Recently, a user of SuperMemo wrote: I would like to know your comments about the article that claims that intense learning can lead to Alzheimer's: Does longer-term memory storage never become overloaded, and would such overload cause Alzheimer's disease and other dementia? (written by Robin P. Clarke, and published in "Medical Hypotheses").
To people familiar with the subject of senile dementia, the idea may seem absurd at first. After all, there has been quite a number of publications on the preventive impact of education in Alzheimer's. Upon closer scrutiny though, the quoted article includes lots of food for thought. At the same time, it is an excellent illustration of how the web can accelerate the progress of science. Inspiration can come not only from on-line claims that could not pass the peer-review test, but also from those that are riddled with errors, falsehoods, poor judgment, prejudice, bad intentions, or other departures from the purely scientific discourse.
Peer review vs. free speech
For over a century, peer-review science journals documented, disseminated, and perpetuated the word on the mainstream progress of human quest for new knowledge. Until recently, scientific publications were growing exponentially. Respected scientometrist Derek De Solla Price noticed that, as all good things in life, exponential growth cannot go on for ever. Indeed, the growth rate in the number of new publication has declined markedly. But Prof. Price, who died in 1983, could not have possibly predicted the new spontaneous growth of publishing on the web (unless he delved into the starry-eyed visions of Ted Nelson). Freed from the peer-review restrictions, web is now populated by scientific, semi-scientific, popular scientific, and pseudoscientific plethora of new research data, hypotheses, models, claims, and wild guesses. Without rigorous peer review, you never know what you get from the web. However, the Internet is now an unsurpassed source of new inspiration. For decades, peer review has been considered by many as a factor stifling free flow of ideas. Publish or perish principle makes scientists often move like wolves in a pack: few dare to step aside. Few want to struggle with peer review and delay publishing at the risk of perishing. No wonder then that peer review kills far out ideas. Continental drift theory proposed in 1910 by Alfred Wegener had to wait until the 1960s to be reborn as the theory of plate tectonics. Examples of ridicule in scientific community abound. As a result, many geniuses left the ranks of academics and moved to less frustrating activities like ... selling vegetables or free-lance free-thinking. Dr Robert Skoyles wrote "I witnessed many a bright student eager to prosper in the academic treadmill. They became deadwood preoccupied with marriage, mortgages and a pension". Having said the above, Skoyles set up his mouthpiece website jam-packed with interesting thoughts on neuroscience and far beyond [2011: the access to the original website was blocked in the Internet archive]. The web frees the genius and provides an open forum for the rebels. Many prominent scientists republish their materials on the net along a much wider volume of half-baked set of propositions, works in progress, word of mouth communications, rough sketches, etc. Others may prefer addressing a wider audience with popular scientific style which, interestingly, may often be a better conveyor of ideas even among his colleagues in the field. After all, many journals impose rigid formats, profile limitations, and other strict rules that enhance formal precision at the cost of the joy of discovery. Metaphoric tools of expression suffer on the way too. The net is full of errors and preposterous proposition; however, it is still a must-read for every mind looking for creative inspiration in any field. The freedom of expression produces a deluge of astrology, psychic "science", wellness advice, hate sites, creationism theories, super-learning, and what not. You will find a million articles on vegetarianism, but you will spend hours trying to sift those written with scientific data in mind, and without a religious or animal-rights bias. Even Google is helpless. Serious-looking articles sport a long list of references from prominent science journals and yet present out-of-this-world claims that have nothing to do with science. That is the side effect of free speech on the net. To the young folks, it may all be awfully confusing. However, to a trained eye, the net is all good news.
TruthRank as a global public review tool
In the wake of the PageRank system employed by Google, it is only a question of time before we see a TruthRank system where linking authors, using a hyperlink variable, testify to the degree of scientific reliability of quoted facts on linked websites. If reliable media link to a questionable material, their own TruthRank drops. Imagine searching Google by the degree of truth! Skeptics did not believe PageRank system would work. It works remarkably well. It made Google a star of search engines. TruthRank is the natural next step in the universal quest for the truth. [On March 1, 2004, Stanford University researchers published a formalization of this concept under the name TrustRank. In March 2005, Google registered the TrustRank trademark]
The quoted article on memory overload was rejected by Nature (and later Science and Lancet) and as such might be badly hit on the TruthRank side. Yet I will try to show that blocking its publication might have a negative side effect of suppressing the freedom of expression. It provides a new perspective, and new inspiration in reference to the hygiene of learning.
