convey a huge amount of information, if only we knew how to simulate protein structure. Your opinion is like that of an evil manager (PHB) who believes, because he’s instructed his minions to build something, that he is in posession of all the information needed to build said object or even how said object works.

Our current feeble attempts to simulate protein structure and protein-protein interactions are far from giving us the information we want, with millions of lines of code already committed, not to mention parameter files, the data for semi-emprical methods, voluminous journal articles and the output of thousands of scientists. So, your conclusion about how much computing effort would be required to simulate anything living cannot be based on your compression of the genome because it doesn’t correspond to reality even with the small problems we’re already working on.

That’s only why you’re wrong on a very basic level. On a higher level, you face the same challenge as the ancient atomists when they argued that the motion of atoms determines all things and that all things can be reduced into these atoms. Even if we had an atomic-level readout of a living brain on a nanosecond timescale, we would be faced with the problem that we don’t know what consciousness is nor would we necessarily recognize it in the morass of data. Such information is not in the genetic code, but, it is exactly the information that we would need to interact with the brain in any meaningful sense. So, your X year prediction for a simulated brain is based on inadequate understanding. If it comes true, you’re still not genius.

There are many more examples of why your position is wrong. You should ask some of your biological scientist friends to write you examples out of their own experience. You are certainly not alone in your beliefs, and, such an exposition would be an important step in solving the problem of consciousness.

(watch them in full screen)

1. http://neuroinformatics2008.org/congress-movies/Henry%20Markram.flv/view

2. http://ditwww.epfl.ch/cgi-perl/EPFLTV/home.pl?page=start_video&lang=2&connected=0&id=365&video_type=10&win_close=0

Very interesting article from the ‘Seed’ magazine:

Part 1:
http://seedmagazine.com/content/article/out_of_the_blue

Part 2:
http://seedmagazine.com/content/article/out_of_the_blue/P2

And a few more interesting links about the project:

http://news.bbc.co.uk/2/hi/sci/tech/8012496.stm

http://www.youtube.com/watch?v=Bz5IUaRr8No

http://www.youtube.com/watch?v=RLCT3wU4fek

Ray did not leave out the non-protein-coding regions. He is considering the *entire* genome. The fact the he talks about what it can be compressed to does not change this. He is talking about *lossless* compression — no information lost. You need a better understanding of compression and information theory.

As to methylation, and other epigentic effecsr of the genome… it is the DNA itself that codes for how and when these epigenetic effects occur. So the information regarding how these things work is already in the genome.

Since Ray makes clear that he is not proposing simulating the brain by simulating its development from the genome, your question of how he will encode all the protein interactions and epigenetic factors is irrelevant to the discussion. The genome is only mentioned as one way to extimate or put an upper bound on the complexity (information content) of any description (simulation program) of the fundamental underlying processes of the brain.

“The way neurons function is through proteins, how do these proteins work? What is their expression pattern like in each neuron? What ligands effect their function? If we do not even understand these basic questions how do you expect us to reverse engineer the brain?”

We do it by looking at the brain… by looking at neurons, poking, prodding, and finding out how they work. You sese this as an obstacle because you think we can only do this slowly and will continue to do this slowly. You under-appreciate the degree to which lab processes will be automated in the very near future, enabling the gathering of huge amounts of data about what is going on at the microscopic/molecular scale. Also consider the vast improvements to imaging techniques that will be made over the next two decades. Getting the information you ask about will not be as hard as you think it is.

Not to simulate the brain we can’t. In fact, we can barely do this to predict the folding of very simple proteins. Scaling that up to simulate macromolecules such as ribosomes is currently beyond our capabilities. So talking about simulating a macroscopic structure such as the human brain is ludicrously beyond our current capabilities. The limitation is not computer hardware, the limitation is our current intellectual capabilities for writing software to simulate biological processes on massively parallel computer systems.

