▲ 198 points
back
82 comments
The best analogy to this field and the many, many problems it might run into is biotech. This is actually
quite funny because in the early days the main analogy to biotech was the computer / state-machine.
For some background, biotech has undergone many booms and busts in the last half century; unlike the last tech bubble, these booms and busts have flown somewhat under the radar.
In fact, Jurassic Park, written in 1990, is actually a book /about/ the late 80’s biotech bubble. There is a really fun part in the book where they are talking about making miniature dinosaurs that will only eat Ingen-brand dinosaur pet food (Ingen was the company behind Jurassic Park).
Anyway, with biotech, the problem has been, again and again, the assumption that a system is radically simpler than it. Biology is so incredibly complicated that it puts the largest engineering book I have, a 1500 TCP/IP protocol book (No Starch) to shame. I mean at least we know how TCP/IP works. With biology, the manual might be closer to 900,000 pages and we only have 40% of the table of contents and maybe 800 pages (out of order) so far.
BEST EXAMPLE: In the 70’s, once large portions of the genome started to be identified and linked to specific genes that were known to cause traits or diseases, it was widely assumed that creating/reading/updating/deleting genes on DNA would be relatively straightforward, particularly as more of the genome was uncovered. However, as was later discovered, many traits or diseases might be the result of 200+ genes that are also used elsewhere. Turn off just 1 gene for a disease, and 1). it won’t do anything because you didn’t turn off the other 199, and 2). oh wow that gene was actually used for something else and now you’ve lost the ability to form eyeballs / are born without anything in your eye sockets.
> Turn off just 1 gene for a disease, and 1). it won’t do anything because you didn’t turn off the other 199, and 2). oh wow that gene was actually used for something else and now you’ve lost the ability to form eyeballs / are born without anything in your eye sockets.
Certainly not on the same scale, but this resonates with my experiences with legacy code.
I think this similarity is more than superficial. Energetic systems evolve over time to become tangled, correlated messes, without some other force counteracting this tendency (ie. refactoring). I wonder if DNA has analogous mechanisms.
Yeah, in some way it feels that there's essentially a huge software (reverse) engineering project.
We have the technical ability to read all the code in our DNA, understand what small parts of it do (e.g. making a particular protein), and model some of the small scale behavior.
And we've got a very, very, very large codebase of mishmash undocumented legacy homegrown code that sort of does what we want but in an unstable and occasionally buggy manner. And we've got a strong wish to fix some bugs (i.e. genetic diseases) and possibly add some features (e.g. longer quality lifespan, increased capabilities). So we'd like to reverse-engineer this system.
The good part is that we only have to do it once and we can cooperate on it; the bad part is that the system is really complex and (more importantly) horribly interdependent; it actually implements pretty much all the practices that we know makes code unmaintainable.
Anyway. The hypothesis I'm trying to make is that this seems to indicate that research on advanced methodologies and tools to analyze and understand large quantities of tangled (and possibly intentionally obfuscated) computer code; work techniques and algorithms for computer(machine learning?)-aided understanding and reverse engineering large quantities of code seem likely to eventually have practical applications in biotech.
Yes, contemporary code behavior is quite far from protein interaction. That's ok - we're quite far from starting to properly reverse-engineer (in this context) biotech as well; with every decade, code (and its analysis) will become more complex and biotech more understood, eventually meeting. And when designing tools for analysis of very complicated systems, the tools will anyway have to be adapted not to the systems but to the analyzer, to the limitations of what structures the human researchers can understand and "keep in their head" and what needs to be automatically summarized/structured by tools.
> I think this similarity is more than superficial. Energetic systems evolve over time to become tangled, correlated messes, without some other force counteracting this tendency (ie. refactoring).
Well... yes. That's entropy. Any system tends towards disorder.
> I wonder if DNA has analogous mechanisms.
DNA works differently, since there's an advantage to reusing code that naturally leads to spaghetti. Although it's not about minimizing energy- just that it's more likely to get successful code by adding onto existing code than adding a whole new section (unless you transclude it from bacteria/viruses).
There's fundamental limits on understanding computation, where the only way to predict what a system will do is to run it.
Is the unmaintainability of legacy code at all related to this? Is the impenetrability of DNA at all related to this?
Though this is a solved problem for our field, if you're using a statically typed language.