Evidence in favor
Glutamate and nitric oxide have been implicated in memory and nerve damage (e.g. nerve damage in multiple sclerosis). The neurological degeneration begins in the entorhinal cortex and spreads through the hippocampus, which, by its role in learning and by its neural capacity, is arguably most susceptible to memory overload. Several drugs have been developed to protect against neurological damage by glutamate over-excitation. Those drugs can, for example, slow down the damage occurring in the wake of a stroke. They work mainly by blocking NMDA receptors involved in the initial potentiation of synaptic connections in learning. Newer drugs are investigated in order to block nitric oxide synthase (i.e. the enzyme responsible for the production of nitric oxide in the course of synaptic transmission). It is true that the dentate gyrus, the site of neurogenesis discovered by Gage and Gould, is the relay station for sensory data arriving in the hippocampus. This exposes it to lots of glutamate able to produce neural damage. It is true that stress of information overload will potentiate this effect. Neural network models, in particular synaptic runaway in associative networks, provide a neat illustration of how the network might have behaved in conditions of overload. It is conceivable then that excessive learning could overload relay networks, produce synaptic runaway, and result in neurodegenerative damage through excess of excitatory neurotransmitters and other harmful byproducts of synaptic potentiation. It is conceivable that learning could do damage to your brain!
The quoted article includes quite a deal of flawed reasoning. Instead of postulating a hypothesis on the basis of evidence and looking for corroboration, Clarke starts off from confident claims that memory overload is the cause of Alzheimer's and later attempts to look for hypothetical explanations how his hypothesis might actually be implemented at the neural or molecular level. Author's view is vividly biased. The editors of Nature (back in 1995) had similar feelings, and their verdict on the article is understandable.
My main problem with the article is that if Robin P. Clarke was entirely right, you should soon expect an epidemic of Alzheimer's among users of SuperMemo. Considering that SuperMemo is already 15 years old, the epidemic should already be underway. The opposite seems to be true. Anecdotal evidence shows that early users of SuperMemo are today doing great or better as far as their mental capacity is concerned. They predominantly seem to occupy prominent positions in society. Naturally, I am aware that SuperMemo is more appealing for high-IQ individuals and the correlation may have multiple roots. With 100-120 minutes of SuperMemo per day, and a 20-year-long incurable learning addiction, I myself should have been struck with memory overload long ago. Although you can consider my opinions highly biased, I would give my right arm to vouch for learning as one of the best factors that would actually prevent senile dementia. Perhaps the trouble will begin only when users of SuperMemo start hitting the golden age of 80? Perhaps they will collapse into dementia in droves? I curiously look forward to that interesting milestone.
