Intelligence that reflects accumulated learning is referred to as

Fluid vs. Crystallized Intelligence: Examples and More

General intelligence consists of two types: fluid and crystallized.

You use both types of intelligences. If you’ve ever encountered a new problem and solved it by recalling a previous experience, your fluid and crystallized intelligences have teamed up to your advantage.

Fluid intelligence is your ability to learn, assess, and navigate new situations. Crystallized intelligence is accumulated knowledge you can recall as needed. Problem-solving uses both intelligences.

Although they’re different, fluid and crystallized intelligence are both equally important — and in some ways connected.

Fluid intelligence, or general fluid intelligence (Gf), is your ability to:

  • reason
  • recognize patterns
  • solve problems
  • adapt to your environment and the world

You use Gf when you encounter new information and learn.

Gf can sometimes predict a person’s academic and job success. When educators or employers conduct aptitude testing, they’re measuring fluid intelligence.

It’s Gf that allows young people to learn faster and more easily than adults. An analysis from 2018 that included 680 studies found a connection between fluid intelligence and reading and math development.

There was a stronger relation to math and complex skills over reading and foundation skills. Yet, math uses both fluid and crystallized intelligence.

What is an example of fluid intelligence?

When you think of a strategy to solve a complicated math problem, you’re using fluid intelligence.

At what age does fluid intelligence peak?

Fluid intelligence increases until adolescence, then gradually declines.

General crystallized intelligence (Gc), or crystallized intelligence, is information you’ve learned in the past and stored. Gc results from experience, education, and cultural background.

School testing of learned facts and job candidate testing of software knowledge are examples of crystallized knowledge assessments.

What is an example of crystallized intelligence?

When you solve a math problem, your existing knowledge of formulas and symbols is an example of crystallized intelligence.

At what age does crystallized intelligence peak?

Crystallized intelligence increases as you age and gain more life experience. It slows in adulthood but can either stabilize or continue to increase throughout most of your life.

Fluid intelligenceCrystallized intelligence
new informationstored information
short-term memorylong-term memory
increases through childhood, peaks at adolescence, then declinesincreases through childhood, and slows with aging, then stabilizes or continues to increase throughout life
functions include working memory, processing speed, reasoning, cognitive control, inhibition, complex skills, attention tasks, creativityfunctions include procedural (practical), declarative (factual), general and specialized knowledge, wisdom

Although fluid and crystallized intelligences are different, they complement each other.

For example, an emergency room doctor uses both. When a patient arrives, the doctor’s fluid intelligence assesses the situation. Much like when solving a puzzle, the doctor’s fluid intelligence uses clues to arrive at an answer — in this case, a diagnosis.

Then, the doctor’s crystallized intelligence provides the previously acquired procedural knowledge needed for care and treatment.

Fluid and crystallized intelligences are also connected in some ways.

For example, high fluid intelligence allows you to learn faster, which results in more accumulated knowledge and in turn higher crystallized intelligence, as one 2019 study explores.

Changes in fluid intelligence can affect the expression of crystallized intelligence.

If you’re an older adult who has experienced a decline in fluid intelligence, you may not be as skilled at applying crystallized intelligence from your long-term memory to some current or new situations.

Crystallized intelligence testing measures general knowledge. Test types include vocabulary, verbal, and language use assessments. Research from 2015 points to text completion examinations as one measure of crystallized intelligence.

Fluid intelligence includes multiple cognitive qualities, so it requires a more varied testing approach. Assessment options include:

  • Progressive matrices: to test for pattern recognition, as stated in research from 2013
  • Digital span sequencing: to assess working memory, as explained in a 2019 study
  • Trail making test: to measure cognitive flexibility, as described in 2008 research
  • Digital fluency test: to assess creativity, problem solving, and inhibition, as noted in a 2016 study
  • Tower test: to measure planning ability, as explained in another 2016 study
  • Cognitive failures questionnaire: to assess attention, memory, distractibility, and executive function, as suggested in a 2017 study

You can increase both types of intelligences.

To increase your crystallized intelligence, you simply need to acquire new knowledge. Education and experience can help it take root in your long-term memory.

According to 2016 research, improving fluid intelligence requires a more comprehensive approach. It helps to think of fluid intelligence as a reflection of the health and fitness of your brain. So to increase this type of intelligence, you’ll need to nourish it well and exercise it often.

