惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Cyber Security Advisories - MS-ISAC
Cyber Security Advisories - MS-ISAC
Application and Cybersecurity Blog
Application and Cybersecurity Blog
T
Threat Research - Cisco Blogs
Latest news
Latest news
Project Zero
Project Zero
TaoSecurity Blog
TaoSecurity Blog
Cyberwarzone
Cyberwarzone
Threat Intelligence Blog | Flashpoint
Threat Intelligence Blog | Flashpoint
Google DeepMind News
Google DeepMind News
P
Privacy & Cybersecurity Law Blog
T
Troy Hunt's Blog
cs.AI updates on arXiv.org
cs.AI updates on arXiv.org
AWS News Blog
AWS News Blog
Hacker News: Ask HN
Hacker News: Ask HN
S
Security @ Cisco Blogs
C
Cisco Blogs
Help Net Security
Help Net Security
I
Intezer
W
WeLiveSecurity
Exploit-DB.com RSS Feed
Exploit-DB.com RSS Feed
腾讯CDC
S
Secure Thoughts
MyScale Blog
MyScale Blog
Recorded Future
Recorded Future
G
GRAHAM CLULEY
L
LINUX DO - 热门话题
A
About on SuperTechFans
C
CXSECURITY Database RSS Feed - CXSecurity.com
IT之家
IT之家
J
Java Code Geeks
The Hacker News
The Hacker News
阮一峰的网络日志
阮一峰的网络日志
Scott Helme
Scott Helme
Recent Announcements
Recent Announcements
AI
AI
Cisco Talos Blog
Cisco Talos Blog
B
Blog RSS Feed
V
Vulnerabilities – Threatpost
C
Check Point Blog
Security Latest
Security Latest
S
SegmentFault 最新的问题
T
The Exploit Database - CXSecurity.com
cs.CL updates on arXiv.org
cs.CL updates on arXiv.org
M
MIT News - Artificial intelligence
T
The Blog of Author Tim Ferriss
Attack and Defense Labs
Attack and Defense Labs
PCI Perspectives
PCI Perspectives
Recent Commits to openclaw:main
Recent Commits to openclaw:main
T
Tailwind CSS Blog
Apple Machine Learning Research
Apple Machine Learning Research

ashishb.net

A day in Luxembourg - the richest country in the world I was asked to install malware during a fake interview Book summary: Breakneck - China's quest to engineer the future by Dan Wang Book summary: How to Teach Your Baby to Read Book Summary: The Discontented Little Baby Book by Pamela Douglas Introducing Amazing Sandbox - run third-party tools and AI agents securely on your machine Why software outsourcing gets a bad reputation? Book summary: The Natural Baby Sleep Solution by Polly Moore A day in Antwerp, Belgium Journey of online influencers Two days in Brussels, Belgium Shortcuts - when we love them and when we don't A visit to Rakhigarhi Three days in overhyped Paris Empty Japan, crowded Tokyo The real lock-in in GitHub is not the code, but the stars 11-day Norwegian Breakaway East Caribbean cruise Sanskrit and Sri Lankan Air Force Use REST with Open API The Achilles heel of American capitalism Costa Rica in 4 days At a juice stall in Sri Lanka A short stay at Warsaw, Poland Best practices for using Python & uv inside Docker Two days in Vilnius, Lithuania How IntelliJ IDEs waste disk space Pregnancy Why there aren't many digital nomads from India Two days in Riga, Latvia To keep your machine secure, run third-party tools inside Docker Family Ties in Your DNA: Some relatives are closer than others Doctors per capita Two days in Tallinn, Estonia Ship tools as standalone static binaries Made in America Two days in Helsinki, Finland Maintaining an Android app is a lot of work The land of good deals Two days in Oslo, Norway FastAPI vs Flask performance comparison Google Search is losing to Perplexity Two days in Dublin, Ireland Continuous integration ≠ Continuous delivery World's simplest project success heuristic London in 5 days It is hard to recommend Python in production Inflation, IRS, Credit cards, and Vendors Temu and the Chinese approach Things to do in Miami Florida Revenue vs Cost Axis Language learning as an adult The unanchored babies of the green card limbo Price variance in the United States A day in Louisville, Kentucky A surprisingly positive experience with Air India Unhospitable Airports Android: Don't use stale views USA = Union of Sales and Advertisement A day in Nashville, Tennessee Minimize Javascript in your codebase A day in Birmingham, Alabama In defense of ad-supported products Real vs artificial world The science behind Punjabi singers Hiking Mt. Fuji The Indian startup bubble is insane Repairing database on the fly for millions of users Book Summary: One up on Wall Street by Peter Lynch It is hard to recommend Google Cloud At the Prague airport Kyoto in three days Migrating from WordPress to Hugo Book summary: Sick Societies by Robert B. Edgerton Statistical outcomes require statistical games Illegal immigrants to Europe via Cairo Tokyo in three days Mobs are Status Games Writing Script matters as much as the spoken language Sri Lanka in 5 days LLMs: great for business but bad business Book Summary: Safe Haven by Mark Spitznagel Mac shortcut for typing Avagraha symbol On a bus with an asylum seeker Nicaragua in 5 days When to commit Generated code to version control Why I always buy a local SIM in a foreign country Use Makefile for Android Four days in Guadalajara, Mexico Android Navigation: Up vs Back Hotels vs Airbnb vs Hostels Currency issues in Argentina Abstractions should be deep not wide Some data on podcasting Always support compressed response in an API service A day in El Calafate - Patagonia, Argentina Hermetic docker images with Hugging Face machine learning models American Elections The sound of "ch" API services should always have usage Limits Hiking in El Chaltén - trekking capital of Argentina
Book summary: How to create a mind by Ray Kurzweil
Ashish Bhatia · 2014-05-09 · via ashishb.net

