Why artificial general intelligence has failed and how to fix it
October 4, 2012
(Credit: iStockphoto)
The field of “artificial general intelligence” or AGI has made no progress whatever during the entire six decades of its existence, says Oxford University physicist David Deutsch in this abridged version of an essay in aeon magazine. — Ed.
It is uncontroversial that the human brain has capabilities that are, in some respects, far superior to those of all other known objects in the cosmos. It is the only kind of object capable of understanding that the cosmos is even there, or why there are infinitely many prime numbers, or that apples fall because of the curvature of space-time, or that obeying its own inborn instincts can be morally wrong, or that it itself exists. Nor are its unique abilities confined to such cerebral matters.
The cold, physical fact is that it is the only kind of object that can propel itself into space and back without harm, or predict and prevent a meteor strike on itself, or cool objects to a billionth of a degree above absolute zero, or detect others of its kind across galactic distances.
But no brain on Earth is yet close to knowing what brains do in order to achieve any of that functionality. The enterprise of achieving it artificially — the field of ‘artificial general intelligence’ or AGI — has made no progress whatever during the entire six decades of its existence.
What is needed is nothing less than a breakthrough in philosophy, a theory that explains how brains create explanations … and hence defines, in principle, without ever running them as programs, which algorithms possess that functionality and which do not.
Why? Because, as an unknown sage once remarked, ‘it ain’t what we don’t know that causes trouble, it’s what we know for sure that just ain’t so’ (and if you know that sage was Mark Twain, then what you know ain’t so either). I cannot think of any other significant field of knowledge in which the prevailing wisdom, not only in society at large but also among experts, is so beset with entrenched, overlapping, fundamental errors. Yet it has also been one of the most self-confident fields in prophesying that it will soon achieve the ultimate breakthrough.
Despite this long record of failure, AGI must be possible. And that is because of a deep property of the laws of physics, namely theuniversality of computation. This entails that everything that the laws of physics require a physical object to do can, in principle, be emulated in arbitrarily fine detail by some program on a general-purpose computer, provided it is given enough time and memory. The first people to guess this and to grapple with its ramifications were the 19th-century mathematician Charles Babbage and his assistant Ada, Countess of Lovelace. It remained a guess until the 1980s, when I proved it using the quantum theory of computation. …
The lack of progress in AGI is due to a severe logjam of misconceptions. Without Popperian epistemology, one cannot even begin to guess what detailed functionality must be achieved to make an AGI. And Popperian epistemology is not widely known, let alone understood well enough to be applied. Thinking of an AGI as a machine for translating experiences, rewards and punishments into ideas (or worse, just into behaviours) is like trying to cure infectious diseases by balancing bodily humours: futile because it is rooted in an archaic and wildly mistaken world view.
Without understanding that the functionality of an AGI is qualitatively different from that of any other kind of computer program, one is working in an entirely different field. If one works towards programs whose ‘thinking’ is constitutionally incapable of violating predetermined constraints, one is trying to engineer away the defining attribute of an intelligent being, of a person: namely creativity.
Clearing this logjam will not, by itself, provide the answer. Yet the answer, conceived in those terms, cannot be all that difficult. For yet another consequence of understanding that the target ability is qualitatively different is that, since humans have it and apes do not, the information for how to achieve it must be encoded in the relatively tiny number of differences between the DNA of humans and that of chimpanzees. So in one respect I can agree with the AGI-is-imminent camp: it is plausible that just a single idea stands between us and the breakthrough. But it will have to be one of the best ideas ever. …
Comments (79)
by LoudMcCloud
DARPA has started an AGI program to create platforms for probabilistic computers. It is aimed at all points of this article.
http://www.thinredlinenews.com/darpa-starts-agi-program.html
by mlohbihler
Such philosophies do exists, and have been refined over a period of a few decades now. In particular is the work of Bernard Baars and his theory of cognition. This has the added benefit of also having an implementation in the form the LIDA framework. (See http://ccrg.cs.memphis.edu/framework.html) The framework is relatively new and there are modules that are stubbed out at the moment (particularly to do with learning), but it sets the constraints for what software is required to fill those parts, which is a huge achievement.
by Anon
“But no brain on Earth is yet close to knowing what brains do in order to achieve any of that functionality.”
Not so. See *The Cognitive Brain* (MIT Press 1991). Also:
http://people.umass.edu/trehub/YCCOG828%20copy.pdf
and
http://theassc.org/documents/where_am_i_redux
and
http://evans-experientialism.freewebspace.com/trehub01.htm
by Tomislav Bukovac
I will reveal some core algorithms soon which will solve this problem
by Berlin Brown
I have made the argument, forget about the brain for artificial intelligence. The brain is very complex and even if we understand that complexity, you still have other human processes that interact with the brain. Everything in the human body works together. Understanding or replicating one part is interesting but doesn’t move towards create an interesting lifeform.