Here is a list of points I find most questionable in Robin P. Clarke's writing:
- No overload axiom. Clarke begins with "According to an uncritically accepted axiom, the human brain's capacity for longer-term memory storage is never overloaded". The introduction is strange in that no one with a dose of understanding of computing can doubt that memory capacity must be finite and by far insufficient to store the daily inflow of information. We all agree that there must be mechanisms that prevents memory overload. This mechanism (naturally) is (among others) forgetting
- Anti-memory drugs. Clarke postulates that drugs preventing memory formation might prevent Alzheimer's. However, memory deficits are exactly the prime pain of Alzheimer's. If the drugs were to be applied preventively, we would simply trade senile dementia for life-long moderate dementia
- Real anti-overload. Memory overload is supposed to lead to a pathology. This is a conclusion derived from the behavior of overloaded neural networks. Nature has already implemented the mechanisms of LTD (long-term depression). We observe the powerful anti-overload mechanisms behaviorally in the form of forgetting and the spacing effect. Among users of SuperMemo, memory overload manifests heavily as memory interference. We tend to obliterate old memories when learning confusingly similar material. There is nothing pathological about such overload. The only painful side effect of forgetting through interference is the need to relearn a given piece of data. New memories by default receive high-priority and overwrite older memories in need of storage. The symptoms of interference are exactly the same as in overloaded artificial neural networks
- Abduction. Clarke seems to reverse the deductive and inductive scientific thinking by applying abduction with the said hypothesis as the starting point. It is as if we wanted to prove God's existence by speculating how we could abolish facts of evolution to prove the coherence of the book of genesis. It is not the evidence that is used to construct the model, but the hypothesis that is used as the searchlight for evidence: "So how could an anti-overload mechanism operate? A possibility is that synapses in a state of LTP thereafter transmit a molecular back-signal to inhibit input from the afferent neurons". There is no need for a back-signal to the input. Potentiated synapse is protected from "hyper-potentiation" by the spacing effect. It employs efficient AMPA-based transmission that prevents NMDA activation, and prevents changes in the status of the system independent of what happens on the input. Spacing effect lasts long enough for the overload to be cleared by long-term network optimization mechanisms that operate in sleep. Clarke goes to science fiction extremes in search for a plausible backup for the hypothesis: "Conceivably a bit of data in search of a storage location involves or has the form of a free radical, neutralized on finding a storage space"
- TV fallacy. Clarke claims that studying produces fewer memories than watching TV or socializing, and uses an erroneous comparison between the sizes of video and text files. The error is double. On one hand, when considering the memory overload we should focus on memories that get actually imprinted in the brain, not on the input stream which largely gets ignored. Secondly, the brain uses efficient internal representations that may actually reverse the storage needed to preserve visual impressions as compared with textual messages (mnemonic techniques make use of this property of visual circuitry). The analogy of storage needed to keep a text file and a TV program ("100,000 times more bits") is thus erroneous. It takes less to store a simple image metafile illustrating the bifurcation of the aorta than actually describe the same vessel in words with similar accuracy. Last but not least, TV watching is so popular in the overworked population precisely for the reason that it is actually the least demanding on the side of active mental processing. Passive TV does not put much burden on the overloaded EC-hippocampal circuits. Tired TV audiences rarely trade their shows for a newspaper, let alone for SuperMemo! SuperMemo would simply be too tiring. It is true that AD patients watch more TV than controls, but research shows that the difference is made at the cost of AD-preventing activities such as reading or exercise. TV is harmful not on its own but by the virtue of taking time away from more challenging forms of leisure
- Education as prevention. Research shows that education prevents Alzheimer. Clarke believes that this may be due to extending the memory capacity in youth. However, middle-age reading is also an efficient preventive factor. If the hypothesis was right, middle-age reading should contribute to memory overload. Moreover, leisure studies show that all middle-age leisure activities help preventing Alzheimer's. Intellectual activities (e.g. puzzles) are a better form of prevention than physical exercise, which in turn is better than passive leisure (e.g. watching TV, listening to music), which in turn is better than unadulterated idleness.
