Cognitive Science and its applications to Artificial Intelligence

AI and Human Cognition

Artificial intelligence (AI) is the ability of machines to perform complex information processing in the context of human mental abilities. As humans are the only known organisms to perform complex tasks such as abstract thinking, causal reasoning, forecasting, etc., human cognition models often inspire machine information processing models.

Most recently, Google’s DeepMind used deep reinforcement learning (a common term in behavioral research) to beat an experienced human player at ‘Go’ – an extraordinarily complex Chinese board game of abstract strategy. ‘Go’ allows for more possible scenarios than the total number of atoms in the visible universe, on a 19 X 19 board. Therefore, any ‘memorization’ of moves is unhelpful, and the player must depend on identifying winning scenarios from new incoming information.

In contrast to IBM’s Watson, which first used natural language processing with ‘search and rank determination’ to outperform human memory & knowledge on Jeopardy, DeepMind used a ‘sample and optimize process’ to learn the novel environment of Go. This process is similar but more formal than experiential factors that humans use to understand new environments.

Being able to perform complex calculations gives DeepMind the ability to learn from subtle yet robust statistical patterns in its observations, which human observation would be unable to detect and therefore learn.

DeepMind’s success is a particularly encouraging step towards a general purpose AI – an intelligence that can optimize across multiple domains of information – a noteworthy quality of human intelligence. (For special purpose business facing AI, see x.ai’s Amy Ingram and Amazon’s Alexa focused on scheduling meetings, and speech recognition respectively).

“Just as antibiotics revolutionized medicine and human health by protecting us from grave biological susceptibility, artificial intelligence promises to shield us from our own undiscovered cognitive limitations.”

The necessity of AI

Haven’t they noticed what’s happening? Are they oblivious to their fate? Are they unable to work together on behalf of the environment that sustains them all? Perhaps, you think, it’s time to reassess the conjecture that there’s intelligent life on Earth – Carl Sagan, Pale Blue Dot

In 1900 the rate of deaths from infectious diseases in the U.S. was 580 per 100,000. In 2010 it was 16.2. Before antibiotics were discovered, human lives were soft targets for bacterial infections. In 1928, Alexander Fleming changed the course of history and human well-being with his discovery of Penicillin which formed the basis of antibiotics, saving hundreds of millions of lives over the decades. Today we take antibiotics for granted. It is easy to forget how exposed our species was to microbes without this discovery.

Just as we are biologically vulnerable to bacterial infections, we are cognitively vulnerable to complex information. Most of our senses are adapted to serve us in our ancestral environments of sub-Saharan Africa and not in the modern world with fast-moving cars and complicated medical diagnoses and procedures, to take two examples.

In the U.S., human error causes 90% of all road accidents leading to roughly 30,000 annual deaths. Even medical professionals – relative experts in precision – are not immune from cognitive vulnerabilities, as 1 in 10,000 patients have a surgical tool or object accidentally left in their bodies during surgery. This is in addition to the issues with misdiagnoses and inability of the smartest, best trained humans to include relevant information from voluminous new research.

As a result, avoidable human error causes significant pushback on systems designed for human flourishing.

The success of systems like Watson, DeepMind, and others provide further evidence of the necessity of AI.

Part of the reason why the arguably rudimentary AI of DeepMind could so easily beat an experienced human player is the limited capacity of humans to process complex information.

In order to improve the human condition, we need the power of artificial intelligence. AI is a tool that will allow computers to take over the parts of our lives that render us cognitively vulnerable.

For example, during the financial crash of 2007-08, one of the major precipitating factors was the loose underwriting criteria adopted by regulators and mortgage lenders who failed to assess the risk associated with their decisions. As we develop our AI tools, we may be able to prevent crashes by analyzing associated risks with the decisions of humans involved.

Other challenges facing human intelligence include the sheer scale at which data and new research on critical issues such as cancer and particle physics are produced. Without AI this data would be impossible to consume.

“The success of systems like Watson, DeepMind, and others provide further evidence of the necessity of AI.”

Future prospects

Businesses that see opportunity in developing and using these solutions will see impact and growth.

Conservative estimates predict that AI will grow from $419.7 million in 2014 to $5.05 billion by 2020, according to Markets and Markets Report. Other predictions, such as the one by Tractica, estimate AI industry revenues to reach $36.8 billion by 2025.

As we initiate AI’s transition into human governed systems, two critical and interdependent aspects of AI need to be improved:

1. AI engineering (performance)
2. AI experience (perception)

As with technologies such as computers and smartphones, AI will also likely have engineering and experience components. Fear of AI has already received media coverage and if not designed around experience, it will likely face backlash from human users once the honeymoon phase is over.

Issues such as trust, comfort, expectation, and safety will take precedence as AIs and humans form closer relationships. AIs need to adapt to human cognition in order to coexist with humans and to have maximum impact on human well-being.

Compared to our intuitive understanding a century or more ago, the world has turned out to be infinitely more complex and interesting. Our new understanding has formed the foundation of new business and enterprise in arenas like medicine, engineering, technology, and communications, as well as many others.

Just as antibiotics revolutionized medicine and human health by protecting us from grave biological susceptibility, artificial intelligence promises to shield us from our own undiscovered cognitive limitations.