Artificial intelligence (AI) is currently the trendiest buzzword, with nearly every major company incorporating AI features into its products or services. However, despite its widespread use, the term lacks a straightforward and clear definition. What researchers might consider a minor improvement in machine learning, the marketing department may portray as a major stride towards artificial general intelligence. This disparity in interpretation makes understanding the true scope and implications of AI a complex endeavor.
Sci-fi writer Ted Chiang’s view on artificial intelligence as “a poor choice of words in 1954” is quite telling. Having written about AI developments for the past decade, I can attest to the truth in his statement. The terms and definitions surrounding AI are so ambiguous that engaging in a meaningful discussion about AI requires first clarifying what one precisely means by it. The nebulous nature of AI-related terminology makes it challenging to have a genuine and focused conversation about this rapidly evolving field.
Let’s delve into the world of artificial intelligence and try to understand why defining it precisely is such a challenging task. We’ll explore the historical journey that led us to this complexity and examine the vast capabilities of AI. I’ll do my best to present the knowledge I’ve gathered in recent years in a simple manner, but be prepared to encounter the same frustration I’ve faced in summarizing such a complex and evolving subject.
- What is AI ?
- How AI Works?
- AI : terms and definitions
- Machine learning vs AI: What’s the difference?
- AGI vs. AI: What’s the difference?
- What AI can do?
- The pros and cons of using AI
- Future of AI
What is AI ?
Artificial intelligence refers to a machine’s ability to learn, make decisions, and take actions, even when faced with unfamiliar situations it has never encountered before.
At its most comprehensive level, artificial intelligence entails a machine possessing the capability to learn, make decisions, and take action, even in novel and unfamiliar situations.
In the context of narrow sci-fi interpretations, AI is often associated with robots and computers displaying human or super-human levels of intelligence, along with enough personality to behave as characters rather than mere plot devices. For instance, in Star Trek, Data is an AI with distinct personality traits, while the ship’s computer is akin to a more advanced version of Microsoft Clippy. It’s essential to note that contemporary AI falls far short of this sci-fi portrayal, lacking the complexity and human-like qualities depicted in fictional works.
In simple terms, a non-AI computer program follows a set of programmed instructions to perform the same task repeatedly and in the same manner each time. Picture a robot designed to create paper clips by bending a small strip of wire. It consistently produces the same three bends in the wire without variation. However, if presented with dry spaghetti, it would be incapable of doing anything other than breaking it. The robot lacks the ability to adapt to new situations autonomously; it can only perform the specific task it was programmed for and requires reprogramming to handle different scenarios.
AIs, on the other hand, possess the capability to learn and tackle more intricate and dynamic challenges, even those they haven’t encountered previously. Take the example of building a driverless car: instead of trying to program a computer to navigate every intersection on every road in the United States, companies are developing computer programs equipped with various sensors to assess their surroundings and react appropriately to real-world scenarios, regardless of whether they have encountered them before. Achieving a truly driverless car is still a significant challenge, but it’s evident that it cannot be accomplished using the same approach as conventional computer programs. It becomes impractical for programmers to account for every individual case, necessitating the creation of computer systems with adaptability to handle diverse situations.
The current state of AI is often referred to as weak AI, narrow AI, or artificial narrow intelligence (ANI). These AIs are specifically trained to perform particular tasks and lack the ability to handle everything. Nonetheless, they still showcase impressive capabilities. Apple’s Siri and Amazon’s Alexa are examples of relatively simple ANIs, but they can respond to a wide range of requests.
Given the current popularity of AI, the term is likely to be used extensively, sometimes in situations where it may not truly apply. It’s essential to approach claims of AI with a critical eye and conduct thorough research to verify whether it genuinely involves AI or is just a set of predefined rules. This brings us to the next important consideration.
How AI Works?
Currently, the majority of AIs rely on machine learning, a process that involves developing complex algorithms enabling them to act intelligently. While other areas of AI research like robotics, computer vision, and natural language processing play vital roles in practical implementations, machine learning remains the foundation.