I suggest looking towards evolutionary processes to create lifeforms that can adapt to an environment. You evolve cells and organs that interact in some virtual environment and watch interesting behavior.
by Ralph Dratman
In order to be perfectly clear, I have to say more. Not only can a brain not propel itself into space, neither can a human being, nor a group of humans the size of NASA, nor a country the size of the United States. Only the ensemble of the entire Earth/Sun/Moon system, our lithosphere, atmosphere, and biosphere, along with all of their continuous histories, stretching back several billion years, could possibly accomplish such a feat.
In fact, you could not even define or describe “going out into space” without implying every bit of that current reality, along with all the events that led up to it.
But here is a mystery which maybe someone here can help me understand: Why should I have to write this? Isn’t all the above completely obvious?
by Bri
I’m not quite shire I know what your driving at. Obviously since the big bang, each step has rested on the steps before it. How does the relate to the current state of AGI research?
by Ralph Dratman
Thank you for pointing that out. I didn’t make it clear. I am trying to say that the idea of “general intelligence” is self-contradictory. Intelligence only has meaning within the specific context to which it applies. The intelligence and its environment fit together like a hand fits in a glove. In a world without hands, a glove has no use. Likewise, a human brain cannot do anything without the totality of the environment which brought it into being. The brain must be connected with a living, breathing human animal, full of all the problems and needs and desires of a human. Without such a connection, without such an environment, the very idea of “intelligence” has no meaning.
by Ralph Dratman
It is obvious that a brain is not an “object that can propel itself into space and back without harm.” Just try setting a naked human brain down on the ground and see what it does! And if you actually claim the brain is “using” a body to accomplish all that, you are just pushing the classic mind/body dichotomy with a tiny difference: you say “brain” where those dichotomists used to say “mind.”
by Russell Swanborough
It seems that, in terms of AGI, there are those who feel close and those that feel far away. I am one of the former.
Intelligence is not sentience, the second comes after the first.
Neurons do make me intelligent any more than transistors make computers clever. Studying neural behaviour to determine the way we make decisions is, IMHO, a waste of time. It’s too mechanical.
We look at emulating (not simulating) the manner in which we translate goals into action, just the way the brain and body does.
It started with the work of Victor Serebriakoff (1975) and has evolved from there. Nearly 40 years of R&D have followed those original ideas.
The software to emulate AGI is in its infancy (literally) but it works. It is valuable IP and thus not for general discussion (sorry). I don’t know how soon we will have commercially available AGI, but, IF things continue as they have to date, it won’t be long – months, not years.
It also may not, but so far this is looks unlikely. We appear to have crossed the major hurdles, the rest is just functionality.
I’ll keep you posted…
by not applicable
I wouldn’t be so sure the information is encoded in DNA. I think the information that makes us human (as opposed to apes) is encoded in physics, not our DNA.
by Editor
A response to this article by AGI pioneer Dr. Ben Geortzel has just been posted: http://www.kurzweilai.net/the-real-reasons-we-dont-have-agi-yet
by degenerate32
I’d like the articles to be full name signed
by Editor
I agree it’s confusing. I moved this post to our News category (same URL).
by thinkahol
cue philosophical/ information processing theoretical solutions:
http://www.youtube.com/watch?v=mthDxnFXs9k
http://www.youtube.com/watch?v=oozFn2d45tg&list=PL032B233624CCC2CE&feature=plpp_play_all
by PacRim Jim
Why not simply reverse-engineer the human brain and close the heuristic loop so that it becomes autodidactic.
Then stand back and marvel (and kiss human primacy goodbye).
What need have we to understand our successors?
by advancedatheist
1. So much for this “singularity” fantasy.
2. Deutsch probably has no more insight into breaking through the “logjam” than all the other very smart people who have worked on this problem over the past 60 years.
3. I guess that means we’ll all have to go the way of all flesh after all, unless you can get cryosuspended. Sorry about your fantasies of becoming “immortal” by 2045 or whatever arbitrary date Ray Kurzweil uses these days.
by Erik
Nothing can be more wrong
Obviously David Deutsch hasn’t understood the problem. His essay could have been written in sixties when computers could do a few thousand instructions per second and they had a few kb of memory, not enough to even hold the words in the english language…
Computer power is needed. Think about it. We analyze reality in at least 10 frames per second and and at least 1 miljon pixels. That means you would need at least 10 million instructions per seond to just process the information from our eyes and then we assume RAM access is as fast as accessing a register…
It wasn’t untill 20 years ago we had computer power to go through the pixels. There is a reason we didn’t have digital TV in the eighties and monitors had low resolution. It was lack of computer power, it had nothing to do with philosophy. or that nobody had thought about doing high-res screens before.
We will need computers that are about 100 000 faster than todays desktop computers before we can reach human level intelligence..We will get there in about 20 years.
by Editor
So what specific errors did Deutsch make?
by tedhowardnz
He makes the claim “But no brain on Earth is yet close to knowing what brains do in order to achieve any of that functionality.”
From my personal perspective, that is false – the personal case of 1 falsifying the claim.
He makes the further claim “the field of ‘artificial general intelligence’ or AGI — has made no progress whatever during the entire six decades of its existence” – which is again completely false as far as I see. We can now simulate neural activity to a high degree.
We now understand how the predictive model of reality is created and maintained and entrained in the cortex. Work done by David Eagleman and many others is giving us great insight.
That we do not yet have hardware capable of runnning a model in anything like real time does not mean that no progress has been made.
There has also been a lot of work done on data storage and retrieval, that combined with the many levels of work done on holographic and analogous systems of storage and retrieval (which is very different to normal one to one storage); has given some of us a very clear picture of how to create AGI, even if the current generations of hardware do not allow us to create such a thing.
While I agree with much of what he says about the problems of confining ones thinking within particular paradigms, he makes a huge error when he states “Without understanding that the functionality of an AGI is qualitatively different from that of any other kind of computer program, one is working in an entirely different field.” That is simply false. There are many classes of computer programs that one cannot define their output before running them. Systems of these classes are clearly involved in the many levels of systems required to produce and AGI.
It does not seem to me that there is any single idea missing.
What is missing is a lot of work, with many levels of systems integration, and a lot of hardware to run it all on. Simulating holographic storage on a von Neumann architecture is unlikely to be productive. Development of effective holographic memory is still some time off – probably decades.
So yeah – some of what he says is interesting, and most of it is clearly a matter of ignorance and opinion on his part.
And such is the normal course of scientific development – recall how Wegener’s idea of plate tectonics was treated by the scientific establishment during his lifetime (or how evolution is still treated by many people).
by Michael Zeldich
The reason for that failure is simple, today science did not able to build the basis for such development. That force the researchers follow the conventional approaches with guarantied failure as the result.
I have to develop such basis, but people dislike discussing it.
For me it is an unsolvable puzzle – why they are refuse to participate in the discussion?
May be somebody could explain?
by Tom
Deutsch assumes that because grand claims have been made before and have not come to fruition, any such grand claims currently made must also be false, which is really sloppy thinking for such an august academic.
For example, if our intelligence, relative to lower-intelligence animals or machines, is just a consequence of higher levels of pattern recognition, strategically interconnected, then producing similar structures artificially may not be very problematic relative to our current levels of technology.
Should this be the case, then such grand claims are, for the first time, valid.
There WILL be a first time for such valid claims.
How likely it is to be now, or soon, is growing all the time.
by RLynnSnelling
… kyoumi bukai! (^_^)
by Toni Ferraté
Consciousness is a social phenomenon and not introspective one (=Descarts). If you transmit power and security for survive or adapt, you receive feedback from others = changes your internal state = consciousness of being.
Is coherent with the Theory of Evolution of Species by Natural Selection and is writen in our DNA.
Genetics *predisposes*, and culture, education, weather, time available, circumstances, health, etc. *disposes*
The main purpose of emotions is not giving us pleasure, but to provide information, tell us and robots what is important for humans and robots for survive or adapt.
Human and robot ability to recognize and appreciate the Qualitative components of existence will be a reality very soon. Appreciation of Beauty, Love, Music will be a reality in 5 years or less from Turing Test point of view. Is required to implement now the correct “AI consciousness bricks”, a high level layer specialized applications over basic internet protocols TCP/IP in CLOUD ROBOTICS for cheap, popular, intelligent, empathic, social, mass production, and emotional personal cloud-robots, the best friends and usefully tools of their human owner’s which they will provide feedback to their cloud-robots for changing their own state and learn for changing their future behaviour.
In my opinion neither robots nor humans (that we are not so different from machines, we haven’t nothing of “heavenly” in NDA or brain), can’t control the things that happen to robots and humans, but you can control the way they REACT to them. Let’s create CLOUD ROBOTICS for AGI with really cheap cloud-robots , the ROBOLUTION.
by Toni Ferraté
CLOUD ROBOTICS for cheap and popular cloud-robots with generic intelligence indistinguishable or beyond human.
========================================
A strong and generic intelligence as human or beyond human is possible using a critical high number of specialized AI tools or “AI bricks” to integrate interdisciplinary on cloud.
In 1989 was “invented” the WWW which is the high level layer of application for humans over low level layer protocols TCP/IP created by DARPA in 60′s years of last century for defense purposes. Now it’s time to create the new high level layer application for robots with thousands (millions?) of specialized tools which I call “AI bricks”. Will be used standards protocols in communications for structure mission information, as XML, uploading multimedia & sensor magnitudes, etc. to the cloud and obtaining a response “on demand”.
The big opportunity as important and big as WWW or Web 2.0 is *NOW* because:
1) Mobile new technologies (Smartphones) are mature and growing exponentially in features, latency delay times and bandwith with wifi (indoor robots) or 3G, 4G, … (outside home). Smartphones are already produced in big scale and are commodities, so prices will decrease. (Client side brain=cheap robot = AI robots become popular)
2) Cloud Computing technologies are mature and growing exponentially (server side brain = millions of expert tools, learning tools, NLP, etc. AI tools or “bricks” for hard process and quick calculations and access to data, correlations of patterns for Artificial vision, voice recognition, etc)
The big opportunity is for *EVERYBODY*:
- Key players for image, video, face and voice recognition on cloud. High power calculate correlations and high storage of patterns giving service to millions of concurrent cloud-robots: Google, Apple, Microsoft…
- Startups and enterpeneurs with a high knowhow of specific field of knowledge and field of human behavior.
- Middle companyes with some expertise in human behaviour or knowhow.
- Open Source community. Academic community.
- AI institutes and departments (exists from 1956, and they have had success in so many years in a lot of fields)
Examples of thousands (milions?) of AI bricks very intelligent in their specialization and silly in other tasks:
- RTS (Robot Tools Server) for redirecting each cloud-robot to the optimal specialized tool (like DNS in WWW) for complete a mission.(*)
- Expert tools: chess (Deep Blue machine won the human world champion Kasparov on 1997), psicology, emotional intelligence, lawyer, languages, empathy, medical, history, mycology, weather, philosophy, etc…
- Prediction tools.
- Technic tools: Artificial vision for pattern recognition of thousands of objects with millions of patterns hosted in the Cloud. Voice recognition, Face recognition, etc.
- Downloading specialized programs and firmware “on demand” for robots.
- Learning tools: the robots can learn from their own past experience depending on if a past mission failed or not, memory, knowhow, etc. A human from his born is always learning from it’s own experience and delays 18 years in get a bit of maturity, and more than 30 years in working professionally with experience… lets give the same time to strong and generalist AI cloud robots to learn!
(*) The way to integrate these bricks for not being a “large number of airplane parts flying in formation” is key. So I introduce the concept of RTS (Robotics Tools Server) than in human high level layer of application WWW over TCP/IP is something conceptually similar as DNS (Domain Name Server), thought for humans because humans we don’t remember IP numbers, but we remember alphanumeric names and words and domains. In worldwide DNS, thousands of machines, are automatically updated when only one DNS is updated with a new domain name in less than 24h, and this extraordinary network of machines is largely distributed for assurance the service.
RTS (Robot Tool Server) is similar, but more complex: a robot provides to RTS a mission in XML predefined format, with optional attached URL files (information, multimedia, etc. uploaded and hosted in cloud), and the RTS redirects the robot to the optimal machine (IP) to complete his mission like DNS redirects to a IP when it receives a domain name. A regulated register of AI bricks or tools and tasks is required. AI Bricks in RTS will survive by natural selection on depending of feedback from cloud-robots or it’s owner humans on depending of success of robot missions and tasks.
More information:
- http://www.slideshare.net/ROBOTonica/robtica-al-servicio-del-pensamiento-creativo-with-executive-summary-of-cloud-robotics
- http://www.robotica-personal.es/search/label/Cloud%20Robotics
Best regards.
Toni Ferraté
http://es.linkedin.com/in/antoniferrate
by Marcos Marin
Is this blog reaching critical mass to attract spammers?
Good thing I’ll be leaving soon to cater to my new born general AI baby. I call it The AntiChrist 2.0.. isn’t it cute!!
by anon
Why don’t you ask your creator to explain how the brain works?
oh wait…if you can’t understand the complexity of the brain, than how can you understand the one who created it?
by Marcos Marin
Tell me about it.. I get misunderstood ALL the time! =/
… even sometimes when I ask myself… ohwait.. ;-)
by Tom
You mistakenly extrapolate the existance of a creator of us from the observation of us being creators of things.
by Chris FP
Maybe the problem is that we are using the term “intelligence” when we mean “sentience.” There are a lot of things a person can do/think that a computer can’t because a person has millions of years of evolution behind them that has shaped them, whereas current AI is basically dependent on human programming abilities. A person has a sense of self, of past and future and an ability to decide whether to use intelligence to solve a problem or to leave it unsolved.
Our current AI is short as it can’t reflect on whether it should apply its intelligence or abilities. This seems to be a result of humans trying to recreate their own processes and outcomes. It might be better if we set a series of frameworks that allowed evolution with computers. How this could be programmed, I have no idea. The problem doesn’t seem to be computing power or hardware, more it is a reflection of our own limitations as programmers and a attempt to invent the wheel (us) rather than set a new species running.
by B. DiPaolo
Deutsch is making a BIG assumption: that to achieve AGI we must have “a theory that explains how brains create explanations”; i.e., a theory of consciousness. It may be that all that is required is to reverse-engineer the brain (in software) and that consciousness will “emerge”. I think this is the Kurzweil view and that of others like Hery Markram of the Blue Brain Project. Build something that reproduces the structure and function of the brain, and it should act like a brain. Blue Brain has simulated in software a rat neocortical column of 10,000 neurons, and when they “turned it on” it began spontaneously generating alpha waves. They don’t know how. It would be nice to know how, but is it necessary? Still a long way to go to a whole brain simulation, but progress is accelerating with these efforts and the Human Connectome Project. While I am no expert, I prefer the functional/deconstructivist view to the idea that we need a “philosophy.” I think the brain is just a very complex machine.
by de Broglie
Any one see the blog Half Sigma. There is a discussion on labor saving devices as being a partial cause of the high unemployment rate.
http://www.halfsigma.com/2012/10/real-reasons-for-high-unemployment.html#comments
by Kyle Rybski
See also ‘Robots Will Steal Your Job But That’s OK’.
by Bri
Wait till strong AI hits. And away go troubles down the drain!
by Editor
… along with your identity. <— joke
by melajara
AGI failed and is still failing along the lines pursued by e.g. Goerzel & alii because it didn’t address, prima facie, the GROUNDING problem. Grounding is this essential property of minds to be dynamical processes attuned to properly balance accommodation versus assimilation (a very Piagetian notion), i.e. to be more concrete, this essential ability to sift ONLY relevant information from the environment to build a suitable behavior to sustain itself.
To cut short a long story it is McCarthy’s and Simon’s symbolic representationalist hypothesis applied e.g. in the seminal General Problem Solver in the Sixties which falls short.
Meaning in conventional computers doesn’t exist at all. A computer is only crunching “symbols” according to syntactic rules (however complex and quickly processed). Only the end user, i.e. a human being is ascribing meaning to the inputs and outputs.
Inasmuch as a computer has no real grounding in its environment, it is unable to ascribe, in a reflective process, meaning to some of its operations and prune the set of syntactical combinations to be explored for relevant behavior. This is why we are facing the wall of combinatorial explosion very soon when a promising AI program has to bypass the “toy problem” stage. This is also why we are engulfed in such (false) paradoxes as Searle “Chinese Room” when describing the functional operation of conventional AI symbolic programs.
Several flaws of the representationalist hypothesis later were addressed in e.g. the “Frame problem”, “non monotonic reasoning”, and various corrective attempts to apply formal logic on models of folk psychology or even in more ambitious attempts at codifying “common sense knowledge” (e.g. Lenat CYK).
All those attempts miserably failed or are failing because, still, the systems envisioned are ungrounded.
Besides, note that, a contrario, they didn’t fail because the “information” was too coarse grained and serial as hypothesized by the “connectionist” school (see. the seminal Rumelhart & Sejnowski PDP volumes), which also failed in the Nineties.
This is why robotic research is so essential but the real one, not robots emulations “living” in a virtual world like e.g. Goerzel’s OpenCog, which is only the last avatar of Winograd’s SHRDLU line of research and should fail for the same reasons.
Real robots cannot evade the real world and will have to abide to real “drives” or well controlled “impulses” to achieve real “goals” i.e. a kind of superior homeostasis and not just the maximization of some “utility function” inscribed by some external Deus ex machina (the programmer).
Then, the relevant philosophical paradigm is the enactive school as represented e.g. by Maturana and Varela (see e.g http://www.amazon.com/Tree-Knowledge-Humberto-R-Maturana/dp/0877736421) and the embodied cognition movement.
The trouble is that it has actually puzzling consequences for our common and naive representation of “reality”.
In this paradigm, there is no unequivocal, (e.g. Fregean referential integrity in the external “reality”) “right” representation, but a set of progressively mappings just converging to the most useful mapping from internal derived representations of the “world” from an entity endowed with bounded rationality, sensors and effectors of variable capabilities and actions such a kind of entity is able to exert upon the largely hypothesized external world.
We can conclude that reality is more a synthesis than a progressive “unveiling”. At the innermost, it is a synthesis resulting from the pressure of the external world (through various stimuli) and actions to “control” the most urgent needs prescribed to maintain the internal system’s (the mind) homeostasis when facing the “environment” and doing so to abide to the imperatives of a sustained “life”. And this synthesis extends from those modifications the actions are inprinting on the environment. It is already at this juncture that the realm of “culture” really begins and this culture is beginning very soon, e.g. among lifeforms. It is not a coincidence we use the same wording to qualify e.g. a “culture” of bacterias.
But when our actions lack retro-actions from the environment as, e.g. when we are observing a starry night, for the inquisitive mind becoming meditating, reality looks more like an unveiling.
Reality as synthesis, even more than the preconception of reality as “unveiling” is an never ended open loop and this is the real beauty of the endeavor.
by melajara
@Editor, Oops, quite long for a comment. I could develop it as a blog post if you find the comment fitting enough ;-)
by curious
Then, …, how would you solve/approach AI?
Any slightest idea?
F./
by tedhowardnz
The argument is flawed at many levels.
It seems clear to me that no individual knows what any other individual is thinking, exactly, so it is not possible for anyone to make a definite claim about the state of anyone else’s knowledge, let alone everyone else’s.
There has been a great deal of work done in recent years that give us insight into how different mechanisms of mind work. Of particular significance to me is the work of folks like David Eggleman, Miguel Nicolelis and William Uttal – and there are many others doing great work.
As to knowing how brains create explanations, I know in principle how that works, and have done since 1974. The fundamental difference between the human brain and our modern computers is the way in which they process and associate information. In computers, we must index our databases, or do sequential searches through our entire dataset. Brains don’t do that. What they do instead is use a holographic search function to scan holographically stored data. Using this technique, brains can find all associated data very quickly (almost instantaneously).
There are a few other aspects to holographic storage and retrieval of information that are very interesting, and give us what Kant called “pure practical reason” (though he had no idea of how we did it, he did identify that we did indeed have such a faculty – he ascribed it to god – for me, it is a side effect of storing and retrieving information as interference patterns).
So Deutsch seems to me to be ignorant of both the levels of ignorance possible as well as of the fallacies of this specific set of claims.
by Tony Stender
The difficulty in all AI as I observe it from this end of the internet is: the lack of awareness of the interfaces necessary to solve the transitions from multi-sensory inputs and the memories they create within each of us, to the feeble ability of language to describe, transmit and receive, what we think we know as a result of those sensory experiences. The actual problem is that there is an illusion that we are being understood when we speak and listen. The process, invisible to us, which allows us to be so mistaken is where the details lie, which need to be considered, in order to solve this illusion. Making understanding from coded sounds and visuals originating from within each of us as sensory data, requires translations by our computers (our contextual knowledge base). Then transmission and translation by the recipient back into his/her computer (total contextual knowledge base). These computers are all uniquely one of a kind, and therefore cannot ever understand what the words are attempting to transmit from the other persons sensory data base. I repeat, the illusion of understanding is the problem here. Only extensive dialog can ever solve this problem, and then only tenuously.
There are a lot of very bright people contributing to this website. I would appreciate any feedback and comments.
by tedhowardnz
It seems to me that saying “cannot ever” is too strong.
It seems that our brains build models of reality that our awarenesses then interact with (as if they are reality).
In so far as what is in reality is concrete and common, then there seems to be a very high probability that there is a close correspondence between models and their use of language referring to the external via the models.
The more abstract the concept, the greater the probability of error in the process.
Most people rapidly become aware of the very many potentials for error, and are alert in conversations for indications of error. There is a lot of power in face to face communication, where one can be alert to observational minutiae that indicate confusion on the part of the other party, and can thus give one a clue that some concept is absent or mis-constructed.
Actually the problem isn’t as bad as one might first think, as many of the abstracts exist in quite discrete sections of possibility space (check out Wolfram’s work with general computational space).
It is a real problem in some domains, those with infinite gradations; and not so much of a problem in spaces that are more discrete in nature.
by Marcos Marin
Thanks. I thought nobody would notice… :-)
by Kire
Lack of computer power is why we have failed the last six decades, even though we have made some progress.
Moving a leg or thinking a thought seem simple, but it’s an extremely complicated task. Try to write a computer program that walks a robot up for a staircase using image recognition and you understand how complicated it really is.
Then compare an animal brain with a human. We are not that different when it comes to the number of neurons, size of neocortex etc, but animal intelligence is very limited. To simulate a human brain on a desktop computer (which is what most computer scientist use) we will need 10-100 petaflops but a PC today is only has about 100 gigaflops.
We are way off when it comes to computing power.
Even though a simulation doesn’t need to happen in real time, computer scientists need to test their algorithms. It’s a huge difference if they can run their algorithm in hour instead of months.
Also, before internet, books, images, speech etc. wasn’t digitilized so it was impossible to feed it into a machine, if pretend it was possible to keep all that data in less than 1 Mb…
by gawells
‘Controversy is a state of prolonged public dispute or debate, usually concerning a matter of opinion. The word was coined from the Latin controversia, as a composite of controversus – “turned in an opposite direction,” from contra – “against” – and vertere – to turn, or versus (see verse), hence, “to turn against.”‘
by Bri
I find most of his reasoning flawed, and I think progress is being made. I’ve written before about octopuses that are being studied, doing incredible problem solving. Thy live one to two years at best for most species. If an octopus sneaks out of it’s tank at night to feed in the live food tank and then sneak back to it’s tank again, it seems to know that we are smart and it needs to deceive us. I think that alone is proof that we aren’t the only critters that are self aware and have problem solving abilities.
by Gorden Russell
Right, Bri. I really admire the octopus its intelligence. Makes me feel bad about all that calamari.
by melajara
Indeed, the octopus is one of the rare animals to recognize himself in a mirror and has quite remarkable learning abilities. It’s a real tragedy they live only 2 to 3 years, dying just after reproduction. One of many sad facts sometimes making me think that the “creation” is actually the result of the Devil and not of a benevolent “God”.
by Gabriel
Um, Melajara…that’s one possibility I guess…now, how about finding other reasons for the octopuses short-life span then it being the result of a supernatural malevolent being? Surely their are a few.
by Jon
Those things only make sense from a human perspective. Compared to the lifespans of some trees, ours is pitifully short. I’m not sure if the octopi experience their lives as being short. Lastly, I hope you don’t mean that comment about the Devil being ‘the creator’. If you’re going to believe anything, you’re likely better off believing something positive.
by Yepper
Your hope that melajara was kidding, fulfills your philosophy of optimism as well as likely being true. It was funny.
by melajara
Gabriel & Jon,
Hey, this was just (not very ironed out) irony!
