Artificial intelligence, or AI, is a field that combines computer science and large data sets to accomplish tasks and solve problems that are difficult in traditional programming. AI is concerned with intelligent machines or computer systems that can accomplish the same functions as human intelligence, but with different approaches and methods. And, in some cases, accomplish things that are particularly difficult or impossible for human minds to comprehend.
In general, today’s AI models are computer programs that are trained to perform specific tasks. In the training process, there are three main components:
- The model (the AI program itself)
- The data (the training material)
- The learning algorithm (the method of training)
For example, an AI model may learn to recognize objects (like cars or people) in images—a crucial component of many modern automotive safety systems. To accomplish this goal, the AI model is “fed” a large set of images containing the objects it needs to identify (in this case, images that contain cars or people). Through an iterative process, the model gets better at recognizing the patterns and features that distinguish the relevant objects (e.g. the wheels of a car or the face of a person). Each round of training improves the model’s object recognition abilities, until it reliably detects the necessary objects.
A brief history of AI
Some of the earliest concepts about artificial intelligence date back to the 1950s. That’s when Alan Turing—famous computer scientist and mathematician—published a paper that outlined a test designed to detect “intelligence” in machines. (This would later become known as the “Turing test.”) In the mid-1950s, artificial intelligence became somewhat formalized as a field of study by researchers interested in creating intelligent machines.
The first few decades of AI research were focused on “symbolic” or “classic” AI—where AI is programmed to follow established logical rules, rather than ingesting data to come up with novel solutions on its own. Symbolic AI can, for example, tell you something like “if a patient has a sore throat, swollen lymph nodes and tonsils, and a fever, then that patient might have strep throat.” To reach such a conclusion, classic AI simply follows its programmed reasoning.
For decades, AI was a slowly developing field (thanks largely to a lack of interest and funding, limited hardware capabilities, and high costs). The rapid growth and interest in AI we see today didn’t really begin until the mid-2010s. The current period of rapid expansion—the era of “machine learning”—came about thanks to huge technological leaps: namely, more powerful computers, new AI architectures and techniques, and the availability of tons of data for AI model training.
The relevance and impact of AI today
You may not realize it, but much of today’s technology uses elements of AI, and has for quite some time. AI powers virtual assistants (e.g. Alexa and Siri), autonomous cars (e.g. Tesla vehicles), and even chatbots on websites. Not to mention that every content recommendation system of the last decade is powered by AI, too—from social media to content-streaming platforms. We’re also witnessing the rise of generative AI—models that can generate novel text, images, and other media.
AI is also proving crucial in scientific research, robotics, healthcare, and other data-heavy fields like finance and manufacturing. But these are only some of the real-world applications of AI. In time, we’ll likely see AI continue to transform industries and enhance productivity.
Different “types” of AI
Defining and measuring “intelligence” is a complex—and subjective—task. But while AI is an emerging field, and there are bound to be more and new distinctions, today it’s best to understand AI according to its two “types,” each of which represents a different degree of intelligence.
Narrow (or weak) AI
Narrow, or weak, AI basically means a system that can perform isolated tasks according to its training on specific data sets. This type of AI has a limited—or narrow—scope.
Narrow AI excels (and even outperforms humans) at well-defined tasks like image and speech recognition, or content recommendation. Once properly trained to recognize specific patterns and relationships, Narrow AI is good at detecting those things in new sets of data, and making decisions, predictions, or recommendations based on its analysis.
Common examples of narrow AI applications include things like:
- Apple’s virtual assistant, Siri, or OpenAI’s chatbot, ChatGPT
- Cars with object detection or adaptive cruise control
- Machines that can analyze medical imagery like X-rays and MRIs and diagnose health issues
- Content recommendations on YouTube, Netflix, or social media platforms
General (or strong) AI
General, or strong, AI refers to systems that are capable of human-like intelligence. It’s also known as artificial general intelligence, or AGI. For now, AGI only exists in theory, though it’s a common trope in science fiction movies. Different researchers, institutions, and organizations are currently exploring various approaches to AGI.
In order for true AGI to exist, it would need to possess the ability to understand, reason, learn, and adapt to new situations—much like a human being. It would also need to be capable of applying knowledge or skills from one area to another in order to solve unfamiliar problems. The scope of AGI is infinitely broad, as opposed to narrowly defined by a few particular tasks and data sets.
The concept of AGI introduces some very serious questions—both about AGI’s potential impact on society (e.g. job displacement, scientific breakthroughs, unpredictability, etc.), and ethical questions about consciousness, rights, responsible behavior, and more. These are just a few of the things we need to consider with the concept (and, one day, the invention) of AGI.
The current state of AI
These types, or levels, of AI help us talk about artificial intelligence in a well-defined way. All the AI you see around you today is technically narrow, or weak, AI—though some instances of weak AI are still quite impressive, like the large language models that power tools like ChatGPT. And, as far as anybody knows, that’s the only type of AI that will exist for the foreseeable future. Nobody knows when—or even if—general AI will arrive. For now, it’s a purely theoretical concern.
That said, AI’s place in modern society (with current narrow capabilities and potential future capabilities) requires us to address certain questions as a society and a species, such as:
- What biases could be inadvertently built into AI systems, and how can we proactively detect and mitigate it?
- How can we ensure AI systems align with human values? What governance models or oversight may be needed?
- How can we ensure AI respects relevant laws, rights, and regulations? What new laws or policies may be needed?
- How will AI impact employment? How can we adapt to AI in the workforce and manage job displacement?
- How can we balance AI innovation with data privacy and consent? How much transparency should AI systems have?
- How should large AI systems be controlled and accessed? How can we expand access to AI tools?
What can AI be used for?
Today’s AI has a relatively narrow scope—it’s helpful in any field where work can be intelligently automated, or where AI can produce novel insights or solutions quicker or better than humans can. In particular, AI has proven useful with detail-oriented and data-heavy tasks.
Today, that mostly looks like data analysis, predictive modeling, and decision-making processes in various industries. There are currently several fields where AI is being applied, including:
| Field | Applications |
|---|---|
| Healthcare | Imaging analysis, drug discovery, disease diagnosis |
| Finance | Fraud detection, algorithmic trading |
| Retail/e-commerce | Chatbots, personalized shopping recommendations |
| Logistics | Demand forecasting, inventory management, supply chain optimization |
| Robotics | Advances in sensors, vision systems, and control systems |
| Education | Personalized learning platforms, tutoring systems, automated grading, educational chatbots |
| Automotive | Self-driving cars, driver assistance systems |
| Media/entertainment | Content recommendation, content creation |
| Technology/cybersecurity | Search engines, chatbots like chatGPT, network security/threat detection |
This is by no means a complete list; there’s potential for AI to expand and impact every industry you can think of (just like the Internet has).
But along with these benefits, AI technology is introducing significant challenges and risks, such as bias and fairness concerns, privacy and security concerns, and a threat to proper attribution (i.e. the “oracle” problem).
Much like any other piece of revolutionary technology, AI’s ultimate impacts on society will likely depend on careful research, foresight, and responsible implementation.
