KNOWLEDGE AND INTELLIGENCE
THE FUNDAMENTAL THESIS OF THIS BOOK IS THIS:
1. a person with an IQ of 200 is vastly more likely to be healthier, richer, happier, more productive, more successful, better connected, and freer from stress than a person with an IQ of 100 .
2. The person with an IQ of 200 has in his brain a massive knowledge base which, in addition to raw facts, words, experiences, and images, has, more importantly, a vast knowledge of methods for acquiring still more knowledge and heuristics (rules) for solving problems and relating to people and opportunities.
3. If the person with an IQ of 100 was given the knowledge base of the person with an IQ of 200, then a key part of the difference between them would disappear. The person with an IQ of 100 would then be able to enjoy the benefits of the smarter person.
4. If the person with an IQ of 200 was given the knowledge base of a hypothetical individual with an IQ of 400 , then he would enjoy far greater success, and so on.
5. Right now, to a limited but still quite important extent, we can use computers and the internet, using methods I will describe, to accomplish 3. and 4. above. In this book I will show you how in great detail.
6. The combination of you plus your computer and network, so configured, will be far more powerful as a problem solver than you alone. Even without your computer, you will be smarter by virtue of remembering more and taking hold of better methods for acquiring still more knowledge.
7. Our computer hardware, software, and networks for accomplishing this merger of YOU + IT will be exponentially more powerful in the near future.
8. In an iterative loop, this cycle of more powerful computers assisting more human beings will create an explosion of knowledge and creativity .
* * * * * * * * * * * * *
I realize some of the assumptions and claims above may arouse skepticism or even out-right rejection. Addressing paragraph 1 above, many of us may know smart people who are miserable or poor or unhealthy or who have made foolish mistakes or who are riddled with anxiety or worse. My claim is that these individuals are exceptions. More intelligent people, taken as a statistical cloud, are significantly to the right along the scales of success than are less intelligent folk. The evidence, which is abundant, substantiating this claim will appear in later chapters.
In paragraph 3 above I claim that embuing the individual with the lower IQ with the knowledge base of the smarter individual would enable him to enjoy the same benefits. Surely I'm speaking approximately. Of course. As they say, your results may vary.
INTELLIGENCE AUGMENTATION: IA rather than AI
My thesis bears a resemblance to Moore's Law and to Ray Kurzweil's more general formulation of it, the Law of Accelerating Returns, discussed in his recent work The Singularity Is Near .
However, my book is predominantly about human minds and human brains , specifically yours, and not primarily about computers (although artificial intelligence (AI) has been a major focus of mine for decades). Through this particular work I am interested in making the human race - starting with you - smarter, rather than producing smarter machines. ( I am also quite interested in artificial intelligence; however, for the next decade or so, AI will be improved mainly by more and smarter engineers working on it.) Even without Moore's Law, the methods described in this book - (among many) augmenting human minds using knowledge ware - can proceed with their own cycle of recursive improvement independent of progress in computer hardware or in invasive brain modification.
Rather than AI, then, this book is about IA. IA is intelligence amplification or augmentation.
Before getting into the nitty-gritty details, let's have some fun.
The Power of Super Intelligence: Let's Go to the the Movies
If I haven't already captured your interest, let me try again.
No one succeeds at showing the power of super intelligence as well as writers and novelists. By presenting some of their efforts, you will see aplenty what happens when common folk are transformed into geniuses.
FLOWERS FOR ALGERNON
One of my favorites in this genre is the book and the movie, Flowers for Algernon by Daniel Keyes. I first discovered this story when I was an undergrad at MIT in the sixties. (Interestingly, Ray Kurzweil was a classmate of mine at MIT.) When it appeared in the early sixties, Flowers swept the top sci fi literary awards, the Hugo and the Nebula, and the movie won an Oscar.
Here's an edited version of the description that appears in WIKI (ref).