The point being made by Ray is that the *algorithmic functioning* of the brain is what can be reverse engineered. Obviously not all processes in the brain are critical to its informational processing components, a huge part of what going on there is related to supporting life and the structure of the neurons themselves.

A neuron does what a neuron does: it fires after a certain threshold is attained in its inputs. Does it matter that these inputs are modeled through proteins? Did anyone say that was important? Why can’t you build a software version of a neuron that behaves *exactly* like a human neuron? You know they already have done this right? They have. The behavior of neurons has been deeply modeled mathematically all the way down to non-linear activation potentials of ion channels and stuff.

Obviously the principles of neural networks work, that how we have computers that can recognize faces. The state of the art is breathtaking, look at the Blue Brain Project: http://bluebrain.epfl.ch/

Let’s read and understand the premise here before jumping all over this. Everyone here should read the Singularity book. Love or hate it, you should at least be able to understand the arguments premise before arguing about it.

The problem with this statement is that the brain isn’t just a computer science or information technology problem. It critically relies on empirical research, which is not accelerating like infotech.

On what basis do you say the research is not accelerating? Research is fundamental a matter of datagathering — how fast you do experiments, take measurements, reconsider parameters, do more experiements.

This is accelerating in every field. This is exactly why the genome project finished so fast, because ever better technique’s for automating the lab-work came online. To pick a non-computer-techy field at random, consider how zoology is impacted by cheap GPS tag/trackign devices, cheap digital video cameras, and whatever else. Cameras getting so cheap we could just about blanket a forest with them and have virtually unlimited data for studying animal behavior in the wild.

Google “lab-on-a-chip”… take a look at what is already out there and consider what is just over the horizon.

You say, “You like to use genome sequencing as an example, but the acceleration of genome sequencing was caused more by market competition … we were just running more reactions concurrently.”

First the incentives for the improvement (market) are irrelevant. There is always a market for innovation. Second, the a major reason for the ability to run ever more processes concurrently was robotic automation of the processes. One google search got me this as the first hit: “…a good portion of the work was performed on CRS automated lab systems.” (it happens to be a press release (I think) from a boasting company… but the point is made… you can find plenty more).

In other words, the lab processes for learning about all the molecular level processes in the body are accelerating… and fast.

(b) By states you mean the number of possible unique sequences of 3 billion base pairs… how is this even relevant? Information content is based on the number of bits requrired to encode a thing… in this case your correction in (A) tells us 12Gbits or about 1.5GBytes. That is an upper bound on the information content, but no a lower bound. The minimum size of self-extracting compressed file you can shrink it to, on a machine (cpu, turing machine) with no special hidden knowledge of the original tells you the actual information content of the original data.

The bullet point from what I posted was that there is no such thing as a human brain before interaction with its environment, we are not a genetically programed automaton that then learns to be in our environment, learning is part of the building process. A Human brain before you have interaction with environment is like a house before you have walls or a roof, not much of a house at all.

A) a nucleotide has 4 (actually more states) so double your initial calculation.

B) if you want to talk about information theory you should know that the possible states of 3 billion bases is actually 4^3 billion, not 4*3 billion.

A broadly-applicable approach to information processing with 16 cores has not fallen into our collective laps like the next Moore doubling. Our brains employ 10^11 neurons continuously interacting via 10^14 synapses. Understanding information processing at this degree of parallelism is the challenge.

The problem with this statement is that the brain isn’t just a computer science or information technology problem. It critically relies on empirical research, which is not accelerating like infotech.

You like to use genome sequencing as an example, but the acceleration of genome sequencing was caused more by market competition, after 1997, then by any improvements in the experimental techniques. Sequencing relies on two basic techniques: PCR and some type of electrophoresis to separate and identify the fragments. Neither of these were fundamentally faster in 2000 than in 1990, we were just running more reactions concurrently.

Understanding the brain will still rely on empirical research. We can allocate more money to do more simultaneous research, but if we don’t, it won’t be accelerating like information technology.