The book is an insightful journey into the contemporary understanding of the human brain and how scientists are trying to replicate it. Major takeaways from the book are listed below.

Thought experiments in the world

  1. Charles Lyell was the first person to propose that steady movement of water carves out gorges and canyons.
  2. This became the inspiration for Charles Darwin’s theory of evolution.
  3. Both of them engaged in thought experiments looking for how things around them attained their states and discovered underlying phenomena.
  4. Similarly, Einstein, after reading about the experiments which concluded that the relative speed of light is always constant, engaged in thought experiments that eventually lead to the “Theory of relativity”.
  5. The human brain is remarkably amazing in its ability to identify such patterns and discover underlying phenomena just by thinking.

Thought experiments on thinking

  1. Experiment 1: It is easy to recite the alphabet going from A to Z, but difficult to do the reverse. The same experiment can be done for things like rhymes and poems. The conclusion is that Our memories are stored in sequential order. They can be accessed only in that sequential order.
  2. Experiment 2: Try visualizing a person/situation which was encountered only once or twice, it is very difficult to visualize the details => Our memories are stored as a sequence of patterns, there are no images, videos, or sound recordings. Memories that are not accessed dim over time.
  3. We can recognize even partial patterns with alterations , our recognition ability detects patterns that survive real-world variations.
  4. Our conscious experience of our perceptions is changed by our interpretations , we continuously predict the future and hypothesize what we will experience and this influences the actual perception .
  5. Routine procedures are stored as organized hierarchies in the human brain, the same ability is used for recognizing objects and situations.

A model of the neocortex

  1. The Neocortex is responsible for hierarchical patterns of information and hierarchical thinking. Some scientists call it the “new brain” as opposed to the reptilian “old brain” which is more tuned towards short-term things like getting food, avoiding becoming someone’s food and having sex.
  2. Humans have a really large forehead and large neocortex compared to other mammals. It has roughly 300 million pattern recognizers.
  3. A human master in a particular field knows about 100, 000 chunks of knowledge.
  4. Pattern recognition theory of mind - Neocortex has multiple layers of neurons, based on sensory inputs, some of the first layer neurons are triggered, and those that trigger beyond a certain threshold, fire the second layer of neurons, the second layer matches higher-level patterns than the first layer, the higher layer can send a back signal and reduce or increase the triggering threshold of the previous layers. This is the prediction and hypothesis part.
  5. Memory is effectively a list of patterns. They will trigger with appropriate input leading to the recall of that memory. Since we capture patterns, our memory is only an approximation of events that happened in the past.
  6. Learning - it’s difficult to learn too many conceptual levels simultaneously. It is difficult to train multiple layers simultaneously. We learn a level and as it stabilizes, we move to the next one.
  7. Misunderstanding - A person tries to convey patterns in his neocortex to another person’s neocortex using “language” which itself is a set of patterns in the neocortex, these differences in patterns cause misunderstanding.
  8. Directed thinking - where we consciously try to direct our thoughts toward understanding or solving a certain problem
  9. Undirected thinking - where we experience sudden recollection of memories, triggers of which appear to be non-logical.
  10. Confabulation - We subconsciously make up stories to justify our actions that cannot be explained logically. This is more pronounced in split-brain patients where the left and right hemispheres are not connected - leading to one part of the brain reacting and the other one confabulating to justify the action.
  11. Culture, society, and profession predict certain norms, and this trains our neocortex to think in certain ways, this ensures social order but at the same time makes it difficult to think differently, which Darwin or Einstein did! In dreams, these norms are usually a bit relaxed.

The biological neocortex

  1. The human brain is unusually large, its a cause of a higher maternal mortality rate among humans compared to other mammals and requires a pivoting gait which makes women biomechanically less efficient walkers than men.
  2. The neocortex is a repetitive structure consisting of an “assembly of neurons” where each assembly is ~100 neurons and connections inside each assembly are similar. The connections between assemblies are dynamic, new ones are formed as needed and old unneeded ones are pruned away.
  3. The neocortex is highly “plastic” - meaning if a part which deals with say, vision is damaged, slowly another part develops the lost set of patterns.
  4. A genome has about 25 million bytes after lossless compression while total connections in the neocortex are ~ 10^15 => most connections are not determined genetically but built over time.
  5. Over time, the field of AI (Artificial Intelligence) has developed the same set of techniques, that are believed to exist in the neocortex, to process real-world information like human speech and written language.