- Nigerian case. Comparing Nigerian and US populations can yield revealing answers in the search for etiological factors of AD, but concluding that "much less data input" seems "highly likely" to account for the absence of AD in Yoruba is quite a stretch. Equally well other factors could play a role such as dietary (say, Omega-3 fatty acid intake), occupational (physical activity), environmental (e.g. oxidative stress risk) or survivability of affected individuals (longer lifespan will result in higher AD incidence), etc. If to neglect the mind-information link, extending this abductive reasoning to other population correlates one could claim that obesity, high cholesterol, or high blood pressure come from information overload
- Dementia-age link. The fact that dementia increases with age is not particularly striking as this is true of most degenerative diseases
- Evidence asked for. Clarke agrees that his theory can be weakened by evidence that "that overload never occurs or else does not manifest as dementia". The evidence for effective anti-overload system is plenty. The mechanisms are well known: forgetting, spacing effect and memory interference
- Data errors need not mean pathology: the fact that neural networks produce errors in data on overload by no means indicates this is related to a pathology. Healthy networks will produce errors as soon as the data input surpasses their capacity. Analogy to neural networks is correct but parallels to Alzheimer's are wrong
- Evolutionary argument: Clarke believes that evolution developed anti-overload mechanisms but fails to list any fact-based reason for which this mechanisms should go awry. Any balance between forgetting, spacing effect and LTD produces a compromise between storing important pieces of information without ever producing a pathological effect
- Anti-overload signal: "Natural selection will favor further mechanisms, which enable local matrixes nearing overload, to signal their lack of spare capacity, thus activating diversion to other locations". This sounds exactly like the job of NREM-REM sleep interplay. Optimizing the storage is the simplest defense against memory interference. Sleep may act as an anti-overload and anti-interference mechanism that does not show the same destructive powers as forgetting. The signal on the "lack of spare capacity" might simply be adenosine-based homeostatic component in the two-processes sleep model
- Animal models: If "the theory obviously predicts that an effective animal model of AD could only be produced by replacing the extremely unstimulating life usual for most laboratory animals with much more data-rich circumstances", why is it that rat funhouse experiments produced animals that are by far more agile, nimble and smart. The more data-rich the environment, the happier the rats
- Bias: Clarke consistently points to cases in which old memories are overwritten but still attempts to use these examples as proof of a link between learning and dementia. This sound like a proof at any cost: "Indeed such delusions [resulting from missing pieces of information] are the primary symptom in some cases" to add two sentences later "the fact that such symptoms are not predominant merely indicates that the anti-overload mechanisms usually do not fail".
- Alzheimer's is the failure of the memory system: "If a person's old memories fade or are lost as fast as new ones are acquired then that person will not develop overload. Any such mechanism of ongoing fading or loss would greatly reduce any memories spanning a lifetime. But not only does this find no support in experimental studies, but is also decisively refuted by experience from old people many of whom have more numerous memories of eighty-ninety years previously than of more recent times". Omnipresent fading is beautifully illustrated by the universal Ebbinghausian forgetting curve, and the experience of the elderly affected by dementia is not a result of a memory system in action. It is a result of the failure of that system
- SuperMemo epidemic. Again: If too much learning leads to Alzheimer, we can expect a massive epidemic of AD among users of SuperMemo. After all, they use the shortest path to stable memories that are supposed to produce the postulated overload. SuperMemo may be the best evidence around against the claim. In its 15th year of existence, it should have already yielded many individuals with memory loads reaching far beyond that of a typical octogenarian. Of what we observe among users of SuperMemo, we could tentatively state that the opposite seems to be true. When reasonably used, SuperMemo can be considered a tool of mental hygiene (for plethora of evidence see supermemo.com)
From chaos to the truth
Some people raised fears that the Internet will make it easier to spread pseudoscience, hatred or pornography. They were right. However, the solution should not be an outright suppression of zany or offensive material. It is true that it is extremely easy to hit a site deriding Islam as a "religion of hate and conflict". But it is equally easy to jump, with a few clicks, to a counter-site with anti-zionist material soaked with no less hate. In that sense, the Internet is less damaging than the printed matter where it is more customary to read lengthy (biased) materials cover to cover. Cover-to-cover implies a more stable imprint. The same refers to pseudoscience. If you use incremental reading to process fringe materials, you always run the risk of picking misguided inspiration. But in the sea of contradiction, the truth is the only survivor in the end. Let's try to reconcile the evidence for and against
Memory overload in learning
However much I have criticized the "memory overload" hypothesis above, one inspiration emerges. Overloading short-term memory circuits can indeed be unhealthy to their neural fabric. If runaway synaptic network model is true, if excess glutamate or nitric oxide are able to damage neurons, if the entorhinal-hippocampal relay is indeed subject to saturation, then running against the principle of mental hygiene could indeed result in the damage to the neural tissue. At the same time, well-paced and well-structured memory training could work as the best way of preventing Alzheimer's. This could work on two fronts: (1) the well-documented trophic effect on the nervous tissue, and (2) the systematic establishment of harmless neocortical long-term memories. Doing harm and good with the same factor (learning) do not need to be contradictory. The same happens to nearly all forms of exercise: you can over-train or apply a wrong training regimen to produce a negative effect.