In machine learning, a computer program is provided with a substantial training data set—larger data sets yield better results. For instance, to train a computer to recognize different animals, a data set with thousands of animal photographs paired with text labels describing them is used. Through processing this training data set, the computer program creates an algorithm—a set of rules—for identifying the various creatures. Unlike traditional programming, where humans define criteria, the computer program creates its own.
This highlights the significance of having existing data, such as customer queries, to train AI models effectively for businesses.
While the specifics of machine learning get more complex, both GPT-3 and GPT-4 (Generative Pre-trained Transformer 3/4) and Stable Diffusion were developed using structured training with machine learning. GPT-3 was trained on a vast data set of 500 billion tokens from books, news articles, and websites, while Stable Diffusion used the LAOIN-5B dataset, containing 5.85 billion text-image pairs.
From these training datasets, both GPT models and Stable Diffusion developed neural networks—complex, many-layered, weighted algorithms inspired by the human brain—that allow them to predict and generate new content based on their training data. For example, ChatGPT uses its neural network to predict the next token when answering a question, while Stable Diffusion uses its neural network to generate images matching a given text prompt.
Technically, these neural networks are “deep learning algorithms.” Although the terms are sometimes used interchangeably, a neural network can be relatively simple, whereas modern AIs rely on deep neural networks with millions or billions of parameters, making their operations intricate and challenging to deconstruct. This can lead to issues with biased or objectionable content, as these AIs often function as black boxes, taking inputs and providing outputs.
AI models can also be trained in alternative ways. For instance, AlphaZero learned to play chess by playing millions of games against itself, starting with only the basic rules and the win condition. As it explored various strategies, it learned and even developed new tactics that humans had not previously considered.
AI : terms and definitions
At present, AI has the capability to execute a diverse range of impressive technical tasks, often through the integration of various functions. Below are some of the key accomplishments it can achieve.
Machine learning refers to the process in which computers, or machines, extract information from the data they are trained on and subsequently generate new insights and knowledge (learn) based on that data. Initially, a vast dataset is provided to the computer, and humans train the machine in various ways. Through this training, the computer develops the ability to adapt and improve its performance based on the knowledge gained from the dataset.
Deep learning is a subset of machine learning, specifically focused on the development of highly autonomous computer systems that require less human intervention. In the context of deep learning, the vast training dataset is used to create a deep learning neural network, which is a sophisticated, multi-layered, and weighted algorithm designed to mimic the human brain’s functioning. As a result, deep learning algorithms can process information and various types of data in an exceptionally advanced and human-like manner, making them capable of complex and sophisticated tasks.
Generative AIs like GPT and DALL·E 2 have the remarkable ability to produce new content based on the data they were trained on. For instance, GPT-3 and GPT-4 underwent training on an astonishing amount of written material, encompassing the entire public internet and a vast array of books, articles, and documents. This extensive training enables them to comprehend written prompts and engage in discussions about various topics, including Shakespeare, the Oxford comma, and appropriate emojis for work Slack conversations, as they have absorbed information on these subjects from their training data.
Similarly, image generators were trained on extensive datasets containing pairs of text and images. This enables them to understand distinctions between dogs and cats, although they may still struggle with more abstract concepts like numbers and colors. Their training data is a vital component in their ability to generate coherent and relevant content based on the inputs they receive.
AIs possess a wide range of capabilities when it comes to working with words, and generating text is just a small fraction of their linguistic abilities. Natural Language Processing (NLP) empowers AIs to comprehend, classify, analyze, respond to, and even translate human communication in a natural manner.
For instance, when you ask someone to turn on the lights in a room, there are numerous ways to frame or phrase the request. Basic language understanding allows a computer to respond to specific keywords like “Alexa, lights on.” However, NLP goes beyond that and enables an AI to interpret the more intricate formulations and nuances that people commonly use in natural conversations.
NLP plays a crucial role in enabling large language models like GPT to comprehend and respond to various prompts. Moreover, NLP can be utilized for diverse AI language tasks, including sentiment analysis, text classification, machine translation, automatic filtering, and more. Its versatility makes it a fundamental aspect of AI’s capability to work with human language effectively.
Computer vision is the remarkable process through which AIs gain the ability to perceive and comprehend the physical world. This is achieved either by analyzing images and videos or directly through the input from their sensors.