The old brain

  1. Optic nerves have 12 output channels - one recognizes edges, another recognizes large areas of uniform color, and another focuses on background => effectively we only see patterns rather than exact details.
  2. The human cochlea catches sound vibrations and extracts about 3000 bands of information.
  3. The Thalamus process information coming from various parts of the body, including eyes and ears) before handing them over to the neocortex.
  4. Neocortex on its own cannot do directed thinking, it requires inputs coming from Thalamus.
  5. The Hippocampus is the area where short-term pattern forming happens, these patterns are transferred to the neocortex for the long term. Alzheimer’s disease attacks the Hippocampus first.
  6. The cerebellum controls quick motions like catching a ball.
  7. The old premammalian brain is addicted to pleasures like food and sex. Neocortex in mammals allows us to control primitive desires. Dopamine and Serotonin play a role in the feeling of pleasure.
  8. Amygdala controls “fear”, that is flight or fight decisions - on detecting danger, it causes a sudden rise in blood pressure, heart rate, and respiration rate.
  9. Feeling happens in both the old and new brain while thinking happens only in the new brain (“neocortex”).

Transcendent abilities

  1. Emotional thoughts take place in spindle neurons - humans have a lot more of them than other animals. Newborns don’t have them but they are developed over the age of 4 months to 3 years. This is when a child learns emotions and morality.
  2. The ability of the neocortex to master signals of fear from the Amygdala plays a role in confidence, organizational skills, and the ability to influence others.
  3. Creativity - A key aspect of creativity is finding metaphors which the neocortex is good at. Learning new patterns from different fields helps in learning more metaphors, making the brain more creative.
  4. Love - Phenylethylamine (PEA) causes the feeling of “love” - high energy level, focused attention, and craving to be with someone. Oxytocin encourages long-term bonding, that is, monogamous relations. Prairie vole is monogamous because of oxytocin receptors while montane vole engages in short-term relations because of lack of them.
  5. Love in humans exists primarily to satisfy the need of the neocortex as primitive lust was sufficient for reproduction. A loved one becomes a major part of one’s neocortex and after spending decades together, a virtual other exists which can anticipate every move of the loved one. When we lose the person, we are still left with patterns in the neocortex which trigger, except triggers change from delight to mourning.

The biologically inspired digital neocortex

  1. Neocortex replaces the process of slow evolution with fast learning and that’s one reason for the advancement of the human race. Since even a single human finds something new, everyone can learn that without relying on it appearing in the genetic code.
  2. A digital neocortex will provide a factor of a thousand to a million speed up over the biological cortex.
  3. Several attempts to do brain simulations at both the functional level as well as the molecular level are being done. By 2020, there will be sufficient computational power to simulate the human brain.
  4. Techniques being used are neural networks, vector quantization (for sparse coding), HHMM (hierarchical hidden Markov models), and evolutionary genetic algorithms.
  5. Watson, Google Translate, and Wolfram are some major works done in the direction of processing and generating real-world information.

The mind as a computer

  1. The brain is slow but massively parallel.
  2. In the human brain, 300 million pattern recognizers can be fired together.
  3. The computers are fast but do not have sufficient parallelism.

Thought experiments on the mind

  1. Consciousness is a heated topic of debate among philosophers. There are no agreements on “Are plants conscious?” or “Are babies conscious?”. The general agreement is that the “ability to act according to one’s free will” (implying nondeterminism) is consciousness.
  2. Panprotopsychism says that everything is conscious, “humans are more conscious than a light bulb”.
  3. The author predicts that by 2029, we can expect to see “digital consciousness”.
  4. Western perspective - consciousness is an emergent property of the complex system.
  5. Eastern perspective - consciousness is a fundamental property, the physical world only comes into existence through the thoughts of conscious beings.
  6. The existence of “free will” is a hotly debated topic - the existence of free will requires non-determinism in actions and experts don’t agree on whether humans have free will, compared to say Watson, or are the actions predetermined. Some philosophers believe that the future is completely deterministic but complex enough to be unpredictable.
  7. There are also disagreements on what constitutes identity - the author holds the position that a snapshot of the neocortex (or brain, in general) is a person’s identity, and hence, identity, in principle, is not unique and can be replicated.

The law of accelerating returns applied to the brain

  1. Most technologies follow the S-curve, that is, slow start, rapid growth, and then maturity. But at the end of its life cycle, it’s replaced by something else, eg. Transistors replaced vacuum tubes.
  2. Most technologies like processing power, storage capacity, brain imaging resolution, etc. are growing at an exponential pace and leading to a “Law of accelerating returns”. Hence, a digital brain will arrive faster than most people expect.

Objections

  1. This chapter talks about objections raised to the previous work of the author - not much relevant to the rest of the book.