What Is Artificial Intelligence & How Does It Work?
I am deeply fascinated by artificial intelligence (AI), not only by the technology itself but the implications it can and will have on society. Whilst it’s becoming more common and people have an ‘idea’ of what it is, many people don’t fully know what it is or how it works.
When I’m interested in something I like to simplify it down so that I can understand it, and in the process I can share those learnings with you. I’ve found a lot of ‘AI explained’ resources to be written by tech-gurus, and are quite hard to understand for the everyday Joe like you and me.
So if you’ve been wondering what the heck AI is, how it works, do we need to be afraid of killer robots, here is some information that will help you, in a blog that I’ve made as easy-to-understand as possible (hopefully). Enjoy!
Note: Whilst I’ve put it into my own words to make it easier to understand, this is mostly a combination of notes from various sources.
Table Of Contents
What Is AI?
AI isn’t one thing, it’s a mixture of many different technologies working together so machines can comprehend, act, and learn with human-like intelligence. Or as John McCarthy puts it, “the science and engineering of creating intelligent machines that can achieve human-like goals.”
You already encounter AI every day — it ranks the websites you see on Google search, blocks spam from your inbox, suggest what Netflix shows to watch, what clothes you might like, sorts your social media feeds and more. That’s just the tip of the iceberg! There are three ‘stages’ to AI advancement and evolution:
1. Narrow Ai
AI that performs a single task or set of closely related tasks. It does one thing really well, but can’t do lots of things at once. Think Siri, weather apps and Netflix recommendations.
2. GeneraL AI
More like what you see in movies, where machines can perform task as good as (or better than) humans. This is when AI has human-level intelligence and can think strategically, abstractly and creatively, with the ability to handle a range of complex tasks. While machines can perform some tasks better than humans (e.g. data processing), general AI does not yet exist. We’re still decades away from that tipping point—known as the ‘singularity’—assuming it happens at all.
3. Super AI (or the ‘singularity’)
The singularity is a hypothetical point in time at which technological growth becomes uncontrollable and irreversible, resulting in unforeseeable changes to human civilisation. Basically, the AI becomes so advanced that we can’t control it or even understand it, and we have no idea what it’s going to do. This is the main thing people are concerned about, but we’re a long way off it.
Benefits Of AI
AI has the potential to transform businesses as well as the relationship between people and technology at large. You think the internet had a big impact on the world? Wait until AI kicks in! Here are a few benefits that AI provides:
Efficiency: AI eliminates friction, improves analytics, automates processes, predicts maintenance needs and lowers costs.
Automation: Automation cuts costs and brings new levels of consistency, speed and scalability, executing tasks without ever needing to rest. It can tackle mundane activities so employees can spend time on more fulfilling high-value tasks and boost labor productivity.
Accuracy: AI removes human error and makes specialised skills more available. E.g AI can now be used to find cancer on MRIs with the same accuracy as highly trained radiologists. Its ability to self learn and self optimise means it continually gets better over time.
Research: AI can uncover gaps and opportunities in the market more quickly, helping businesses introduce new products, services, channels and business models with a level of speed and quality that wasn’t possible before.
Customer service: From 24/7 chatbots to faster help desk routing, AI can curate information and provide high-touch personalised experiences that drive growth, retention and satisfaction.
Intelligence: AI adds intelligence to existing products. In most cases, AI will not be sold as an individual application. Rather, products you already use will be improved with AI capabilities, much like Siri was added as a feature to a new generation of Apple products.
A quick demonstration…
Here’s a story so you can see the power and speed of AI. There is a game called Go, a Chinese strategy game that is over 2500 years old and still played today. It’s kind of like chess on steroids, with way more variables (which is why what follows is so impressive).
AlphaGo, an AI created by Google’s Deepmind was taught how to play the game. In 9 months it went from unable to beat just a reasonable Go player, to then beating the European world champion (ranked #600), then beating the world champion, then beating everyone while playing simultaneously.
It gets crazier… then they created AlphaZero, which crushed AlphaGo (the one that just beat all the world champions simultaneously), 100 to 0. One hundred! And all it did was learn by playing itself! The exponential rate of improvement is wild… here’s how it works.
Key Terms (How Does It Work?)
Algorithm
A set of instructions that tells a computer what to do. It is a well-defined procedure with an input and an output that allows a computer to solve a problem.
Machine Learning (ML)
A key subset of AI. It’s a set of algorithms that can change themselves, find patterns, draw conclusions and ‘learn’ from experience, without human intervention. An ML system can do its own thing without its creators needing to program it. There are a few subsets:
Supervised Learning — You feed the algorithm loads of label training data (e.g an image of a cat with the label ‘cat’), and supervised system learns to associate input (e.g image) with the correct output (e.g ‘cat’ or ‘not cat’).
Unsupervised Learning — You give the algorithm unstructured, unlabelled data and it makes conclusions on its own. A common use case is clustering, where input data is divided into groups or patterns and rated on similarity.
Reinforcement Learning — A machine learning model that interprets the environment, takes action and learns through trial an error. For example if the AI is playing a game, if it makes the right move it is rewarded. If it makes the wrong move, it is punished. It uses this feedback to learn by itself and get better and better over time.