Brain health strategies include:

  • improved diet
  • regular exercise
  • adequate sleep
  • mindfulness training
  • stress reduction

Brain fitness activities include:

  • learning a new language
  • playing strategy games like chess
  • trying cognitive training
  • changing your routine
  • learning new skills and hobbies
  • socializing more
  • learning a musical instrument
  • challenging yourself mentally
  • using your mind to remember things rather than apps and smart devices

General intelligence has two main categories: fluid and crystallized.

Fluid intelligence is your ability to process new information, learn, and solve problems. Crystallized intelligence is your stored knowledge, accumulated over the years.

The two types work together and are equally important. They both increase through childhood. Fluid intelligence decreases with age and crystallized intelligence remains stable or continues to increase with age.

You can increase both types of intelligence. You could improve your fluid intelligence by keeping your brain healthy and fit. Adopting a lifelong learning habit can help you build up more crystallized intelligence.

Crystalized versus Fluid Intelligence | Lifespan Development

Intelligence is influenced by heredity, culture, social contexts, personal choices, and certainly age. One distinction in specific intelligences noted in adulthood, is between fluid intelligence, which refers to the capacity to learn new ways of solving problems and performing activities quickly and abstractly, and crystallized intelligence, which refers to the accumulated knowledge of the world we have acquired throughout our lives (Salthouse, 2004). These intelligences are distinct, and crystallized intelligence increases with age, while fluid intelligence tends to decrease with age (Horn, Donaldson, & Engstrom, 1981; Salthouse, 2004).

Figure 8.16

Adapted from Horn, Donaldson and Engstrom (1981)

Research demonstrates that older adults have more crystallized intelligence as reflected in semantic knowledge, vocabulary, and language. As a result, adults generally outperform younger people on measures of history, geography, and even on crossword puzzles, where this information is useful (Salthouse, 2004). It is this superior knowledge, combined with a slower and more complete processing style, along with a more sophisticated understanding of the workings of the world around them, that gives older adults the advantage of “wisdom” over the advantages of fluid intelligence which favor the young (Baltes, Staudinger, & Lindenberger, 1999; Scheibe, Kunzmann, & Baltes, 2009).

The differential changes in crystallized versus fluid intelligence help explain why older adults do not necessarily show poorer performance on tasks that also require experience (i. e., crystallized intelligence), although they show poorer memory overall. A young chess player may think more quickly, for instance, but a more experienced chess player has more knowledge to draw on.

Seattle Longitudinal Study: The Seattle Longitudinal Study has tracked the cognitive abilities of adults since 1956. Every seven years the current participants are evaluated and new individuals are also added. Approximately 6000 people have participated thus far, and 26 people from the original group are still in the study today. Current results demonstrate that middle-aged adults perform better on four out of six cognitive tasks than those same individuals did when they were young adults. Verbal memory, spatial skills, inductive reasoning (generalizing from particular examples), and vocabulary increase with age until one’s 70s (Schaie, 2005; Willis & Shaie, 1999). However, numerical computation and perceptual speed decline in middle and late adulthood (see Figure 8. 17).

Figure 8.17 Seattle Longitudinal Study ages 25 to 88

Cognitive skills in the aging brain have been studied extensively in pilots, and similar to the Seattle Longitudinal Study results, older pilots show declines in processing speed and memory capacity, but their overall performance seems to remain intact. According to Phillips (2011) researchers tested pilots age 40 to 69 as they performed on flight simulators. Older pilots took longer to learn to use the simulators, but performed better than younger pilots at avoiding collisions.

Flow is the mental state of being completely present and fully absorbed in a task (Csikszentmihalyi, 1990). When in a state of flow, the individual is able to block outside distractions and the mind is fully open to producing. Additionally, the person is achieving great joy or intellectual satisfaction from the activity and accomplishing a goal. Further, when in a state of flow, the individual is not concerned with extrinsic rewards. Csikszentmihalyi (1996) used his theory of flow to research how some people exhibit high levels of creativity as he believed that a state of flow is an important factor to creativity (Kaufman & Gregoire, 2016). Other characteristics of creative people identified by Csikszentmihalyi (1996) include curiosity and drive, a value for intellectual endeavors, and an ability to lose our sense of self and feel a part of something greater. In addition, he believed that the tortured creative person was a myth and that creative people were very happy with their lives. According to Nakamura and Csikszentmihalyi (2002) people describe flow as the height of enjoyment. The more they experience it, the more they judge their lives to be gratifying. The qualities that allow for flow are well-developed in middle adulthood.

