Slashdot Mirror
Mathematical Model Suggests That Human Consciousness Is Noncomputable
KentuckyFC (1144503) writes "One of the most profound advances in science in recent years is the way researchers from a variety of fields are beginning to formulate the problem of consciousness in mathematical terms, in particular using information theory. That's largely thanks to a relatively new theory that consciousness is a phenomenon which integrates information in the brain in a way that cannot be broken down. Now a group of researchers has taken this idea further using algorithmic theory to study whether this kind of integrated information is computable. They say that the process of integrating information is equivalent to compressing it. That allows memories to be retrieved but it also loses information in the process. But they point out that this cannot be how real memory works; otherwise, retrieving memories repeatedly would cause them to gradually decay. By assuming that the process of memory is non-lossy, they use algorithmic theory to show that the process of integrating information must noncomputable. In other words, your PC can never be conscious in the way you are. That's likely to be a controversial finding but the bigger picture is that the problem of consciousness is finally opening up to mathematical scrutiny for the first time."
Nope, just a bad copy of it.
Get free satoshi (Bitcoin) and Dogecoins
But they point out that this cannot be how real memory works; otherwise, retrieving memories repeatedly would cause them to gradually decay.
Memories do decay upon recall. People misremember something and convince themselves that the misremembered notion was correct.
Retrieving memories repeatedly would cause them to gradually decay is talked about in a radiolab episode.
http://www.radiolab.org/story/91569-memory-and-forgetting/
Eyewitness accounts have been proven to be wrong over and over again. The assumption of a non-lossy memory is just false.
Not retrieving memories is what causes them to decay. Ever hear of refresh?
Because I'm a human being and it's a PC. Duh...
I think machines will eventually acquire their own form of consciousness, totally separate from ours. and I reckon it's just fine, and much more exciting in fact than trying to replicate our humanity in hardware that's just not compatible with it.
"A door is what a dog is perpetually on the wrong side of" - Ogden Nash
Baloney. What a stupid argument. Here is it, summarized:
1. Here is one mathematical model of a way that memories could work.
2. This method would be computable.
3. But that would mean memories degrade the more you remember them
4. But memories don't degrade the more you remember them.
5. Therefore memories are not computable.
Assignment for the student: find the flaw in this argument.
http://www.geoffreylandis.com
One of the most profound advances in bullshitting in recent years is the way researchers from a variety of fields are beginning to misuse mathematical terms in order to give their ideas a facade of intellectual responsibility. Since no one has yet come up with an agreed-upon definition of what this "consciousness" is as an objective observable phenomenon, trying to talk about it in mathematical terms is nothing more than intellectual masturbation.
Tom Swiss | the infamous tms | my blog
You cannot wash away blood with blood
Ouch. Just. Ouch. No. Noooo. NOOOOO.
There is so much wrong with this statement I don't even know where to start. It implies that the memory is overwritten with the memory of recalling the memory, which is a huge and ridiculous assumption. Memory likely works much more like ant paths. The details that are recalled more frequently are reinforced, and can be remembered longer. It could also be compared to a caching algorithm; details used more often are less likely to be lost, or need fewer hints to retrieve them.
And then using this assumption to declare something as non-computable demonstrates a lack of understanding of the concept of computability. The only way that conciousness could be non-computable would be if there is a supernatural element to it. Otherwise, the fact that it exists means it must be computable.
No. All Turing complete means is a universe Turing machine can execute anything any other Turing state machine can. People misunderstand "Turing complete" can think it means someone that is Turing complete can do "anything." That is NOT what it means.
Anonymous Cowards suck.
That allows memories to be retrieved but it also loses information in the process. But they point out that this cannot be how real memory works; otherwise, retrieving memories repeatedly would cause them to gradually decay.
I remember hearing a radiolab episode on NPR talking about how memories actually get modified every time you recall them.
http://www.radiolab.org/story/91569-memory-and-forgetting/
Maybe the radiolab episode is completely wrong, but I don't think it's fair to assume memories are lossless without providing some evidence of this.
Everything is computable given the right models and starting conditions.
"Does the Turing machine with a given description halt?" That's been proven not computable on a Turing machine. And we lack a model more powerful than a Turing machine.
Hmmm. Do you find yourself occasionally having to re-learn your address or phone number ?
There seems to be a step missing from A (that's not how memory works) to B (therefore uncomputable). The premise that memory isn't lossy sounds like rubbish, even IF it's perhaps not so simply a question of 'read errors'
I recently watched this talk, Modeling Data Streams Using Sparse Distributed Representations, which seems to be able to represent memory in a layered and lossy way perfectly fine in a computer.
Belief is the currency of delusion.
Here's a a critique. (It's on arxiv; no need to sign up for "Medium")
The paper isn't impressive. It make the assumption that human (other mammals, too?) memory isn't compressed, and is somehow "integrated" with all other information. We've been through this before. Last time, the buzzword was "holographic". We've been here before.
The observation that damage to part of the brain may not result in the loss of specific memories still seems to confuse many philosophers and neurologists. That shouldn't be mysterious at this point. A compact disk has the same property. You can obscure a sizeable area on a CD without making it unreadable. There's redundancy in the data, but it's a lot less than 2x redundant. The combination of error correction and spreading codes allows any small part of the disk to be unreadable without losing any data. (Read up on how CDs work if you don't know this. It's quite clever. First mass-market application of really good error correction.)
Stop it. Just stop it, people.
Memory doesn't work that way. It's a live feedback loop that reinforces itself through the conscious mind. There is some lossy drift but stuff that maps to the real world is indeed corrected if lossily. Ancient stuff from when you were a kid (Gee, what did Koogle taste like) drifts and drifts.
Something from when you were a kid,
like Orange Julius taste, drifts but may suddenly be reset when you stumble across one at a mall somewhere (or Dairy Queen, whoever bought them). His model is a solution to a problem thatsn't a problem. It doesn't matter how clumsily intertwined actual brain processes are for this.
Furthermore, he conflates consciousness with deep thought. I could grant his proposition of complexity yet it would not matter one bit for the subjective conscious experience. The subjective perceptual experience may still be magic w.r.t. grounding in real physics, but it is there and not some.purely informational process (i.e. Searle is still undefeated) and there is nothing requiring consciousness to be synonymous with all this complicated brain activity.
Your unconscious mind does the vast bulk of difficult processing, then passes it through consciousness for some kind of review.
There is no evidence consciousness, however miraculous and awesome, need be particularly complicated in and of itself, nor is "what it does" as part of your larger, largely subconscious thought process, particularly valuable.
From an importance point of view, it is vastly overrated as information processor.
Your thinking, in other words, could be supra-Turing in computational model, yet the consciousness itself perfectly mundane, experencing these supra-Tuting-generated thoughts and doing a computationally mundane thing with them.
(-1: Post disagrees with my already-settled worldview) is not a valid mod option.
Then you're in science fiction land...woo hoo! I like scifi as much as the next /.er but your imaginations of the possible existence of a civilization that can fully digitize continuous data is worthless to a **scientific discussion**
That's the problem. Hard AI, "teh singularity", and the "question of consciousness" are so polluted in the literature by non-tech philosophers throughout history that the notion of ***falsifiability*** of computation theory get's tossed aside in favor of TED-talk style bullshit.
Falsifiability kills these theories *every time* and hopefully this research in TFA will help break the cycle.
To be science it must be able to be tested. It must be a premise that is capable of being proven or disproven. "hard AI" proponents like Kurzweil and the "singularity" believers ignore this part of science.
So happy to see this research
Thank you Dave Raggett
Your cite of the "recent" study fits with my memory from the old school. There are several kinds (at least two) of memory: long- and short-term; one is chemical, the other electrical. Each reference to the protein carrying the memory rewrites it to include the information from the new conscious understanding and context, thus changing the protein when it is recreated. I am surprised that this method of decoding/recoding has not been looked into.
Are brains Turing machines?
Then it's magic. If the brain is formed by neurons that work within a certain logic and mathematical model, then it's computable.
Repeatedly recalling an event, as for story telling, restores a subtly _altered_ copy of the memory. This has been shown by many experiments about the plasticity of human memory.
I hardly understand a goddamn word of TFA and have never heard of the "Integrated information theory", but I know that TFA's proposition must be false because the brain is based on the laws of physics, which are computable. Q.e.d.
Wrong. Your short term memory can remember, without effort, 7 digits easily. It therefore is not so complicated that it changes before you put it back.
Remembering the order of events over several days, on the other hand, does not fit in short term memory in a coherent fashion. So you may gradually put several things on a week long trip on the same day. When telling the story, someone else who was there says "No, it was the next day, because [reason]". You didn't have a forceful enough memory to record a separation, and they gradually blur the line while you remember, and now they are on the same day. Something made an impression on the other person that put it on a different day - maybe the same day as something you remember happened on a different day.
There are lots of other little things, like pattern recognition. If someone's number is nearly the same as yours, you may repeat theirs when saying yours, because you call them more than you call yourself. And then you have to struggle to recall your own number, and it is vulnerable to change in that moment. But if you're not sure, you pull out your phone and problem solved - you store it correctly. There's more, but I don't see you coming out of this any better if I keep typing.
The argument for compression describes essentially Kolmogorov Complexity. The idea is that the K.C. of something (and everything can be reduced to a binary string) is the length of the shortest program (if you look at it algorithmically) that can describe that object (reproduce that binary string) and stops. In TFA, the example is reducing the description of an infinite sequence of numbers to a finite program that calculates the odd primes and adds one to each. The number pi is infinite in length and random, but not complex since there are small programs that caculate it; an infinte truly random string of numbers would have infinite K.C. because the shortest program would be "print -infinite string". The K.C. of an object is not computable (it's related to the Halting Problem), essentially you never know if you have the shortest program to describe an object.
So here are some observations
a) the whole premise rests on the assumption that the brain is a Turing complete computer ie. the brain is a computer too. So if the brain is a computer, why couldn't other Turing complete computers mimic it? In fact, K.C. theory uses the idea that there is a Universal Turing Machine that can mimic all other Turing machines. If the brain is not a Turing machine then you can't really make any comments about it's compression abilities, etc. because algorithmic theory is grounded on the Turing assumption; ... it just seems like a pointless example; BUT
b) the TFA implies that compression is lossy. Well, not all compression is lossy and the example provided (prime plus 1) is not lossy at all, it's perfect. So what is the point of that example except to suggest that memory must be perfect compression?
c) the assumption that memory is/must be perfect compression seems extremely flawed. Memory is not perfect and most memory seems to degrade over time (see witness reports, personal experience, etc.)
So ... the whole paper seems riddled with discontinuities or inaccuracies. Really it seems like it would have been better to say:
"The brain compresses information in a lossy fashion. We don't know how. Assuming a Turing process is occuring, then the brain is looking for the best compression it can but it can never know if it has the best or not. A computer will be in the same boat." BUT
THE FEAR
Basically the article is making (a flawed?) claim "Machines can never be conscious." The argument plays very well to a religious and research oriented crowd. First: machines can never be made in the image of man. We are not gods. Second: There is no requirement to consider ethics in AI . No matter what the AI seems, it is not, _can't be_, conscious. Therefore, should you create a robot that walks, talks, acts, and feels like a human ... well, it isn't conscious, so do with it as you will.
"Consensus" in science is _always_ a political construct.
2024 is not a date or time. In the multiverse it is a place.
That year, 2024, is the point in space/time where the natural progression of human consciousness & technology & science converge and we will take a step forward equivalent to the first humans to make artwork or speak language...only this is not an inward step, but an outward one.
Conspiracy theorists talk about "predictive programming" and it's bunk of course, but humanity has known this all along. The parallel is all humans who will be alive during the coming transition & all humanity..."we" have always known something was coming. Books like Childhood's End and films like 2001-A Space Odyssey or Contact are really a primer, like an introduction. Those works intentionally prepare our minds for 2024, even though the people who made it may not be conscious of it!
If you can understand intentionality of will without consciousness then you're on your way to becoming a Pleiadians ;)
Thank you Dave Raggett