The story of Flowers for Algernon centers on Charlie Gordon, a mentally retarded janitor who volunteers to take part in an experimental intelligence-enhancing treatment. Poor Charlie has just enough smarts to eat, sleep, and walk to work to do his janitorial job. Meanwhile, a group of scientists have developed a drug that seems to produce the capacity for super intelligence and they have tested it on a lab mouse - Algernon. Charlie volunteers for the experiment barely understanding what is going on. The story is told from Charlie's point of view and is written as a journal, in progress reports or progris riport as he initially spells it. Charlie is quite fond of Algernon and takes a great interest in his well being. After Charlie is injected, his progress parallels that of Algernon, who had been "enhanced" earlier.
Succeeding entries in Charlie's diary trace his ever-increasing comprehension and intelligence in the aftermath of the treatment, as he passes through normalcy and then reaches super genius level, becoming vastly more intelligent than the doctors who invented the procedure (he learns 20 languages, reads books at one page per second, writes a piano concerto, makes Nobel-worthy theoretical discoveries and disproves the hypothesis of the experiment that he was the subject of. He discovers both the advantages of intelligence, including his sexual-emotional relationship with his former teacher, Alice Kinnian, and the disadvantages, as he discovers that the people he thought were his friends had only viewed him as 'entertainment', and now resent his superior abilities.
Yet, all else seems to be well, until Algernon's enhanced intelligence begins to fade rapidly.
As Charlie himself confirms, the neural enhancement is only temporary, and he too is doomed to revert to his original mental state. Algernon's death seems to presage Charlie's. In a vain but herculean effort Charlie furiously combs the literature of neurobiology and cognitive psychology in a futile effort to stave off his certain mental decay and impending death.
Ah - triumph and tragedy - on the silver screen. Whatever else you may think of Charlie, he was much happier when he was smart.
I can only begin to hint at the catalytic effect that this movie had on my undergraduate studies of neurobiology and artificial intelligence. However intense my studies of electrical engineering, psychology, neurobiology, and AI were up to that point, after seeing the movie and then reading the book, they were attacked with a singlemindedness that bordered on religious zeal. To pursue these studies I literally moved into the MIT undergraduate library (which was open 24/7) for the next several months. (You must be curious: I also slept in the library, showered and kept a change of clothes in the gym, and ate all my meals at the grill. I entertained my dates in the piano practice rooms.) My path was obvious: first solve the intelligence-enhancement problem . Then solve all of the rest of the world's problems using this newly found intelligence.
This sentiment is echoed below: The development of machines with superhuman intelligence is the last development that the human race need accomplish. The rest will be done by them. or by this quote from I. J. Good Let an ultraintelligent machine be defined as a machine that can far surpass all the intellectual activities of any man however clever. Since the design of machines is one of these intellectual activities, an ultraintelligent machine could design even better machines; there would then unquestionably be an 'intelligence explosion,' and the intelligence of man would be left far behind. Thus the first ultraintelligent machine is the last invention that man need ever make.
Another inspirational example is the education of Superman. While not acclaimed primarily for his intelligence (after all,he's no match for his arch nemesis, Lex Luther,) Superman is no slouch in the smarts department. In addition to Superman's xray vision, speed, and collosal strength he is also shown (in the wonderful 1978 film Superman: The Movie starring Christopher Reeve) to possess great wisdom, and we are shown precisely how he acquired it. That wisdom was passed to him by his father, Jorel, a respected leader of the destroyed planet Krypton. Jorel teaches Superman during his long space voyage to Earth and also in his Fortress of Solitude when his image is evoked by holographic recordings contained in crystals given to him by his father. He learns the history and detailed knowledge accumulated by the civilizations of his forebearers, and uses that knowledge to advance the cause of truth, justice, and the American way. (His values are also nicely augmented by the teachings of his adoptive parents on earth, who cherish and nurture their wonderful gift from space.) While some may question the current justice of the American way, there is no questioning Superman's having benefited greatly by his father's imparted knowledge and wisdom.
STAR TREK: THE VOYAGE HOME
Spock's mind has been wiped clean by his death and resurrection in the previous Star Trek movie. In an effort to re-educate him from an extremely low level of functioning he uses a tutoring machine of the future. In short order we watch him relearn all of history, astrophysics, and the vast body of knowledge accumulated by the Federation. Having regained his former brilliance within a few days, he is only stumped by really hard questions like that posed by Captain Kirk: Spock. How do you feel? Despite that gap in his abilities, he is later able to easily impart the formula for transparent aluminum.