You can't right-click on a gene and "find all references".
Civilization (II)’s “modern” techs are like a time capsule of the late 80s / early 90s.
- Cure for Cancer is a wonder of the world and requires “genetic engineering”
- You can build SDI lasers in every city, and they work. In 2018 the current state of the art is “it will probably shoot down an ICBM, maybe”
- Fusion power and room temperature superconductor are basic research projects. The technology immediately preceding superconductor is ... plastics
- SETI project is a wonder of the world and gives a huge boost to scientific research. The obvious choice now would be to change it for “the internet” but hindsight is 20/20
Part of that is because Civ II is meant to model a large span of technologies over time using a fairly uniform mechanism. It would be infeasible to include the vast numbers of technological branches that have occurred in modern times while not making the game one that takes place mostly in modern times... So it only picks a few representative technologies so show you.
A genetic algorithm designed an FPGA logic that couldn't possible work under usual rules, but actually did and used less resources than any human design. As it turned out, the genetic algorithm exploited some analogue properties of the FPGA, which a human designer would never attempt to do.
Genetics seem to favor efficient systems over understandable and easily extendable systems.
Paper "Adrian Thompson, An Evolved Circuit, Intrinsic in Silicon, Entwined With Physics."
So true about TCP/IP. Even rocket science is a joke compared to most biological systems. That said, even a set of rules as simple as TCP/IP can lead to complex, valuable systems (e.g. WWW).
TCP/IP rules leading to the web is nothing compared to physical rules leading to biological life, true. Even in 2018, one could say that computers' potential has barely been tapped.
the manual might be closer to 900,000 pages
Knowing a thing or two about neurophysiology, this seems like a vast underestimation (since a biology book should also cover that, likely most complicated of all, topic). So vast I wouldn't even dare putting a number on it, because there is so much we don't know about the brain, it is hard to even estimate how much we don't know. (Which by the way also puts some AI claims regarding replication of brain functionality in a different light)
> many traits or diseases might be the result of 200+ genes that are also used elsewhere.
What are examples of 200+ gene diseases? At this level of genetic sophistication it should probably be called a trait which causes disadvantages in some situations.
An excellent book which covers this topic is Permutation City by Greg Egan (1994). If we can clone all of you (physical characteristics, brain, memories, personality) into a simulation, what are all the implications?
In it, a perfect replica of a person is scanned, and when it boots up, it continues it's existence from the time of the scan. Hacking on the model is possible, so you can avoid certain memories, update "physical" appearance, etc.
Given how prevalent virtualization has become recently, I thought it was a fairly modern book. I was surprised to find out it came out in the nineties, and not only has it held up well, it would be considered visionary even if it had been published yesterday.
You should try the Takeshi Covacs books by Richard Morgan. There is a TV show coming out in two days (good timing here) based on the first book (Altered Carbon). It'll probably be a mess since the book has quite some violence. But the thoughts above are played out in a interesting and entertaining way. All three books play in the same universe with the "same" main character but are not a trilogy.
Robin Hanson recently wrote a non-fiction book on the topic looking at what standard economic and sociological models say would happen if this were possible.
I've recently played through the game "Soma" which had a similar concept. Warning spoilers:
The main character goes to a doctor to get his brain scanned following a car crash. After the brain scan he wakes up in a distant future and discovers that his brain scan was used as a sort of programming template and that there are many (modified) versions of him living on in robots. Throughout the game this has many more implications.
Also at the dawn of humankind the earth sends out a satellite called the Ark which houses some kind of simulated reality inhabited by the brain scans of selected humans since earth has become uninhabitable.
Especially since this topic covers about half of the episodes in the most recent season of Black Mirror.
This report by IEEE is actually a very good collection of topics and research currently being done in this field. They discuss specific problems and topics like the neocortex, IIF [1], neuromorphic engineering [2], pose cells, SLAM [3], and more.
For anyone interested in research being done in AI, ML, consciousness, etc., these are great articles written by actual scientists and researchers who are doing the work (as opposed to the hyperbolic articles or tweets you see online these days about AI).
[1] https://en.wikipedia.org/wiki/Integrated_information_theory
[2] https://spectrum.ieee.org/semiconductors/design/neuromorphic...
[3] https://spectrum.ieee.org/robotics/robotics-software/why-rat...
