Four FAQs about Artificial Intelligence

Artificial intelligence, or AI — few topics are receiving as much attention in technology and political circles, and the news is full of extreme claims. Some argue that AI will immediately revolutionize the world economy, while others see AI as an existential threat straight out of a science fiction movie.

Despite the media attention, the average person doesn’t understand what AI is or how it works. This article will explain the basics by examining four frequently asked questions (or FAQs) in layman’s terms. It won’t argue for or against AI, instead laying out the facts and allowing readers to decide what to make of it. The first step is defining precisely what AI is.

1. What Is Artificial Intelligence (AI)?
Artificial intelligence is a broad term describing a machine’s ability to learn, reason, generalize, and infer meaning. The underlying technology is called LLM, an acronym for large language model. Contrary to what some believe, AI doesn’t allow machines to “think” in the human sense of the term. Instead, the machines use algorithms, pattern recognition, and complex computing processes called neural networks to facilitate “deep learning” and make decisions as a human would.

This is a vital distinction. Human intelligence allows people to contemplate any problem, while AI tools are programmed to tackle very specific problems and cannot do anything outside their prescribed niche. For instance, Stockfish is an AI-powered chess bot that outperforms any human player by a wide margin. It can solve chess puzzles, play chess, and evaluate the state of any chess board. It cannot play other games, generate text, or offer solutions to geopolitical crises. That flexibility requires human intelligence.

Truly anonymous AI is called general AI and would theoretically be capable of deciding to eradicate humanity and rule the world itself. Modern technology is nowhere near general AI, however, and experts disagree on how such a system would even work. Therefore, the doomsday scenarios aren’t happening anytime soon.

2. What Can AI Do?
The applications of AI are many and varied. For example, a fundamental AI process called machine learning is responsible for the recommendations you see on sites such as Netflix and YouTube. AI determines the behavior of NPCs (or non-player characters) in video games. One of the most popular forms of AI is generative or gen-AI, AI that produces text, images, art, or music in response to a prompt.

The most well-known example of gen-AI is ChatGPT by OpenAI. Users type a prompt into a box and allow the AI to generate a text-based response. You can ask it to compose a story, generate an essay online, or answer a question. It’s a powerful tool companies are using to streamline employee training, expedite customer service, and even automate job roles that no longer require a human touch. Naturally, business leaders are excited about the efficiency gen-AI offers their businesses.

Of course, individuals can benefit from gen-AI too. Content creators can use AI to get a jumpstart on their projects by automating SEO keywords and refine them with editing tools such as Grammarly GO, while doctors can utilize AI to quickly and efficiently read medical scans. Many people also find gen-AI tools fun to play around with, generating absurd images like babies floating in lava or outrageous stories about celebrities.

3. What Are the Limitations of Current AI Technology?
AI technology isn’t foolproof, and every gen-AI tool on the market is prone to errors. These are called “hallucinations” in the AI world, and the industry actively downplays how prevalent they are. Some vendors don’t mention them, allowing users to assume that every response is always 100% accurate. Others acknowledge the possibility of mistakes by hyping marketing slogans such as “98% accurate!”.

A two-percent error rate doesn’t sound that bad, but the practical effects of each error can be catastrophic. One of the most famous examples occurred in June 2023, when lawyer Steven Schwartz filed a legal briefing citing two cases, Varghese vs. China Southern Airlines and Shaboon vs. Egypt Air, that didn’t exist. Both cases were hallucinations invented by ChatGPT, which Schwartz used for research purposes. He was fined $5,000 and saw his case dismissed for submitting non-existent evidence. While the other side was able to act as a fact-checker in Schwartz’s case, other hallucinations frequently go undiscovered and contribute to the spread of misinformation.

Gen-AI is trained on internet search results, which become outdated as time goes on. ChatGPT most recently crawled the internet in April 2023, leaving it unprepared to answer questions about current events that have happened since. Likewise, AI bias is a recurrent problem since information found on the internet isn’t always reliable.

4. How Much Does AI Cost?
ChatGPT is free for end users, so many people assume the technology is free. It isn’t. Gen-AI is powered by computers demanding constant water and air conditioning, so companies like OpenAI rack up large energy bills. The machinery also has to be housed somewhere, to say nothing of the people responsible for updating the software.

Companies like OpenAI monetize gen-AI using “tokens,” with each token generally costing a fraction of a cent as of January 2024. This wouldn’t be too bad if each query was one token, but that’s not how it works. Instead, the number of characters in a query determines how many tokens it is. For example, the word “hamburger” is three tokens: “ham,” “bur,” and “ger.” It’s easy for a single query to cost dozens of tokens, which becomes a significant revenue stream when thousands of people submit questions.

Unsurprisingly, more powerful AI tools charge more per token. The base version of ChatGPT is free, a fact that made the name ChatGPT virtually synonymous with gen-AI. However, ChatGPT for Enterprise, ChatGPT Turbo, and GPT4 are not free and offer more power than the base version. Similarly, competitors such as Google’s Gemini often outperform OpenAI’s software at comparable price points. When employers offer gen-AI to their employees, they frequently choose the inferior free version or limit how many tokens can be included in queries to save money without anyone knowing the difference.
The Final Verdict

Artificial intelligence is a deep topic, but the information above should suffice as an introductory primer. The next time you see a report on AI, you’ll be ready!