Tacit knowledge is knowledge that is pragmatic or practical and learned through experience rather than explicitly taught, and it also increases with age (Hedlund, Antonakis, & Sternberg, 2002). Tacit knowledge might be thought of as “know-how” or “professional instinct.” It is referred to as tacit because it cannot be codified or written down. It does not involve academic knowledge, rather it involves being able to use skills and to problem-solve in practical ways. Tacit knowledge can be understood in the workplace and used by blue collar workers, such as carpenters, chefs, and hair dressers.

What is artificial intelligence (AI)? – Amazon Web Services

Artificial intelligence (AI) is a branch of computer science that deals with solving cognitive tasks normally reserved for human intelligence, such as learning, problem solving, and pattern recognition. Artificial intelligence (AI) is often associated with robotics and science fiction, but in fact it has long gone beyond science fiction films. Today, artificial intelligence is part of advanced computer technology. One of the prominent scientists in this field is Prof. Pedro Domingos. He identified five groups of scientists who contribute to the development of machine learning: symbolists, whose field originates in logic and philosophy; connectionists who take knowledge from neuroscience; evolutionists developing the methods of evolutionary biology; Bayesians who apply mathematical statistics and probability theory; and analogues whose research is based on psychology. In recent years, advances in statistical computing have led to the further development of artificial intelligence in a number of areas, which are collectively referred to as "machine learning". Similarly, progress in the field of neural networks has led to the development of an additional field called "deep learning". Machine learning and deep learning are two fields of computer technology that have been pioneered by the study of artificial intelligence.

In a broad sense, these approaches are divided into "guided" and "voluntary" learning. In the first case, data with a given result is used, and in the second, without it.

Companies produce data for machine learning and deep learning every day, and with more data, AI is getting smarter and evolving faster and faster. Data is pulled from data stores such as Amazon Redshift, collected using crowdsourcing platforms such as Mechanical Turk, or loaded dynamically using Kinesis Streams. In addition, with the development of the Internet of Things and sensor technologies, data that was hardly used before is now available for analysis, and its volume is growing exponentially.

Machine learning is commonly referred to as analysis methods based on Bayesian theory that are used for pattern recognition and learning. At the heart of machine learning is a set of algorithms that use the provided data to train and predict, optimize a utility function under uncertainty, recognize hidden structures in the data, and classify the data into a concise description. Machine learning is often used in cases where the use of precise software algorithms does not provide sufficient flexibility or is inefficient. Ordinary computer code processes the input data according to the algorithm laid down by the developer and returns the appropriate response. A machine learning system parses the input to look for patterns and generates an aggregation code (machine learning model) that returns the “correct result” based on previous inputs (as well as outputs in the case of guided learning). The accuracy of a machine learning model largely depends on the quality and quantity of data accumulated over time.

Using qualitative data, the model can analyze multivariate problems with billions of possible options and find the optimal function that, given the input data, will predict the correct value. As a rule, machine learning models predict the response with statistical certainty and are quite reliable. Such estimates should be considered when deciding whether to use machine learning models or any single prediction.

How do we use machine learning at Amazon? actively uses machine learning systems to solve practical problems. Machine learning technologies help to expand the scope of activities, improve the operation of services, improve the quality of logistics and speed of delivery. launched the AWS platform to provide other companies with the flexibility and cost-effectiveness to leverage these benefits in their IT infrastructure, and continues to make machine learning technology available to others.

The very structure of Amazon. com's development teams and commitment to solving utilitarian business problems with machine learning help build simple yet powerful machine learning tools and services. These tools are first tested on critical processes within before being made available to other companies, just like other IT services.

Using Machine Learning for Business

Machine learning is often used to predict outcomes based on historical data. For example, companies can use machine learning to predict future fiscal quarter sales based on demographic information, or estimate which customers are most likely to become disillusioned with, or most loyal to, a particular brand based on a customer profile. Such forecasts allow you to make better decisions, improve product quality and reduce customer retention costs. Machine learning complements business intelligence systems, which provide information about a company's past performance but focus on forecasts of future trends.

Successful implementation of machine learning technologies in a company consists of several steps. First of all, it is necessary to determine what problem the system should solve, i.e. what forecasts can be useful for the company. Then you need to collect historical data based on business indicators (transactions, sales figures, customer losses, etc.). This data will be used to build a machine learning model. The machine learning model will then make predictions that can be used to make more informed business decisions.