I'd like to add this article (with somewhat of a cheeky title): "The impossibility of intelligence explosion"
https://medium.com/@francois.chollet/the-impossibility-of-in...
It's written by François Chollet, creator of the Keras DL framework. In the article it is shown how the environment and intelligence are interrelated. Some of the points are expressed in the IEEE Special Report as well (sensorimotor integration). There are many correlations with the recent push towards simulation in AI - Atari, OpenAI Gym, AlphaGo, self driving cars, etc. It's a new front of development, where simulation will create playgrounds for AI.
The main point is that intelligence develops in the environment, and is a function of the complexity of the environment and task at hand. There is no general intelligence, or intelligence in itself, only task-related intelligence. An intelligence explosion can't happen in the void (or in a brain in a vat, or in a supercomputer that has no interface to the world, and can't act on the world). The author concludes that AGI is impossible based on environment and task limitations.
An interesting take because we're focusing too much on reverse engineering "the brain" as if it exists in itself, outside the environment. We should learn about meaning and behaviour from the environment and the structure of the problems the agent faces. Meaning is not "secreted" in the brain.
A related 'trend' in Cognitive Science is called Embodied Cognition [1]. Intelligence develops together with the body that it inhabits, and, as you mention, the environment that this body is living in.
Maybe Dolphins are as 'intelligent' as we are, but having fins instead of hands and living in a maritime environment just make it impossible for them to invent fire, printing presses and automobiles.
Sure... sorta. But there is a problem. You see, without continuous input from the environment in the form of highly corellated input... we know what happens to human consciousness. It rapidly dissolves and disappears. Consciousness is intimately and (thus far) inextricably bound up in its embodied existence. Facial paralysis has profound effects upon the ability of people to feel emotions 'internally', eventually resulting in both an inability to even recall ever feeling those emotions or having an ability to recognize them in others. Should a consciousness divorced from a feedback loop created with the same environment we share (or at least similar in most respects, a simulation might work OK, I don't think we know) be created, the odds that we could even recognize it as conscious are very low. Maybe a general measure of the systems tendency to reduce entropy in some region either within or near itself?
We are feedback loops, and when the loop is broken, we stop being us.
Most people would accept being able to see and interact with the world a prerequisite to consciousness. I understand this as an I/O problem. There must be continuous high-bandwidth input (visual, audio, tactical) as well as very detailed and dexterous output.
I wonder to what degree virtual realities can play the I/O function. I think for the next half century virtual realities will mostly operate at a lower level of detail then meatspace. Can a mind function well stuck in a lower detail world?
Alternatively, cybernetics could serve. You bring up a real concern, but it's a solvable problem.
That's not entirely true.
That certainly is how we evolved, but somehow we evolved a consciousness that isn't purely a function of its inputs; people who are "locked in" remain conscious.
That was quite the trick...
No, it's protected under the DMCA. If you wish, you may read the complaint at ChillingEffects.org.
Tom Scott did a video about that — “Welcome to Life: the singularity, ruined by lawyers”
You wouldn't download a brain!
I read something along the lines of files having no meaning until we interpret them with programs. The idea is that files are almost schemaless and we bestow a schema on them by the way we consume them with programs. I believe this may have been a SQL book...
In any case, what does "copying the brain" mean if we don't fully understand how it works? How can we bestow meaning on the raw data, given we can even collect it, without knowing exactly how the brain itself interprets it?
What if there is no meaning. What if "meaning" is just a word we invented to describe whatever our brain does with information.
Ideally if you are copying a brain, you are going to copy the I/O schema (spine connecting to a "body", eyes, ears, nose, mouth, ect). When you run a brain as a computer program you don't want a disembodied mind, you want a person who "really-is" in a VR environment.
I'm pretty sure we know how disembodied brains would work - consider any quadriplegic. They may as well be disembodied, yet they can continue to mentally function at a high level.
Agreed 'copying the brain' would have to copy/checkpoint the software running, or you risk failing to connect meaningfully with the environment. It'd be hard for a deaf/dumb/quad brain to learn to speak or interact from scratch.
Why bother? It's a great general purpose organic computer that you can power with sugar and fit into a head...but there's no reason that some other solution to the problem might be arrived at that works along entirely different parameters. There's lots of evidence that brain design has some deep flaws, extreme limitations and severe compromises. Look at the crazy stacked, hemispherical design of mammal brains -- nonsense!