Evolutionary reservations could be cut short by the observation that human populations (even as recently as in the 19th century) did not use to predominantly live to the ripe age of eighty; therefore, it could be cheaper to retain the present flawed memory overload handling than to perfect the system for the sake of longevity. Beyond active reproductive years, longevity does not translate to an evolutionary advantage. Even genetic predispositions related to APOE4 allele, APP, or presenilin genes would do little harm enough to stay afloat in the population. Especially that they seem to surface only in industrial societies where lipoprotein metabolism malfunction is rampant.
It seems plausible then that abuse of neural networks could lead to Alzheimer's while proper use of the same networks would lead to a long term benefit. In the same way as exercise leads to better health while immoderation in sports can be dangerous. Well-managed touch typing leads to high typing speeds while poorly-managed typing can equally well lead to repetitive strain injury. If there was to be a link between memory overload and AD pathology, using SuperMemo along the minimum information principle could be compared to moderate exercising, while cramming would be a form of running a marathon without preparation.
Whatever is said about the healthy brain may not necessarily apply to the optimum strategy for the treatment of Alzheimer's. On one hand, spaced repetition could help establish memories vital for survival. On the other, all forms of mental rest might also have a therapeutic value.
If we take all the above data into consideration then we could propose a few rules that could ensure your learning adventure does not go in a wrong direction:
- knowledge representation - few things are as stressful in learning as wrongly formulated material. If it is not comprehensible, if it is hard to remember, if it is complex, it is literally able to give you a headache. See: 20 rules of formulating knowledge for learning
- circadian rhythm - mental capacity rapidly declines in periods of intense intellectual work. This is particularly true if we learn new things. Sleep might be the chief anti-overload protection mechanism. It helps unload separated neural representations from the hippocampus. It optimizes the long term neocortical overlapping representation. Learning with a fresh mind after a good night sleep will then be recommended. Learning in condition of sleep deprivation or mental fatigue would be considered wrong. See: Good sleep for good learning
- information overload - rational and methodical approach to information overload is vital. Well-managed incremental reading might be the best way of avoiding information fatigue syndrome. It also works great for resolving equally stressful contradiction (neural interference). See: Devouring knowledge
- pleasure - the brain has undoubtedly developed mechanisms that excellently resolve the complex optimization issue where vital survival information needs to be coded in a limited space of neural memory. In humans, those mechanisms are particularly sophisticated and provide for an impression that we cannot run out of memory space in lifetime. Apart from lower level mechanisms such as the spacing effect, forgetting, LTD in conditions of interference, etc. we have definitely involved the reward system that acts in the process of learning. For example, in recognizing information overload (too much information), boredom (too little relevance of information), etc. For these reasons, the pleasure of learning can and should be the primary indicator of the adherence to the principles of mental hygiene. If you love learning, probably no damage awaits you in the process. If you get tired, bored or restless, take a break, take a nap, go for vacation, or rethink your learning strategy.
Memetics in action
Once a meme virus of memory overload in AD is unleashed on the net it will propagate well in circles that oppose industrial growth, mass media culture, globalization, global transport, public schools, television, etc. It will also spread well among young folks who would just pick any excuse to do less learning. A counter meme of the therapeutic value of learning will propagate well among those involved in promoting learning tools or those who develop learning materials (here goes SuperMemo). When these memes clash, the truth is likely to see the light. The truth here may be analogous to the truth about the value of jogging. It can damage your knees. It can boost your glucocorticoids. Joggers smile less than walkers. Yet well-managed jogging might be the best known cardiovascular exercise.
Ignorance is always a good medium for false memes. Incremental reading loaded with contradiction helps resolve such ambiguities. The net is a wonderful breeding ground to promote psychic powers, astrology, reincarnation, UFO, tarot magic, dianetics, dowsing, neurolinguistic programming, clairvoyance, etc. However, there is no choice but the submission to free speech and competing ideas. The truth will come out in the end for those who care to see it.
Great sites for skeptics:
- Skeptic's Dictionary (by Dr Robert T. Carroll)
- Quackwatch (by Dr Stephen Barrett)
- James Randi Educational Foundation
Year 2011 follow-up
Robin Clarke submitted his reservations to this article. They can be found here: