21Mar/079
Relative Advantages of AI and Human Brains
Advantages of computer programs over humans, which some might call, “why we use computers at allâ€:
More design freedom, including ease of modification and duplication; the capability to debug, re-boot, backup and attempt numerous designs.The ability to perform complex tasks without making human-type mistakes, such as mistakes caused by lack of focus, energy, attention or memory.The ability to perform extended tasks at greater serial speeds than conscious human thought or neurons, which perform approx. 200 calculations per second. Computing chips (~2 GHz) presently have a 10 million to one speed advantage over our neurons.The in principle capacity to function 24 hours a day, seven days a week, 365 days a year.The human brain cannot be duplicated or “re-booted,†and has already achieved “optimization†through design by evolution, making it difficult to further improve.The human brain does not physically integrate well, externally or internally, with contemporary hardware and software.The non-existence of “boredom†when performing repetitive tasks.Advantages of human brains over hypothetical AIs:
Present AIs lack human general intelligence and multiple years of real-world experience.The computational capacity of the human brain is estimated at 2 * 10^16, or 20 million billion calculations per second, which is twenty times greater than the supercomputer Blue Gene’s predicted achievement of 10^15, or 1 million billion calculations per second, by 2005. However, the human brain may not have a computational advantage over computers for much longer. Ray Kurzweil, for example, predicts that the computational capacity of the human brain will be accomplished on supercomputers, or clustered systems, by 2010, followed on personal computers by 2020.The human brain has already achieved a high-level of complexity and “optimization†through design by evolution, and thus has proven functionality.Advantages of minds-in-general (AIs) over the human brain:
(The following are not advantages of specific AI approaches, but rather advantages of minds-in-general over the human brain.)
An increased ability to acquire, retrieve, store and use information on the Internet, which contains most human knowledge.Lack of human failings that result from complex functional adaptations, such as observer-biased beliefs or rationalization.Lack of neurobiological features that limit human control over functionality.Lack of complexity that we have acquired from evolutionary design, e.g., unnecessary autonomic processes and sexual reproduction.The ability to advance on the design of evolution, which is continually constrained by lack of foresight, the requirement to maintain preexisting design, and a weakness with simultaneous dependencies.The ability to add more computational power to a particular feature or problem. This may result in moderate or substantial improvements to preexisting intelligence. (AI does not have an upper limit on computational capacity; we do.) Note that the speed of computational power is predicted to continually increase exponentially, and decrease exponentially in cost, every 12-24 months, in accordance with Moore’s Law.The ability to analyze and modify every design level and feature.The ability to combine autonomic and deliberative processes.The ability to communicate and share information (abilities, concepts, memories, thoughts) at a greater rate and at a greater level of complexity than us.The ability to control what is and what is not learned or remembered.The ability to create new modalities that we lack, such as a modality for code, which may improve the AI’s programming ability-by making the AI inherently native to programming - far beyond our own (a modality for code may allow the AI to perceive its hardware machine code, i.e. the language used to write the AI, and other abilities).The ability to learn new information very rapidly.The ability to consciously create, analyze, modify, and improve abilities, concepts, or memories.The ability to operate on computer hardware that has powerful advantages over human neurons, such as the ability to perform billions of sequential steps per second.The capacity to self-observe and understand on a fine-grained level that is impossible for us. AIs may have an improved capacity for introspection and manipulation, such as the ability to introspect and manipulate code, which would be the functional level comparable to human neurons, which we can’t think about or manipulate.The most important and powerful capacity of minds-in-general over the human brain is the ability to recursively self-encapsulate and self-improve its intelligence. As a mind becomes smarter, the mind can use its intelligence to improve its design, thereby improving its intelligence, which may allow further improvements to its design, thus allowing further improvements to its intelligence. It is unknown when open-ended self-improvement may begin.Think about the differences and what they mean. The items on the above lists are not controversial - they’re either known facts or follow directly from the nature of the hardware. It’s the policy consequences that are controversial. But take the time to ignore the policy implications (if any), and by ignoring I mean not commenting on, and leaving this post just as a place to meditate about known differences between human brains, computer programs, and hypothetical AIs.