Neural Networks (or ‘Deep Learning’)
A set of artificial algorithms that work like a human brain, designed to recognise patterns. Whilst machine learning (above) almost always requires structured data, neural nets rely on layers of artificial neural networks. As data moves through different layers, it extracts different information and finds patterns. This is similar to how our human brain works to solve problems – running queries through different concepts and related questions to find the answer (or possible answers).
APplications of machine LEARNING & algorithms
By combining algorithms with machine/deep learning together, AI can accomplish complex tasks that are becoming very common today. Some examples include:
Computer Vision — How computers ‘see’, enabling them to identify and process objects in images and videos. Whilst humans easily see a picture or an object and know what it is, AI requires a lot of training and has to analyze thousands of pixels and figuring out what it’s looking at, let alone the context.
LiDar — Another form of computer vision is LiDar, which acts as an eye of the self-driving vehicles. It provides them a 360-degree view of the surrounding helping them to drive themselves safely.
Speech Recognition — How computers ‘hear’. It enables computers to recognise and translate spoken language into text.
Natural Language Processing (NLP) — How computers analyse, understand and manipulate human language as it is spoken, and to take action based on spoken instructions (like Siri).
Natural Language Generation (NLG) — How computers ‘write’. The ability to turn structured data into understandable written text, similar to that of a human being but at a faster pace of thousands of pages per second.
Generative Adversarial Networks (GANs)
GANs pit networks against each other. One network serves up content (typically videos or images), which the second evaluates as real or fake. They work together to generate fake audiovisual outputs that look realistic: artwork, video and deepfakes.
Data Science
The science of analysing and systematically extracting or dealing with data sets that are too large or complex to be dealt with by traditional data-processing software.
Turing Test
A test to ascertain whether a computer has the ability to “think” like a human.
AI Ethics & Concerns
AI is moving at a blistering pace and, as with any powerful technology, organisations need to build trust with the public and be accountable to their customers and employees. “Responsible AI” is the practice of designing, building and deploying AI in a manner that empowers employees and businesses and fairly impacts customers and society—allowing companies to engender trust and scale AI with confidence.
Data security
Data privacy and the unauthorised use of AI can be detrimental. Companies must design confidentiality, transparency and security into their AI programs at the outset and make sure data is collected, used, managed and stored safely and responsibly.
Transparency and explainability
As AI technologies become increasingly responsible for making decisions, businesses need to be able to see how AI systems arrive at a given outcome, taking these decisions out of the “black box.” Clear governance and ethics can help with the development of practices and protocols that ensure their code of ethics is properly translated into the development of AI solutions.
Control
Machines don’t have minds of their own, but they do make mistakes. Organisations should have risk frameworks and contingency plans in place in the event of a problem. Be clear about who is accountable for the decisions made by AI systems.
Bias
AI isn’t inherently unbiased. In the U.S., the AI community skews white and male. This affects how AI systems are built and designed, as well as what training data they are fed. Data can often be fundamentally biased itself. When bias creeps into algorithms, it can reinforce and even accelerate existing inequalities—especially in regard to race and gender.
Job Loss and Wealth Inequality
One of the primary concerns people have with AI is future loss of jobs. By the year 2030, about 800 million people could lose their jobs to AI-driven robots. Some would argue that if their jobs are taken by robots, perhaps they are too menial for humans and that AI can be responsible for creating better jobs that take advantage of unique human ability involving higher cognitive functions, analysis and synthesis.
Industries AI Could Reshape
Manufacturing
Automation and optimising the manufacturing process by using robots to improve quality and lower costs. It could also displace up to 20 million jobs globally by 2030.
Logistics
Using machine learning to forecast transit times, demand levels, and shipment delays. AI-guided systems also work inside fulfilment warehouses.
Transport
Self-driving vehicles, sidewalk delivery bots, delivery drones and more.
Healthcare
Quickly interpreting charts and medical data, accelerating drug discovery, enhancing chemistry, altering genetics and even conducting surgical operations.
eCommerce
Driving online sales and increasing conversion rates and recommending products. The Alexa voice assistant/salesbot has 10,000 employees working on it.
Advertising
Finding and segmenting audiences, continuously testing variations, and adjusting spend (by scooping up a lot of personal data which is not usually well-received by users).
Cybersecurity
Identifying phishing attacks and malware, trawling through codebases to look for bugs and more. A cruel irony that both attackers and defenders can use the technology, so with more AI cybersecurity, there may also be more social engineering attacks.
Financial Services
Fraud detection, personalised banking services, credit scoring, risk management, automation back-office tasks (like form-filling and claims processing), robo-investing and enhancing high-frequency trading with algorithms.
Insurance
Using machine learning for underwriting, lending, and creditworthiness assessments. Ping An uses AI to expedite claim estimates. It’s loaned over $72 billion—and shrunk average approval times from five days to two hours using AI.
surveillance
Computer vision for facial recognition, license plate recognition, and crowd monitoring. It’s obviously a massively controversial yet rapidly growing industry.
Smart Homes
It’s not an exaggeration when people say that nearly every part of the home is being connected to the internet, from the toaster to the doorbell. Smart speakers, thermostats, and security devices all rely on some degree of AI to collect and analyze data.
Data Labeling
A new industry that’s been kickstarted by the rise of AI. We produce a lot of dirty data. In order for computer vision or other AI systems to make use of that data, it needs to be cleaned and labeled.
Human Resources
Tools like Bravely allow companies to run daily surveys to gauge employees’ moods. Bravely provides on-demand, confidential coaching sessions. When an employee books a session, Bravely uses an algorithm to ingest data—identity, role description, support preferences, experience, and urgency—and then pairs them with a coach.
AI Examples & Emerging Tech
Here are some really cool companies and examples of AI in action today.
Neuralink
This will blow your mind (enter the cyborg). They surgically insert a small chip into your brain, and attach artificial neurons to the real neurons in your brain. It can receive information from the brain but also send information to the brain, and can be used to cure all sorts of brain disorders, from alzheimers and deafness, to allowing paralysed people to walk again (seriously).
Tony Robbins Interviews Sophia
I love Tony Robbins — but this likely isn’t the motivational speech or interview you’re used to! Sophia is a social humanoid robot developed by Hong Kong-based company Hanson Robotics and is the first robot to receive citizenship of any country. Tony Robbins interviews her, and it’s pretty interesting to watch! Check out the video here.
World’s First AI City
Terminus unveiled plans for an “AI City” called Cloud Valley in Chongqing, China with carbon-neutral living, low-energy-consuming, complete 5G infrastructure, and a robot-friendly park.
Flippy
Flippy is the world's first autonomous robotic kitchen assistant that can learn from its surroundings and acquire new skills over time. It can flip burgers, use a fryer, clean, switch tools and it cooks perfectly and consistently every time.
Amazon One
When you hover your hand over the device, Amazon One uses computer vision to analyze the ridges, lines, vein patterns, and more in real time. Then it sends an encrypted image to the cloud to create a unique palm signature.
Graze
Graze is a fully autonomous lawn mower that is 100% electric and solar powered.
Federated Learning
Federated Learning Could Be the Next Big Thing for Data Privacy. Federated learning allows your iPhone to wake to your “Hey Siri” but not your friend’s.
Super Maps
Live View, an augmented reality (AR) walking tool that provides direction arrows and distance markers. Google is expanding Live View to include information about landmarks in 24 prominent global cities. Google Maps recently added an AR social location-sharing feature for Pixel users to see how lost their friends are on the way to a meet-up.
Recommended Books & Resources
If you’re interested in AI, there are a few resources that are worth checking out!
YouTube Series
If you don’t have the interest or patience to read in-depth books, check out the YouTube series The Age of AI produced by Robert Downey Jr. It’s super interesting and goes into lots of different examples of AI, from artificial limbs to reducing waste, automating pizza and much more.
Books
The Fourth Industrial Revolution by Klaus Schwab
AI Superpowers by Kai-Fu Lee
Life 3.0 by Max Tegmark
Artificial Intelligence by Melanie Mitchell (quite technical/dense)
Podcasts
The Lex Fridman Podcast
AI in Business Podcast
Data Skeptic
A Timeline History Of AI
Just in case you’re interested, it’s pretty cool to see how AI advances over time! Check out this timeline:
1936
Alan Turing describes a hypothetical universal computing machine.
1943
Warren McCulloch and Walter Pitts publish the first paper about logic units, describing the concept of an artificial neuron (the basis of neural nets).
1948
Manchester Mark 1, the first program computer with storage, is invented.
1950
Alan Turing asks “Can machines think?” and develops The Turing Test, which gauges whether a computer can pass as a human.
1956
Logic Theorist, the first AI program, is invented. The Dartmouth workshop (founding event of AI) takes place and the term “artificial intelligence” is coined.
1966
MIT AI Lab creates ELIZA, the first chatbot.
1972
AI Winter sets in. Optimism and progress fades, as logic-based expert systems can’t do what’s been promised. Funding dries up for roughly two decades.
1979
Kunihiko Fukushima creates an artificial multilayered neural network.
1982
Japanese government announces Fifth Generation, a decade-long plan to develop AI. This triggers FOMO, helping thaw the AI winter in the U.S.
1986
Carnegie Mellon introduces Navlab, a semi-autonomous car.
1994
AI program Chinook is declared Man-Machine World Champion in checkers.
1997
IBM’s Deep Blue supercomputer beats world chessmaster Garry Kasparov.
2004
DARPA holds a self-driving vehicle contest in the Mojave Desert.
2011
IBM’s Watson beats Ken Jennings in Jeopardy!
2012
University of Toronto team submits AlexNet, a deep learning algorithm, to the ImageNet AI competition. The model outperforms competitors and becomes the basis for a deep learning revolution.
2014
Google spends hundreds of millions of dollars to acquire advanced AI research lab DeepMind.
2017
Google’s AlphaZero requires just four hours to become the world’s best chess player.
2020
OpenAI launches GPT-3, the largest language model ever trained, in closed beta. GPT-3 has 175 billion parameters.