THE LAWNMOWER MAN
Of course, massive knowledge may not necessarily be used only for laudable aims. In the 1992 movie The Lawnmower Man A scientist performs experiments involving intelligence enhancing drugs and virtual reality on a simple-minded gardener. (I myself was quite the lawnmower man in my early teens. I cut all the lawns in our neighborhood in Berkeley to pay for my ham radio kits and my chemistry gear.) Back to our story - the scientist puts the gardener on an extensive schedule of learning, and quickly he becomes brilliant. But at this point the gardener has a few ideas of his own on how the research should continue, and the scientist begins losing control of his experiments (ref IMDB). The plot concludes when the Lawnmower man takes over the world and enslaves mankind. Not quite the hoped-for result.
Well, by now the image should be plain: an ordinary person, living an ordinary life, transformed into a phenomenal intellect with all its attendant benefits and consequences. There are numerous other examples from pop culture and science fiction literature - David Bowie's The Man Who Fell to Earth, John Travolta's 1992 movie Phenomenon, The Terminator films, and many others. But, you've got the idea.
Now, can we do it for you? Just roll up your sleeve while I get the syringes of intelligence-enhancing drugs. Then take off your hat, we'll paste on some scalp electrodes, and then connect them to my virtual reality machine.
Let's get back to earth and present day reality.
Picture 1: Earth and a human head (an ordinary guy)
On the left is planet earth, the third planet from the Sun, a minor G3 star in the Orion Arm of the Milky Way. Perhaps otherwise undistinguished (although this is disputed (Rare Earth ref)), over the past few thousand years, several phenomena of interest have occurred. First, the human race has greatly expanded its numbers from a few million to about six and half billion. In doing so the human race has subjugated the planet's non-human inhabitants and surface resources for its own purposes. Using these items for food stocks, energy sources, transportation, residences, and commerce, humanity has profoundly altered the surface and biosphere of planet earth. For the moment let me set aside these minor phenomena.
The enabling special feature of mankind that has enabled it to do the above is its artifacts and knowledge . Giraffes have long necks, spiders have webs, we have our artifacts and knowledge. This banter you read about the importance of bipedal locomotion and prehensile thumbs is really only of historical interest.
By artifacts I mean all of our stuff - all of it. houses and other buildings, cars and roads, planes, every machine, all clothing, all food (almost none of it is natural by the way; it has all been highly hand-crafted through generations of selective breeding.) : everything created by us and by our machines. Then there are the artifacts that explicitly transmit knowledge: televisions, newspapers, books, libraries, computers and network servers, journals, magazines, movies, schools, and universities. More interesting, as a medium of knowledge storage and transmission is the totality of our 6+ billion brains (increasing in number by about 100 million per year).
The knowledge piece is interesting - for many reasons. We not only know things, but we have a sense for what we know and don't know - an inventory. For example, I know that I can read CT scans of patients in the emergency room (as can every ER doc) but that I can't read German, even though I could a bit in high school. Furthermore, among my friends I have some sense for what they know and don't know. That I have a feel for whether I know something or not is perhaps not so surprising. After all, I was there when it went in - usually with great effort.
Here, in a nutshell is how humans (or any other intelligent agent) work.. It's like baking a cake. (I know that I don't know anything about cooking - ask my daughter.) You need an oven connected to an energy source. You get a recipe. You buy the ingredients and put them in a pot in the sequence and manner dictated by the recipe. Then you wait and eat your cake. There it is: everything from tying your shoes to building the International Space Station or the 28 kilometer long Large Hadron Collider (LHC): knowledge + artifacts. The knowledge tells the agent how to manipulate the artifacts to create new artifacts.
KNOWLEDGE POTS AND MENTAL MODELS
Each of our heads is about the size of a pot. In the course of growing up and going to school we fill up the pot with knowledge and then use it to get on with life: to do our work, to accumulate artifacts, to mate, and to otherwise try to enjoy life. Some cognitive scientists suspect that the pot is not empty even at birth (ref: Pinker) but let's defer that.
So, what's in the pot that enables us to bake a cake or build a space station. Knowledge. Yes, but what sort of knowledge and how is it stored? Let me use my computer as an example. (I know much more about computers than about cakes. Food for me is just what keeps the brain going.)
What enables me to fix my computer when it stops working? I have a detailed mental model of how it works from my school courses and subsequent reading of computer science. I know that when it is turned off, my files are only stored on the hard drive and no where else. I know that the information that enables it to boot up (to load the operating system into DRAM) is stored separately on the BIOS. These and thousands of other relevant memories are stored in my cerebral cortex. Now, in what form are they stored: as sentences in English (definitely not in German or Mandarin)? No, not even in English. They are stored in a form that has been called mentalese (ref: Fodor, Pinker etc.) Note: I'm not talking about the physical encoding of memory in the neural tree. We'll save the neruobiology for later. I'm talking about a higher level description of human memory. I know its there because I personally can't work without it. Furthermore, it has been extensively described in the literature of cognitive science and psycholinguistics. That too we will hit in a later chapter. Here in Chapter 1, I'll just give you a flavor for what is meant by mentalese.
THE MIND'S EYE
In your mind's eye please imagine each of the following:
a hunchback walking with a limp (as in the Hunchback of Notre Dame)
Santa Claus walking after a full meal
an old model T Ford, belching smoke, being driven down an unpaved road with potholes
That's mentalese, folks - and it ain't English! For me I see little faint, sketchy movies that I can watch on command and even manipulate with my mind's eye. Its as though the original movie was highly filtered to transmit it via an old acoustic modem over my telephone. (Unfortunately, the older I get the fainter the movies are; for example, when I play chess with my son, he doesn't need a board at all, a skill that I had at his age but, alas, no longer.) Interestingly , I can even use these mental models to reason with and to simulate contemplated actions. Watch out Mr. Hunchback! A visious german shepard is charging after you, slavering and snarling. Can you get away? Look at that. The hunchback was actually a portable jet pack and he's rocketed up in the air.
Do we reason with mental models? We do. Here's another more example. Here's a picture of a computer's hard drive and a detailed blow-up of a magnetic domain with the drive head over it.
(Picture of a platter of a hard drive and a magnetic domain.) I can't talk about hard drives without seeing a whirling tan platter in my minds eye. I know that the zeros and ones are stored as little magnetic domains on the platter, and they are divided into tracks and sectors on the platter. I had lots of experience in my youth with magnets and phonograph records so those images helped me understand hard drives. How about debugging a circuit in the computer? That knowledge starts with the image of electricity flowing like water, only on a wire. If the wire is broken, I can see that the electricity cannot flow. It doesn't matter whether the wire is a 120 kilovolt power line on a tower or a 45 nanometer copper channel on a chip, its more or less the same mental model - at least for the purpose of debugging. Do these models actually exist and are they, in fact, used by the mind for reasoning. Absolutely, x psychologist did experiments in which he asked his subjects to solve IQ test visual imaging problems and then he measured the time it took them to solve the problems. The time was precisely what would be predicted if the subjects mentally rotated the models in their mind's eye. The experimenters were even able to calculate the speed of rotation in the mind in degrees per second. I can even use my mental models to predict what would happen if I hit the hard drive with a hammer or put it in liquid nitrogen. A more challenging task would be to predict what would happen if I jogged with it or whether it would work inside an MRI scanner. Little mental movies.
How accurate are these grainy little models of reality? Not bad - good enough for some purposes, not enough for others. (I was climbing up a steep talus slope in the sierra nevada mountains a few years ago with a friend late in the afternoon. I looked at the very steep wall of snow and ice that was beside us and in my mind's eye I thought we could slide down it on our bottoms. In my imagination I remembered taking falls like that while skiing, stretching out my legs for stability. When I launched myself down the slope this time, I fell at high speed rotating rapidly and broke my arm in three places. Little mental movies. What do we have mental models of? Everything that we know; everything we have learned in school and otherwise, particularly those areas that we have really studied and attended to. For me, I've always loved science and technology but as an adult the utility of knowing about other things has become apparent: history, business, current events, organizations and governments, large and small, and especially people and relationships. The sum total of the knowledge in each of our heads is formidable. Don't forget the skills like getting a job, mowing a lawn, fighting a court battle, behaving at a meeting, driving a car, riding a bike, and so on. (There is a major difference, by the way, between knowledge that we can talk about like the plot of Moby Dick versus say riding a bike or juggling. The former is called declarative knowledge, which we can consciously evoke on command, and procedural or non-declarative knowledge which is just in there somewhere (like the cerebellum or basal ganglia) but which we can't evoke (except for tacked on descriptions of the primary experience). Obviously we also have detailed knowledge of lots of people: what they know and don't know, their likes and dislikes, their opinions, detailed images of their faces, bodies, and how they move, and their physical capabilities. Besides our family members, we also know people on tv and the movies, and also strangers that we encounter. That guy in the grocery aisle - is he from Alpha Centauri with a complement of totally alien internal organs - or, more likely, is he just like you and me. And, if so, what does that mean? Perhaps he has a cochlear impant, a titanium left knee, was a cryptographer and a former ambassador to Poland.