While computer vision is undoubtedly crucial for the development of self-driving cars, its applications are much broader and immediate. For instance, AIs can be trained to distinguish between various common skin conditions, detect weapons, or even add descriptive text to enhance the online experience for people using screen readers. The potential of computer vision extends far beyond automotive applications and brings about transformative impacts in various fields, enriching the way we interact with technology and the world around us.
Robotic process automation
Robotic Process Automation (RPA) is an optimization technique that harnesses the power of AI, machine learning, or virtual bots to carry out routine tasks that would typically be handled by humans. For instance, a chatbot can be programmed to address common queries and guide customers to the appropriate support personnel, or it can automatically send updated invoices to suppliers at the end of each month.
While RPA resides at the intersection of traditional automation and artificial intelligence, intelligent automation (IA) takes a leap into the realm of AI. IA involves crafting workflows that not only operate automatically but also possess the ability to think, learn, and enhance themselves without human intervention. For example, IA can conduct an A/B test on your website and autonomously update the content with the best-performing version—and subsequently run another A/B test with a newly AI-generated iteration. Intelligent automation brings unprecedented efficiency and adaptability to processes, unlocking a new level of productivity and agility for businesses.
Machine learning vs AI: What’s the difference?
AI and machine learning are often used together, but they have distinct differences. To understand artificial intelligence vs. machine learning, consider that machine learning is a subset of AI.
AI is a broad term encompassing any form of thinking or reasoning carried out by machines, making it challenging to draw precise boundaries around its scope. However, one essential component of AI is machine learning, where computer programs can autonomously “extract knowledge from data and learn from it.”
Many present-day AI applications are either entirely based on machine learning or heavily rely on it during the training phase. For instance, at its recent WWDC conference, Apple chose to refer to new features as machine learning rather than artificial intelligence. This approach is more technically accurate, though it lacks the futuristic allure of AI.
In addition to machine learning, AI encompasses various subfields, such as natural language processing, robotics, computer vision, and neural networks, each contributing to the multifaceted landscape of artificial intelligence.
AGI vs. AI: What’s the difference?
When discussing AI, the common focus is on narrow AI or weak AI (unless influenced by excessive MCU viewing). Let’s explore the distinction between narrow AI and the more ambitious concept of artificial general intelligence (AGI).
Artificial Narrow Intelligence (ANI) is designed and trained to excel at a specific task, but it lacks general intelligence.
ChatGPT is a remarkable example of ANI, and it falls within the scope of AI by most definitions. However, its capabilities are constrained. While it can engage in captivating conversations, its knowledge is limited to the data it was trained on.
Unlike ANI, ChatGPT cannot be utilized to navigate an autonomous car and provide directions to a destination. Similarly, you can’t expect a self-driving car to compose poetry.
An Artificial General Intelligence (AGI), also known as strong AI, represents the ultimate goal in AI research.
AGI envisions a computer or robot with genuine intelligence capable of reasoning, communication, learning, and performing a wide range of tasks in a manner similar to humans. Unlike specialized AI, AGI is not limited to a particular domain; it would possess adaptability and versatility.
For example, an AGI could engage in literary discussions about authors like Ted Chiang and also navigate a car to transport you home. Steve Wozniak’s Coffee Test humorously illustrates the level of flexibility an AGI should possess, where it would be able to autonomously make a cup of coffee in an average American home.
However, achieving true AGI remains a significant challenge, and we are still far from creating machines with such comprehensive capabilities.
What AI can do?
AI has reached a stage where it offers practical utility to individuals working across diverse fields. As a regular human reading this article written by another regular human, here are some of the ways you can leverage AI in your everyday life:
- Virtual Assistants: AI-powered virtual assistants like Siri, Alexa, and Google Assistant can help you with tasks, answer questions, set reminders, and control smart home devices.
- Language Translation: AI-based language translation tools facilitate seamless communication with people who speak different languages, making travel and international interactions smoother.
- Recommendation Systems: AI algorithms power recommendation engines that suggest movies, music, books, and products tailored to your preferences, enhancing your entertainment and shopping experiences.
- Email Filtering: AI-powered email filters can categorize and prioritize your emails, reducing clutter and ensuring you don’t miss important messages.