Besides, if we need a brain, there's literally trillions of them already all over the planet. In fact they're so common they're already commoditized, everybody has one!
Be able to make a brain, hopefully means being able to make a better brain - or even, to make our brains better.
Copy at what level?
Cellular? Molecular? Atomic? Subatomic?
The subtitle appears to be "Intensive efforts to re-create human cognition will transform the way we work, learn, and play." So, functional copying, presumably.
I believe the open worm project works at the functional neural level. That's today's state of the art, who knows what level of fidelity we need to get to to get digital immortality.
When they do “copy” it and successfully put it in a human, it will make Pet Sematary the movie look like Driving Miss Daisy
Hard problem of conciousness
https://en.m.wikipedia.org/wiki/Hard_problem_of_consciousnes...
And one of the linked articles was explicitly engaging with that.
https://spectrum.ieee.org/computing/hardware/can-we-quantify...
We've no idea how neurons actually work but hey, sure, let's copy the whole brain!
Most articles concerning brain recently are some kind of infotainment porn.
Downvoted as you probably didn't click the link and left the flippant comment. It links to a page of interesting articles related to this topic.
Yeah there’s still a lot of fundamental research that needs to be done here. We’re probably 3-4 Nobel prize worthy breakthroughs before we have the necessary knowledge to do such things with brains.
We kind of have an idea of how individual neurons work, the problem is that all the interesting aspects seem to appear in tangled systems of millions (or much more) of neurons, and we don't understand that.
If we want to study the collective behavior of neurons, then we do pretty much need to analyze or simulate large-ish portions of brains.
Identical twins handily prove that copying a brain still produces different people.
It’s true for siamese twins, and it’d be just as true for cloned genetic stem cells 3D printed from a snapshot of the exact cellular structure of your brain from a moment in time, transplanted into your body.
If I copy your brain, you won’t be inside it, even if it’s effectively you, as far as anyone else can tell.
So, I can comfort myself with a replica of each dead parent. But the parents that raised me, those people are dead. They won’t be there to see me sharing christmas with their atomic-precision duplicates.
Virtual simulations aren’t people, aren’t human beings, no matter fidelity of the simulation. Maybe you can hold a conversation with a hologram, but then what?
Would one expect to experience what it feels like to become and persist as a haptic-enabled hologram? Pretty sure that’s a bogus concept, turtles on down.
Why do you assume that identical twins have identical brains?
Brain physical structure is altered during lifetime experience, learning changes large physical properties of neurons e.g. forming some new synapses and pruning others.
Even at birth, identical twins have different fingerprints, different retinas and yes, different cellular structures of their brains. And that only diverges later.
It's not a given that a brain "3D printed from a snapshot of the exact cellular structure of your brain" would be sufficiently good copy, we know that there are other things that matter (e.g. various intra-cell properties and chemical concentrations within particular locations of those neurons) and there likely are some other things that matter but which we don't know yet.
However, assuming that we could make a sufficiently good copy, it would be undistinguishable from you. It would have the same behavior, reactions, memories, skills and understanding; it would have the same beliefs as you do - including a belief that it's you and that it has been you since your birth, supported by memories of living your life. If the copy is sufficiently good, barring the signs of the operation itself (scar tissue? machinery? video evidence of it being done) neither your relatives nor (new?) you yourself would be able to make a distinction.
Yes, you could argue that the new copy is different from the previous one, that it's not the same, that it lacks continuity. But that's arguing about what do we mean when we say "the same". If you didn't know that your parents had died, you'd have no way to tell when their copies come to celebrate christmas with you.
I wonder if person died, and his clone waking up can be compared to a person getting into a deep coma from an accident and waking up a year later. In both cases a consciousness disintegrated and a consciousness was integrated?
>If I copy your brain, you won’t be inside it, even if it’s effectively you, as far as anyone else can tell.
If you wake up tomorrow, will you be inside of your brain, even if it's effectively you, as far as anyone else can tell?
What if your awareness is reconstructed daily from the structure of your synapses? If you copy that structure, that copy is you, in the same way that you are now almost the same person as the one that went to sleep yesterday. (Quite like the movie 'The Prestige'.)