How big is the whole collection? How many models and what is the level of detail and complexity of each model. Various estimates have been proferred. No one knows with any precision and, of course, it depends on our age and extent of experience and education. To get in the ball park, let's just say several thousand. Furthermore, the complexity of each model depends on our detailed experience with the item in question. Just like the linguist Benjamin Wharf's notorious eskimos (who were alleged to have dozens of different words for snow corresponding to their detailed perception of same) my models of snow just extend to what I like to ski on versus what I'm apt to break a bone on. Let's set a side for the moment the question of how many bytes it takes to store a model, and the more important question: is there anything else stored in the cerebral cortex that enables us to solve problems. For now just note that there's a lot more to a human's knowledge base than just mental modes. For starters there are also values, opinions, states of belief, hyptheticals, non-visual knowledge and experiences, multi-sensory memories of personal events forming a chronologic, autobiographical memory, and much more. (note: discuss Gordon Bell's total memory project later)).
This is the latest of my incarnations as a student of artificial intelligence. My professional career has tended to alternate between computer science and medicine with a periodicity driven by various externalities. (During my era as a researcher in the Stanford Computer Science Department in the 1970's and 80's, it was common to speak of knowledge bases and inference engines. Here's the idea.
Picture: box labelled KB and box labeled Inf. Engine.
The KB was seen to contain a large collection of knowledge encoded in some form: for example if-then rules, logical statements in predicate calculus, linguistic parse trees,semantic networks, frames describing situations, or scripts describing typical sequences of events. The Inference Engine would pluck out one chunk of knowledge from the KB at a time and use it to make an inference.
MYCIN, for example, was a computer program designed to diagnose and treat infectious diseases like meningitis. Assisting with augmenting and testing the knowledge base of MYCIN was my initial job when I came to Stanford in 1976 after I finished my residency in internal medicine.
MYCIN's inference engine would pull a rule out of the knowledge base, which consisted of a few hundred rules, and execute it. Here's what I mean. The program had as its top level goal trying to determine whether a particular patient had meningitis, and if so what treatment to recommend. Somewhere, along its chain of reasoning it might be executing a rule like this one:
IF the patient is immunocompromised,
AND the patient has a positive TB skin test,
OR the patient has suggestive lung lesions,
THEN suspect that the patient has tb meningitis.
This nugget of medical knowledge would be processed word by word and phrase by phrase by the inference engine. So, for example, when it hit the word immunocompromised, it would have to determine whether that statement was true or false (or more specifically, its certainty or probability). To do that, it would then pull out of the knowledge base every other rule in the KB that had anything to say about whether the patient was immunocompromised. For example, there might be another rule that would be plucked out from the KB.
IF the patient has been on high-dose steroids for more than 2 weeks
OR the patient has been on cancer chemotherapy
OR the patient has a low white blood count
THEN the patient is immunocompromised.
In this fashion, the inference engine would back-chain through the relevant rules in the KB until it arrived at a diagnosis and therapy. (By the way, MYCIN was quite good at doing its task as we showed by doing detailed comparisons with expert diagnosticians. It was a problem for MYCIN and for every other expert system. Their expertise was extremely limited. Furthermore, it took a lot of hand-crafting by experts to keep the knowledge bases up to date.)