- Health Diagnostics: AI assists in medical diagnostics, helping doctors analyze medical images and detect diseases more accurately.
- Personalized Content: AI is used in content platforms to recommend articles, videos, and social media content that align with your interests and preferences.
- Fraud Detection: AI plays a crucial role in financial institutions, detecting fraudulent activities and safeguarding your financial transactions.
- Smart Home Devices: AI-enabled smart home devices can automate tasks, control appliances, and optimize energy consumption, making your home more efficient and convenient.
These are just a few examples of how AI is increasingly becoming a part of our daily lives, making tasks more efficient, enjoyable, and productive.
The pros and cons of using AI
The use of AI comes with both advantages and disadvantages. Let’s explore the pros and cons:
Pros of using AI:
- Efficiency: AI can automate repetitive tasks and processes, leading to increased efficiency and productivity.
- Accuracy: AI algorithms can process vast amounts of data with precision, reducing human errors and improving accuracy.
- Personalization: AI enables personalized recommendations and experiences based on individual preferences and behaviors.
- Cost Savings: Automating tasks with AI can lead to cost savings for businesses by reducing labor and operational expenses.
- Data Analysis: AI can analyze complex data patterns and provide valuable insights that help businesses make informed decisions.
- Accessibility: AI-powered tools and applications make technology more accessible to users, regardless of their technical expertise.
- Healthcare Advancements: AI contributes to medical breakthroughs by aiding in diagnosis, drug discovery, and treatment planning.
Cons of using AI:
- Job Displacement: Automation through AI can lead to job displacement and unemployment for certain industries and job roles.
- Bias and Fairness: AI algorithms may perpetuate biases present in the data they were trained on, leading to unfair outcomes.
- Lack of Creativity: AI lacks human creativity and intuition, limiting its ability to think outside predefined patterns.
- Security Risks: AI systems can be vulnerable to cyberattacks, posing significant security and privacy risks.
- Dependence on Data: AI relies heavily on data, and without quality data, it may provide inaccurate or unreliable results.
- Ethical Concerns: AI applications raise ethical dilemmas, such as the use of AI in autonomous weapons or invasion of privacy.
- Unemployment: As AI continues to evolve, there are concerns about significant job displacement and its impact on the workforce.
Understanding the pros and cons of AI is crucial for making informed decisions about its adoption and ensuring responsible use. Striking a balance between harnessing AI’s capabilities and addressing its challenges is essential for maximizing its benefits while minimizing its drawbacks.
Future of AI
The future of AI holds immense promise and potential, with advancements and challenges lying ahead. Here are some key aspects shaping the future of AI:
- Advancements in Deep Learning: Deep learning algorithms are continuously evolving, allowing AI systems to tackle more complex tasks and achieve human-like performance in various domains.
- Autonomous Systems: AI-driven autonomous systems, such as self-driving cars and drones, are expected to become more prevalent, transforming transportation and logistics industries.
- Natural Language Processing: Improvements in natural language processing will lead to more sophisticated AI chatbots and virtual assistants, enhancing human-computer interactions.
- AI in Healthcare: AI is poised to revolutionize healthcare, aiding in early diagnosis, personalized treatments, drug discovery, and managing patient records.
- AI and Robotics: AI-powered robots will play an increasingly important role in industries like manufacturing, agriculture, and healthcare, performing tasks with precision and efficiency.
- Ethical AI: The future demands responsible AI development and deployment, addressing concerns related to biases, transparency, and accountability.
- AI and Education: AI can revolutionize education by offering personalized learning experiences, automating administrative tasks, and enabling more interactive teaching methods.
- AI and Climate Change: AI technologies can contribute to climate change research, renewable energy optimization, and smart energy management.
- AI in Finance: AI-driven algorithms will continue to impact the financial industry, enhancing fraud detection, risk assessment, and customer service.
- AI for Social Good: The future will witness AI being harnessed for social good, addressing challenges like poverty, healthcare accessibility, and disaster response.
As AI continues to advance, interdisciplinary collaboration, ethical considerations, and regulatory frameworks will be essential in shaping its future impact on society. Embracing AI’s potential while ensuring its responsible and inclusive deployment will be crucial for a positive and transformative future.