March 26th, 2007 - 09:19
1. IanC Says:
March 21st, 2007 at 9:12 am
I continue to fail to comprehend why it is presumed that any capacity available to a highly-developed AI wouldn’t be at least ‘nearly’ as available to a human sentience. For example, imagine a nanite sensor system embedded with a prosthetic neocortical system designed to analyze the brain and directly input that information into the consciousness of the human. This is no more or less readily available than is any AI’s ability to modify its own hardware. And as any AI will be optimized to its hardware, the conclusion is inevitable that enhanced optimization will be as readily available to the human as it is to the AI.
Given that Turing-test positive AI is predicted by Ray Kurzweil to be available by 2025, and that companies such as Cyberkinetics posit that they will have full I/O BMI’s at least by then, I have to stipulate that the conjecture on the Singularity has to include at least the possibility of a non AI singularity-style hard-takeoff.
2. Michael Anissimov Says:
March 21st, 2007 at 9:31 am
Many capacities that would come automatically with any AI, advanced or not, would not be available to humans because of their different hardware. To build AI capable of self-modification, you simply give it some way of accessing its own source code. Automatic and simple. Of course, then comes the question of whether the AI can do anything useful with that code, but that’s another topic…
To make a human capable of full self-modification would be a mammoth undertaking. You’ve need to have a way of representing all the internal cognitive dynamics in a way that is understandable to the human operator. You need nanomachines that have read-write access to every neuron in the human brain. You need to write the software that navigates the machines, the engineering that ensures the machines don’t drill through tissue or bump into each other, the software for internanite communication, the knowledge of which modifications give rise to which high-level mental changes, etc. It’s insanely harder than just telling an AI, “hey, there’s your source code!â€
Even stupid algorithms can be run at 10 million times the serial speed of our neurons. This is why computers can calculate huge equations and we can’t. :) It’s not more or less easier to give ourselves that ability. If it’s more or less easy to give ourselves the abilities of software programs, then why do we even need computers?
And as any AI will be optimized to its hardware, the conclusion is inevitable that enhanced optimization will be as readily available to the human as it is to the AI.
The entire bottom list goes into things that more or less would come by default for advanced AI. They’d need special programming to exploit them fully, sure, but there’s no inherent hardware-level limitations, for, say, forking off multiple trains of thought, in the way that there are for human beings. To me, your statement reads, “Anything another mind can do, I can do too.†Why? If you replace “AIâ€, with “aliensâ€, can you do what aliens do too?
Is anyone else understanding what Ian is saying? Am I missing something?
Given that Turing-test positive AI is predicted by Ray Kurzweil to be available by 2025
2029. Not that I agree with the prediction though, or even think that the idea of a fixed date has any merit.
I have to stipulate that the conjecture on the Singularity has to include at least the possibility of a non AI singularity-style hard-takeoff.
It’s definitely possible, sure. But I tend to doubt it, because 1) airplanes came before enhanced sparrows, 2) cars came before enhanced horses, 3) houses came before enhanced caves, etc….
Moral of the story? Nature fails. Why work within nature’s paradigm when we can build superior forms from scratch?
Also, the FDA will not permit brain research or meddling beyond a certain point. One negative incident in the news, and the whole business will be shut down. Do you know how we know so much about the brain? Horrible experiments on other primates, mostly. We selectively cripple brain areas, take tests, and see what happens. To understand the human brain well enough to enhance it appreciably, we’d need to do the same.
3. IanC Says:
March 21st, 2007 at 10:12 am
What I’m saying is that any enhancements that can be made to Turing-Test positive AI will likely be as readily made to human sentience.
Read/write access to every individual neuron *WON’T* be a necessity. Simple hormonal activation of growth responses would be sufficient to integrate hardware into the neural framework. Look at how adaptively responsive the brain has been to the Braingate & 16-dot optical implants as examples.
As to the Mendelev before prosthetics argument, that is obviated by advances in MRI imaging & synaptic language processing. That coupled with modeling makes it fairly simple, just an engineering problem.
Re: 2025 vs 2029 — either way the point is moot. DARPA wants Cyberkinetics to produce a working I/O brain prosthesis within 20 years. So either way, it’s more than a single-pony show.
As to being optimized to its hardware… I defy you to prove that it won’t be. Also, keep in mind the ‘adage’ that as you approach AGI, there will necessarily be certain levels of “sloppiness†required to be accepted.
The point is relatively simple; an integrated wetware/hardware system will likely be just as capable as a purely hardware system.
And unlike going ‘pure’ AI, at least with an integrative approach we have a wide referent base material.