Using Machine Learning in Your Organization

Anomaly Detection

Identify objects, events, and data that differ from expected results or stand out strongly in their category of objects.

Fraud detection

Build predictive models to detect potentially fraudulent transactions and deceptive or incorrect descriptions.

Customer churn

Identify customers who are most likely to refuse services. This allows you to interest such customers in advance through promotions or personalized service.

Content customization

Use predictive analytics models to recommend products and display customized content on your site based on previous customer activity.

Deep learning is a special case of machine learning that uses multi-level algorithms to analyze data more deeply. In such non-linear algorithms, not just an explainable set of relationships, as in simple regression analysis, but distributed representations of data that interact with each other in depending on a certain set of factors. Given a large amount of input data, deep learning algorithms can determine the relationships between elements. For example, these relationships can be between shape, color, words, etc. These relationships are then used for prediction. In the case of artificial intelligence, the effectiveness of deep learning algorithms lies in that they can determine many more relationships between objects than a human can put into a program, and also find relationships that humans may not be able to detect at all. Given large enough inputs, a network of algorithms will make predictions or interpret very complex data structures.

Image and video segmentation and classification

Convolutional neural networks perform better than humans in many tasks, including object classification tasks. After examining several million images of known content, such a system of algorithms will be able to determine the subject of new images. Many photo storage services use facial recognition algorithms based on the use of deep learning capabilities. They formed the basis of Amazon Rekognition, Amazon Prime Photos, and Amazon's Firefly Service.

Speech recognition

Amazon Alexa and other virtual assistants are able to recognize a request and return a response. Humans are known to be able to recognize voice from a very early age, but computers have learned to recognize voice and respond to it relatively recently. The task of speech recognition is enough it is difficult to solve using only traditional mathematical and computational algorithms, since they do not allow to cover the entire set of accents and speech patterns. With the use of deep learning, algorithm systems cope with this task more efficiently and are able to recognize the content and intent of speech.

Natural language comprehension

Natural language processing systems are designed to understand human language, text sentiment and context. This allows the algorithm to recognize more complex features such as emotion or sarcasm. seek to automate customer service using text or voice bots, such as those used by Amazon Lex.

Recommendation services

The pages of online stores often display content customized for each user: products, films or news that may be of interest to him. Previously, these materials were handled by company employees, creating links between products. However, with the advent of big data and deep learning, this work can be done instead of people by programs in which appropriate algorithms can determine products that are potentially interesting to the user based on data on previous purchases and viewed products and comparing this information with information about other users.

>> Learn more about MXnet, an open source deep learning framework, and how to get started.

AWS support for Internet Explorer ends on 07/31/2022. Supported browsers: Chrome, Firefox, Edge and Safari. Read more »

The battle of artificial intelligences

When the word "artificial intelligence" is used, many people think of intelligent humanoid robots that strive to conquer the world. However, it is not intended to replace humans. Its purpose is to expand human capabilities, it is a process, the ability to analyze data. Simply put, artificial intelligence (AI) is a system or machine that can mimic human behavior in order to perform tasks and gradually learn from the information it gathers.

AI has many implementations, for example:

  • chatbots use AI to analyze customer requests faster and provide appropriate responses;
  • "smart assistants" use AI to extract information from large datasets in free form and optimize planning;
  • Recommender systems automatically select similar programs for viewers based on previously watched ones.

The topic of artificial intelligence is surrounded by marketing noise and is replete with vague terms: neural networks, machine learning, deep learning...

Artificial intelligence (Artificial Intelligence, AI) is a technology by which a computer system can simulate the actions of a person in a similar situation. The date of the emergence of AI is considered to be 1956, when Marvin Minsky, Claude Shannon, John McCarthy and Nathaniel Rochester announced this direction at the Dartmouth seminar. Usually, the scope of AI includes systems that perform human functions - speech and image recognition, search for solutions and relationships, learning.

The main difference between AI and other computer systems is the lack of predetermination: AI can produce different responses to the same input. It is almost impossible to understand how the AI ​​came to a conclusion.

There are many ways to implement AI, for example based on multiple rules or expert systems. Since the 1980s, the most widespread technology has been machine learning (Machine Learning, ML). Machine learning refers to any technique that allows systems to "learn" without explicitly programming that behavior.

The meaning of machine learning is that a computer system receives a large set of data as input and information about which of them contain the desired characteristics. Such data can be, for example, information about financial transactions, as well as which of them are fraudulent.