In this book, however, I'm predominantly talking about reasoning by human beings and not by computers. So, what was the purpose of this digression from mental models into AI? It's this. One might well ask, harkening back to our primary goal, well ok, you give the individual with the IQ of 100 a large KB. So what? Isn't that just like giving an ordinary guy a really big encyclopedia and saying now, go get your Nobel prize ? Isn't the common man with the average IQ just a relatively under-powered, low horsepower inference engine - perhaps with just the added advantage of a giant tank of gas? What would poor Charlie from Flowers for Algernon do with a big encyclopedia, even an on-line one with lots of pictures and movies?
Rather than speculating about Charlie (and we will also examine that subject in later chapters) let's talk about you, my reader, who have no trouble following my argument and perhaps are even prepared to spar with me on grounds that I haven't even conceived of. Can we make the two of us a lot smarter and intellectually more capable? I'll bet we can.
Metaknowledge is one of the most important concepts I will present in this book: knowledge about knowledge . In fact, this whole book contains metaknowledge, since our primary subject is knowledge. (Consider title change to Terabyte Knowledge and Metaknowledge.)
Metaknowledge is what teachers learn when they're taking courses on teaching. Here is a student. How do you impart knowledge more effectively. The techniques may be relatively the same regardless of the subject or domain. If you go too fast and speak in a monotone and don't take questions, you will bore students in not only math but also sex education.
Besides having more mental models, revved up Charlie (post-injection) also learned how to learn . As the fog lifted, he began to read, not only math books, but books about motivational and cognitive psychology, neurobiology, pedagogy, didactics, linguistics, philosophy, and epistemology. There he might have learned invaluable principles of health and physiology that promote knowledge and skills acquisition, about the importance of building an entire framework of knowledge within which to embed each newly acquired particle ofknowledge. He would have learned about the importance of writing, speaking, problem solving and reformulation as methods for broadening understanding and heightening retention. You get the idea.
Beyond these very general and widely accepted principles of education, there is a large body of more specific, domain-level metaknowledge. By domain I mean a specific field like clinical medicine or theoretical physics or field biology or class action litigation or city government. Obviously each of those fields has its own large body of highly specific knowledge. But that knowledge not only has a specific organization and character but each domain has its own specific metaknowledge that is widely shared, although perhaps not explicitly taught or codified by its practitioners. Its easiest for me to give you examples from clinical medicine, since I've worked in it for decades. Every few years physicians in most specialties need to take recertification exams. When I prepared in 2005 to take the recertification exam offered by the American Board of Emergency Medicine I knew exactly how to study, having done it every ten years or so previously. Besides the specific details of diseases, pathologic processes and injurious agents (everything from rabies to radiation), there are scripts that every ER doctor is familiar with from his training. What are the sequence of events in a cardiac arrest resuscitation, the stages of shock from blood loss, the signs of stroke, methods for distinguishing delirium from dementia from psychological or hysterical dysfunction? The list is endless, but for each pattern, you not only learn the specifics of the disease but how to learn that particular class of ailment and how to retain it. This was hard fought knowledge, gained over years of study, but once gained it is relatively easier to freshen up.
One of the core competences of ER physicians is intubation , that is rapidly placing a plastic breathing tube into the wind pipe (the trachea) of a patient who cannot breath. Once respiration has stopped, that patient will incur irreversible brain damage within minutes. Its not so difficult unless - and this is frequently the case in the ER - the patient is obese, cannot open their mouth, their neck is injured, their semi-conscious and fighting you, their mouth is full of vomitus or saliva, or you just cannot see the inside of their trachea no matter how hard you look. OK. I and all of my colleagues around the USA have prepped for this situation endlessly. Here is an example of some domain specific metaknowledge about learning intubation. Every time I learn it I use the same set of about four diagrams, three tables consisting of sequences of procedures, perhaps a dozen descriptions of drugs and their side-effects, and also a detailed equipment list and review of manual skills. That's it. When I work, I've recently reviewed all that information and, preferably, I've got those pictures and lists right at hand. Furthermore, for other emergency situations I know exactly how long I've got to get needed information, where to look for it, and who to call to get it fast.
Metaknowledge is knowledge of the tricks of the trade and every trade has its tricks. (to be continued) EARLY DRAFT copyright 2007 by Robert L. Blum