4. Konrad Says:
March 21st, 2007 at 10:13 am
One listed advantage was: “The non-existence of “boredom†when performing repetitive tasks.â€
Boredom is not the problem. Boredom is the solution. The problem is a non-trivial (or poorly-defined, depending on your viewpoint) goal system which uses heuristics to choose between sub-goals.
Anybody who thinks that’s not a problem for AI should be asked to define the term “friendlyâ€.
5. Hawkeye Says:
March 21st, 2007 at 12:20 pm
Comment on predictions:
I tend to view a prediction as a “sell by†date. The person making the prediction thinks so many variables will coalesce by that date that it is “inevitable†that said event will occur. It doesn’t mean it won’t happen sooner. I’ve had a few interactions with people making predictions and it’s always with a caveat of, “Well, of course it could happen sooner.†I guess it doesn’t mean they don’t push it out as far as they deem likely, but it tends to be weighted outbound on the calendar if they can help it. That way their prediction is still valid for the longest period of time. There doesn’t seem to be any penalty for being off or late.
6. Nato Says:
March 21st, 2007 at 5:28 pm
Today, there’s more raw connectivity between large numbers of Internet connected PCs than between PCs and their owners.
Ubiquitous connectivity – especially of the wireless kind – suggests it will be hard to conceive of individuality or autonomy in AI.
Combine this with software industry trends toward web services rather than downloaded code, and I get a picture of //personified companies//, where a single corporate persona interacts online, or through various robotic telepresent shells, with large numbers of customers simultaneously.
With that much connectivity between AI “brainsâ€, it’s
7. Tom McCabe Says:
March 22nd, 2007 at 5:47 pm
“Simple hormonal activation of growth responses would be sufficient to integrate hardware into the neural framework.â€
This will simply not work for any reasonable kind of I/O input or basic reprogramming. How would you do something like give humans four-dimensional vision by injecting chemicals into the brain? These capabilities aren’t built in with a handy chemical switch; they need to be engineered just like any other complex thingy we can build.
“Look at how adaptively responsive the brain has been to the Braingate & 16-dot optical implants as examples.â€
This is NOT a new capability or even a real interface, since the data only goes one way: in. This is an enhancement to the human brain in the same way that plugging in a new printer is an enhancement to a computer; sure, you can do more on that specific system, but nothing that hasn’t already been done.
“I defy you to prove that it won’t be.â€
You do know that proving a negative is impossible, right?
“Anybody who thinks that’s not a problem for AI should be asked to define the term “friendlyâ€.â€
He’s only listing reasons why AI is more powerful, in the same way that uranium is more powerful than gasoline. It can generate more wealth but can also explode more powerfully.
“The person making the prediction thinks so many variables will coalesce by that date that it is “inevitable†that said event will occur.â€
That is not how Kurzweil makes his predictions. What he does is to extrapolate current trends, so his target date is somewhere around the middle of probability space. That is also not how DARPA makes predictions. What DARPA does is essentially:
1. Break the task into parts.
2. Squeeze the parts into tasks that have been done already, regardless of how well they fit.
3. Add up the times it took to do the tasks.
That is also not how most “futurists†make their predictions, which is to pull a number out of their ears with no supporting data at all. That is also not how Hollywood makes predictions, which is dreaming it up with no concern as to whether it’s even possible. That is also not how cultists make predictions, which is:
1. If they’re totally nuts/suicidal, in the next several years so as to encourage their followers.
2. If they have a lot of commercial interests or some degree of sense, then predictions will be as far off and vague as possible to avoid disproof.
Anyone who’s studied probability or has a good amount of experience knows better than to say “XYZ is inevitable by date Qâ€, because there are too many unknowns to account for.
8. Michael Anissimov Says:
March 22nd, 2007 at 6:52 pm
And unlike going ‘pure’ AI, at least with an integrative approach we have a wide referent base material.
Just like the first flying machines were integrations of birds and bicycles. The Wright Brothers had to wipe away some blood, muscle, and feathers from their shirts after the maiden flight was done, but at least they observed the power of integration.
I defy you to prove that it won’t be.
I’m the most awesome dude in the universe, I defy you to prove I’m not! ;)
Boredom is not the problem. Boredom is the solution.
In some narrow cases, maybe. Usually, no. If I want to do something, and boredom is preventing me from doing it, then I’d be silly to say that boredom is good. What’s the difference between saying “boredom is good†and “aging is good� You can contrive scenarios where both have positive qualities, but in general, never expect the default to be optimal!
http://www.acceleratingfuture.com/michael/works/beyonddefaults.htm
Ubiquitous connectivity – especially of the wireless kind – suggests it will be hard to conceive of individuality or autonomy in AI.
Different machines does not equal different minds. A single AI could exist across machines. And a general AI has to start somewhere, as a point source. Security measures will ensure that the first AI doesn’t run across copying itself willy-nilly unless something is seriously wrong, like the team completely lost control. The idea of multiple AIs emerging simultaneously comes from the sci-fi idea of “gremlins in the net†and the motivation behind it was to create a more interesting story. It’s a literary device, like Asimov Laws.
ombine this with software industry trends toward web services rather than downloaded code, and I get a picture of //personified companies//
These will be intelligent agents, not general AIs. Once general AI pops up, bam, Singularity. Transition to a new era. Not more of the same.
9. Konrad Says:
March 22nd, 2007 at 8:43 pm
“If I want to do something, and boredom is preventing me from doing it, then I’d be silly to say that boredom is good.â€
Look, this is not a question of semantics. How can you say that the boring task is what you really want, when your boredom is making you want to do something else? Do you see my point? Your brain chooses between subgoals all the time, and that directs your attention to using heuristics evolution thought up. They’re not the best approach, but they are NOT the problem. It’s simply one attempt at a solution.
Removing bored is trivial. Take some Ritalin and see for yourself. Take enough and you will realize you are approaching the behaviour of someone with OCD or a crack addict. Boredom is simply an attempt to redirect processing power. Remove it, and you see it wasn’t the problem to begin with.
10. IanC Says:
March 22nd, 2007 at 10:20 pm
Michael: The aerofoil shape is based on the pattern of movement of a bird’s wing. A far, far better example would be the use of biomimetics in robotics. Every time we attempt to model the motion of our machines on an already existing organism, we find the results to be far more effective than when we attempt to “re-invent the wheel.†Why should AI be any different?
Re: Your statement is that Seed AI will be able to improve itself solely through code. This implies that it will inevitably optimize itself to its then-hardware. I defy you to prove that is not the case, or even give a plausible reason why it isn’t. And for the record, gentlemen, science is all about falsifiability.
Of course, I shouldn’t be surprised at the irrationality of your statements regarding AI. You’ve yet to acknowledge my comments regarding your susceptibility to anthropocentrism-in-reverse.
Tom: There is no functioning ‘data-in’ BMI at this time. The Braingate chip is a solely data-out. The 16-dot & 60-dot optical implants are retinal prosthetics. So in this you are absolutely, 100% incorrect. Not only *IS* there no viable data-in BMI, there is no research being conducted or hypothesized *FOR* data-in BMI. (That is a falsifiable negative, by the way, gentlemen.)
Further, Tom: Regarding the use of hormone factors to accelerate hardware integration… I am only left with the conclusion that you are being intentionally dense. Seriously, the segment you quote in and of itself falsifies your response. Are you even reading it? My response to your criticisms will be that you ought to actually do so. I apologize for the harshness of my tone here. To re-emphasize; hardware integrated into the brain will be adapted into its functions. This is more than merely adding a calculator and a database to human experience. This can actually fundamentally alter what it means to experience the world. The example of neocortical augmentation is a perfect demonstration of this.
Konrad is essentially absolutely correct regarding boredom. Any general AI *WILL* develop something approximating the sloppiness of a human mind. While it will inevitably be capable of multitasking far more than are we, the sheer complexity of the information load of the actual world means any AI meant to interact with the world in a way we do will be, assuming Turing-Test positive *ONLY*, hardly likely to be any more capable of multitasking than are we; at least not in any appreciable manner. No, Michael, that’s not anthropocentrism. That’s estimation based on information complexity and what I know of AI progress.
11. Michael Anissimov Says:
March 23rd, 2007 at 12:24 am
Konrad, the lack of boredom in the software underlying this blog is what makes it possible. It is also what makes the NEC Earth Simulator possible. If a certain useful piece of information requires a long math calculation on the part of the AI, it will do it. Not get bored and quit halfway.
Ian, you are saying that any advantage an AI has by virtue of its hardware, would be trivially implemented in a human intelligence augmentation project. This is wrong.
12. IanC Says:
March 23rd, 2007 at 8:34 am
Michael, you’re quite right that it is wrong. There’s nothing trivial about it. Nowhere have I mentioned or implied a “human intelligence augmentation project.†I have simply stated that any enhancement available to an AI will be likely as available to a human being.
If you disagree with this statement, given that it is based on already existent trends which I have already referenced, could you please explain why in an empirical sense? Are you saying that DARPA will change its mind for some mysterious reason? Or that the inestimable opportunities possible to true data-in BMI will *NEVER* be developed? How does that make you any different from “The AI establishment†you have criticized elsewhere? Or is there simply a total disconnect in your thinking where it comes to the implications of AGI co-existing with data-in BMI?
I know for a *FACT* that I and at least three others would be willing to pay to be involved in human trials for early data-in BMI, despite any potential risks. And I’m hardly a sociable person. There is in the USA government interest in its development; there is also private interest in its development.
So I have to ask again; why the bias against its manifestation, and why the equal bias against even contemplating the implications of it?
(By way of allegorical allusion to my thoughts here, you’re behaving quite like a cyb from L.E. Modessitt Jr.’s “Adiamanteâ€.)
March 28th, 2007 - 10:36
Michael wrote, “Konrad, the lack of boredom in the software underlying this blog is what makes it possible.”
Your blog software isn’t intelligent either. It’s just a dumb algorithm that doesn’t face the kind of problems which can be solved through boredom. It doesn’t make sense to talk about boredom for anything that has a trivial goal system.
It’s a different story when a problem requires you to divide finite resources between several sub-goals. Show me a heuristic approach to resource allocation, and I’ll show you boredom.
Consider the software which provides the interface between your blog and the net. Normally, it just responds to request on a first-come, first-served basis. But if it’s smart enough to recognize and resist a DDoS attack, it does so by learning to ignore certain requests. In other words, it *can* get bored of certain repetitive tasks, IFF resource allocation comes into play.
March 31st, 2007 - 02:04
Quite a nice summary, I’ll save it for future reference. Would be nice if I could verbatim reproduce such lists on a party; ideal device sure to kill of the rest of the chitchat.
April 5th, 2007 - 09:47
Michael; regarding the I/O BMI & it’s potential for human cognizance augmentation:
http://www.popsci.com/popsci/science/0e54d952c97b1110vgnvcm1000004eecbccdrcrd.html
Dr. Berger indicates that he’s “15 years” from human trials with this concept. It’s only a biomimetic, but it’s a direct I/O BMI with the hippocampus. Similar techniques could *EXTREMELY* easily be adapted for the neocortex. The implications of this are obvious and direct.
October 30th, 2011 - 13:34
I precisely desired to appreciate you once again. I’m not certain what I could possibly have undertaken in the absence of the entire secrets discussed by you regarding my field. Completely was a real alarming case in my circumstances, however , noticing a expert mode you processed that took me to cry for fulfillment. I’m just grateful for your support and in addition trust you realize what an amazing job you have been getting into teaching many others through your blog post. Probably you haven’t met any of us.
March 21st, 2012 - 03:00
I would like to thnkx for the efforts you’ve put in writing this web site. I am hoping the same high-grade website post from you in the upcoming as well. Actually your creative writing abilities has inspired me to get my own website now. Actually the blogging is spreading its wings quickly. Your write up is a great example of it.
April 2nd, 2012 - 15:15
What i do not understood is in reality how you are no longer really a lot more smartly-favored than you may be now. You’re so intelligent. You already know thus significantly relating to this topic, produced me personally believe it from so many numerous angles. Its like women and men are not involved until it’s one thing to do with Lady gaga! Your own stuffs excellent. All the time care for it up!
April 14th, 2012 - 10:02
Thanks for discussing your ideas right here. The other thing is that every time a problem arises with a personal computer motherboard, individuals should not have some risk associated with repairing it themselves for if it is not done properly it can lead to permanent damage to all the laptop. It’s usually safe just to approach a dealer of any laptop for your repair of the motherboard. They will have technicians who’ve an know-how in dealing with notebook motherboard troubles and can have the right analysis and undertake repairs.
April 19th, 2012 - 03:04
Hmm is anyone else having problems with the pictures on this blog loading? I’m trying to figure out if its a problem on my end or if it’s the blog. Any feedback would be greatly appreciated.