The system processes data according to a special algorithm, accumulating them. Then all decisions are made in accordance with the accumulated “knowledge” - ML systems, like the human brain, create many interconnections in the process of their learning.

The unusual way the AI ​​works not only makes them vulnerable to attacks, but also means they can fail in unexpected ways.

To improve the efficiency of machine learning, the arrays of accumulated data and their processing began to be carried out at several levels, forming entire networks of conditional "neurons" - layers that mimic the human brain. This technology is called deep learning - Deep Learning, DL. Specific implementations of deep learning systems are commonly referred to as neural networks .

The power of artificial intelligence is growing, and it's easy to forget that its protection is not perfect. However, scientists have found that other algorithms can take advantage of the vulnerability of AI. As a result, an attacker can affect the operation of an autonomous driving system, a text mining program, or computer vision.

In September 2019, the US National Institute of Standards and Technology published its first-ever warning about an AI attack on a commercial algorithm. Security analysts determined that the target was Proofpoint, a program that uses machine learning to detect spam emails.

While tech buffs are talking about Tesla's autopilot and watching Siri talking to "Alice," cybercriminals are actively working to introduce AI into fraudulent schemes. Consider some of them :

  • Deepfakes (Deepfakes) - the use of neural networks to create fake audio and video content that is almost impossible to distinguish from the genuine. This technology is becoming a powerful weapon in modern information wars, as a result of which a person can no longer rely on what he sees or hears.
  • Cracking captcha (CAPTCHA, Completely Automated Public Turing test to tell Computers and Humans Apart, public Turing test) - a small computer test, introducing letters and numbers into a special field that you see in a distorted picture - which is easy for a person to pass , but difficult or impossible for a computer.
  • Bypass spam filters and antivirus software . The introduction of AI has become a new stage in the confrontation between cybercriminals and developers of security solutions. Based on the successful experience of implementing neural networks for malware recognition, cybercriminals began to use deep learning to develop ways to hide from detection. Back in 2015, spam filters were bypassed using a neural network that used a generative grammar to create email messages that are almost indistinguishable from those written by real people.
  • Password Guessing . The principle of password guessing tools like HashCat and John the Ripper is that they generate the next password according to given rules, calculate its hash, and then compare it with the hash of the real password. In a large number of cases, a dictionary of frequently used passwords is used to generate passwords, on the basis of which various options are formed. To increase the efficiency of this process, the use of generative adversarial networks (GANs) trained on a large data set from password leaks makes it possible. Such a network will generate variations corresponding to the statistical distribution, which can significantly speed up the process.
  • Breaking into systems . The use of AI allows attackers to improve traditional hacking methods by creating unpredictable ways to carry out cyberattacks. One of the tools that implement this strategy was the DeepHack tool presented at DEFCON 2017, implemented on the basis of a neural network. DeepHack allows you to conduct network penetration testing (pentest) without the need to collect preliminary information about the system. The tool uses a neural network that independently creates commands and allows you to fully automate the process of hacking databases on websites.

Forecasts and recommendations

The high effectiveness of the use of AI to commit crimes has caused an increase in the popularity of these tools in the cybercriminal environment. Gartner predicts that by 2022, 30% of AI cyberattacks will use AI algorithms.

  • Attempts to restrict the spread of neural networks or regulate their use at the state level seem to be ineffective and may rather lead to the opposite effect.
  • It looks more promising as much as possible wide information and public discussion of the problem , which will create a more critical perception and reduce the level of influence of fakes.
  • It is extremely important to develop counterfeit detection technologies and improve methods for detecting the use of artificial intelligence to automatically block fraudulent resources, fake photo, audio and video materials.

How technology helps businesses secure AI programs.

As AI proliferates, so do new opportunities to exploit its vulnerabilities. This has led to the emergence of companies that look for weaknesses in algorithms in order to notice malicious activities before they cause damage.

Yaron Singer - Harvard professor, co-founder and CEO of Robust Intelligence: "If you develop machine learning models, then you have no way to simulate a real cyber attack or conduct a penetration test." Singer says: “In many applications, even small changes can lead to very different results. But the problem is even deeper: the crux of the matter is how machine learning is done.”

The goal of Singer's research is to change the input data of a machine learning system in such a way that it behaves incorrectly and develop systems that will be secure. AI cheating is based on learning from examples and capturing subtle changes that humans are not capable of.

In addition to the verification system, Singer demonstrates a way to outsmart the Internet fraud detection system as part of the search for weaknesses.

Learn more