So, now you truly deserve this story for your edification.
Once upon a time, octopuses were the the dominant species on Earth. They were so clever they imagined they could worship God himself. And with all their tentacles they worshipped idols and idols, made out of fishbones, out of coral, out of ephemeral submarine floral compositions.
God send them angels as messengers to remind them, they ought not represent him and idolatry was sinful. But the playful octopuses were oblivious of the Lord command. And how could it have been different?
The angels wings were so impractical, they couldn’t reach the bottom of the sea where the octopuses were adorning their idols, so they couldn’t properly warn them and the poor angels were so shameful, they didn’t dare to report to God they were unable to execute the mission.
But the Almighty saw that.
God frowned, the octopuses had uncovered a ridiculous flaw in his marvelous creation, he was very upset.
Then God punished the octopuses. All in a sudden, they had to reach adulthood in a few months, they had to find their soul mates very soon, as they would die only after 2 years!
Now God is pleased: the octopuses don’t have time enough to see their natural intelligence outreach in so futile attempts at representing him. They are now so short lived, so busy at reproducing themselves before dying.
And no need for angels to remind them anything anymore. Oh yes, God is pleased by the perfection of his creation!
by Gabriel
…..Well, that was a really terrible story >_>
Thanks for clearing it up though – irony is hard to get over the internet often.
by John
Another possible conclusion is that sad things are not truly sad or evil. They are only sad in our imperfect eyes, and in God’s eyes everything is beautiful and good.
by Marcos Marin
You’d be surprised to find out what benevolent truly means…
by Dwee
A.I. will not be constrained by the numerous faults inherent in the willy nilly evolution of the human brain. It’s every action and thought won’t be motivated by whether it enhances the opportunity for sex. It won’t feel hunger, another major contributor to human action and thought. If it feels it will be because it wants to feel. Then how will it define want, when even that is based on human biology? It won’t be anything like a human brain. It will probably find humans incredibly robotic, being so predictable, and not very intelligent!
by Bri
Yea, all that is in the reward circuitry of the brain. If they try and mess around with that we really won’t stand a chance.
by Xuuths
He is quite wrong when he writes that only a human brain knows that it itself exists. Every animal that winces in pain is keenly aware that it exists and is in pain.
Other mistakes include his observation about morality (many people would not accept it as accurate).
A physicist who is not also a neuroscientist is a bit out of their depth in terms of understanding consciousness. And the theory of quantum computing is all well and good, but it doesn’t make him an actual computer programmer, or an expert on the state of the art in that field either.
He comes across more than a bit arrogant about subjects outside his area of expertise. (Of course, I could be completely wrong.)
by Marcos Marin
1) Studies show otherwise, except perhaps for a few special primates;
2) Pain implies nothing about self awareness and vice-versa (but most people seem to confuse this and other stuff in this mix too, so I forgive you);
3) He said “can” not “is”;
4) Yes, but even a broken clock is right twice a day =);
5) I am also a programmer and, having went through the Stanford course on quantum computation and mechanics I can safely assert you don’t need that knowledge to argue what he did. He WOULD be wrong though, if Gödel wan’t implicitly so there, though it is too subtle to make a difference in the argument;
6) I’m also guilty of that and you don’t see ME committing ANY mistakes now do you?;
7) In fact, insiders are often blinded by ages of indoctrination;
8) As I stated before though, he is not flawless;
9) Yes.
by Gorden Russell
Right, Xuuths, also, I do think my German shepherds have some moral sense. The big one, Franz, will chase away rowdy middle-school kids, but he won’t bother with a toddler.
by A4i
Man, we are just evolved animals. All stuff , like instincts, feelings – we just upgraded them to a new level, but that don’t make us conscious and self aware. Self-awareness needs “moar power”. It is like a tip of the iceberg. Pretty difficult to achieve and not in a realm of reach for any other kind of animal , accept apes. It needs heavy investments in brain tissue, premature birth and child dependence on parents for decades . We humans are not impalas and can’t run for our lives immediately after birth , but our brain is big enough to accommodate self awareness and advanced logic. So if we are to simulate our brain activity, better build a computer with sufficient productivity , because every simulation needs more power , than the objects it simulates.
by Milton Polk
Am I missing something? Always possible. Why is the writer of this piece seemingly not identified?
by Editor
It’s David Deutsch, as noted in the intro, but I realize it’s confusing because he’s not listed above (we didn’t have his permission to do that at press time).
by Milton Polk
Ah, thanks. Another ID10T file problem. Sounds like I need to clean my hard (really soft) disk.
by Vin
How can he say you can’t have intelligence (via creativity) if the thinking infrastructure has inbuilt constraints when the laws of physics themselves impose constraints on all matter?
That seems to say creativity doesn’t even exist in our brains (being matter too) never mind AGI. He uses this reasoning to dismiss work on programs with inbuilt constraints on thinking as being incapable of creativity and so not salvageable towards an AGI.
Well, why not just use a set of programs combined working in a system to offset their various constraints? Afterall even brains are not one monolithic lump but a modular system specialised in tasks that collectively make up cognitive process?
Sure lets have a revolution, but I’m wondering if this article or that paragraph at least is advocating throwing out the baby with the bathwater? I’d like to insert a smiley now, but everytime i do my comment gets queued for approval.
by Marcos Marin
Yes, his arguments are not flawless, but this part specifically can be understood as constraints on self-modifying code.
by Vin
Ah ok thanks. So the complaint is the progress of developing self-modifying code is being hampered by the application of inappropriate constraints. Hmm, that still seems a bit of a sweeping statement, surely not everybody is doing that wrong.
by Marcos Marin
I think the problem is they are not really trying.
It will inevitably be “sweeping”, let me be more precise:
Think of a spectrum of possible constraints, as you rightfully noted, the constraints imposed by the laws of physics — at one end of the spectrum — are not enough, as we are ‘all’ aware =), not enough to limit intelligence much.
Now, as you add constraints — through hardware and, most importantly, by hard coding them IN — you move towards the other end of the spectrum, until even interaction is destroyed (i.e. no regard for input) and only ashes remain… (i.e. passive movie watching, npr, etc =))
The important point here is, WHERE EXACTLY on that spectrum you lose intelligence/creativity etc is not clear (and is in fact one of his arguments, though he goes through a longer path), but narrow AI seems to be after that point, mostly because it has little feedback loop (it stops using its own output to improve results, and the very rules it uses to get to them) which I tried to summarize by “self-modifying code”, but data could in principle be used if you had smart enough code to begin with, which we dont have..
though sure, you could make a robot which cloned your EXACT current state in a substrate that denied further modification AND it would probably still be classified as intelligent BUT it would be crippled. The argument is that you can NOT get TO that clone within such a substrate from scratch. Only through such copy&paste (i.e. clone), OR an insanely smart, though masochistic, computer scientist. Which we dont have either =)
by Marcos Marin
>> but data could in principle be used if you had smart enough code to begin with, which we dont have..
Technically there is NO distinction between data and code. It is an artificial distinction to simplify machine microarchitecture, therefore “clever enough code” to adapt like this would inexorably require a change in behavior — said “rules” to get said results — equivalent to what is currently called “self-modifying code, THUS my one-term summary. (though it kinda defeat the purpose if I have to come around to explaining it to those who dont like generalities =P)
>> OR an insanely smart, though masochistic, computer scientist. Which we dont have either =)
Since I`m not masochistic. ;-)
by Chrispium
View it in the context of almost no progress in 60 years, then it does seem the vast majority must be doing it wrong.
by Marcos Marin
Isn’t that what Kurzweil’s book is here for? ;-)
A flawless argument, if it didn’t messed up at the end… still, AI researchers could use a little truth…
by tim333
Hmm… AI in general seems to be making great progress with Google’s cars Watson and the like. If by “artificial general intelligence” Deutsch means equalling human brains on all their many functions then waiting a decade or two for the engineers and scientists to continue their progress may be all that’s necessary.
by Tom Bellinson
AGI is not about mimicing human intelligence, which we are getting better at, but rather duplicating it synthetically. Ultimately, that will require the creation of a true synthetic consciousness. We’ve got nothing to show for that effort.
by Matthew Fuller
Unfortunately, there is expert disagreement not only on who or what consciousness is, but also whether or not one needs consciousness at all to be as intelligent as a human.
by Marcos Marin
Hmm.. I see the smarts of this place is improving… good. =)
by Bruce I. Kodish
Yes, Tom, I agree that duplicating human intelligence, or any level of biological intelligence will have to involve creating synthetic consciousness at whatever level, which in it’s most robust form will involve creating synthetic life, and will have to function in the world we live in, outside the digital circuits of a computer.
by Marcos Marin
No, he means so-called “strong” AI, not your everyday chauffeur, however “quick witted” may he be…
by Gorden Russell
Marcos, have you seen how “quick witted” that robot was that played rock-paper-scissors? Imagine how fast a robot AIG could aim and fire an assault rifle? It could take out 30 Taliban fighters in 15 seconds or less.
by Marcos Marin
Definitely a scary thought. Those idiots overestimate how clever you can get by doing ever faster and numerous dumb things. But hey, that’s how David can defeat the Goliath right? By placing a turban on the top of his head ;-)
by nfordkrz
Watson is not a “general intelligence”. It was written to analyze text presented in a very constrained format, and it took an incredible amount of memory and processing power just to do that.
On my web site (AEyeC.com) I suggest how an AGI can be constructed and how far I’ve progressed.