My Mother Asked Me to Explain the Fuss Around AI

Google, ChatGPT, Bing AI, How its connected, why its now everywhere and what is all about?

Can’t say for sure why she asked me about that. Me and my mom living in different countries for quite a while, I'm sitting in Berlin, Germany and she is in Kyiv, Ukraine. And suffering from the “Incoming” alarm almost everyday. Probably it is a way for her to go away from that rude, unfair and ugly reality. Well, let me help her and do simplified explanations.

It is interesting how life turned tables, decades ago she taught me about the world, and now its my time it seems.

Luckily, I’ve studied AI, and since then its my job and my hobby…

Here is the disclaimer: In this article Im gonna do a dramatic simplifications and I will provide personal opinions which is sometimes based on personal perception, so my dear nerds, do not be mad at me… at least for this time.

Artificial Intelligence. What is that?

If it is Artificial Intelligence — most likely its written in Microsoft Power Point, if its Machine Learning — most likely its written in Python. — The Internet

The joke above is losing its power nowadays but still there is a “seed of truth”.

Before any try to explain what artificial intelligence is, it looks like to me good point is to focus on problem of definition of what exactly is the Intelligence? I don’t know, and it seems nobody really does. What humanity has is some sort of assumptions and some general explanations thats all. Otherwise If humanity would exactly know what intelligence is — then creation of it is not a big deal.

But most of us agreed that Artificial Intelligence (AI) is a branch of computer science that focuses on the creation of intelligent machines capable of performing tasks that would typically require human intelligence. Or in short: the machine that will somehow “understand” what to do. And the “understanding” here is also a big problem.

But lets go from wide definition to the more strict challenges being addressed by AI systems (at least the ones which I can remember/aware now):

• Natural Language Processing: Understanding, generating, and interpreting human language.
• Computer Vision: Object recognition, image segmentation, and facial recognition.
• Robotics: Navigation, manipulation, and human-robot interaction.
• Healthcare: Disease prediction, medical imaging analysis, and drug discovery.
• Autonomous Vehicles: Path planning, obstacle avoidance, and decision-making.
• Recommender Systems: Personalizing content, products, or services based on user preferences.
• Finance: Fraud detection, algorithmic trading, and risk assessment.
• Education: Adaptive learning systems and automated grading.
• Gaming: Developing intelligent agents that can play complex games like Go and Poker.
• Cybersecurity: Intrusion detection, malware analysis, and network security.

As you may see from the list above, we doesn’t have some “one to rule them all” system but different set of tasks.

And of course, there’s a lot of talk now about General Artificial Intelligence or GAI/AGI, which aims to be like humans or better. But there’s no known success yet. If try to be funny, I could compare its like “sex topic” between teenagers: noone had it but everyone is talking about.

Ok, so if humanity trying to build artificial intelligence, how they measure their success in that?

That is really good question, there are bunch of methodologies which tend to be the marker of a quality of AI. But lets focus on classic one — the Turing Test.

The Turing Test

The Turing Test is a way to measure whether a machine’s intelligence is indistinguishable from a human’s. The name comes from Alan Turing, a mathematician and computer scientist. He proposed a test in the 1950s, and called this method “Imitation Game”.

Think of it like a game show. You’re the judge. Behind one curtain is a person, and behind another curtain is a computer. You can ask them both any question or have any conversation with them. If, by the end of the game, you can’t reliably tell which one is the machine, then the computer has passed the Turing Test.

So does any AI or chat-bot system which is called so have passed that test?

Well, actually some of them — yes. And first of them was far from creation of LLM like ChatGPT — Eugene Goostman.

…a computer algorithm claiming to be a 13-year-old boy called Eugene Goostman passed the Turing test. — The Guardian [1].

The chat-bot was designed to emulate a 13-year-old Ukrainian boy and was said to have convinced 33% of the judges that it was human. However, this claim has been subject to controversy and criticism.

And here we could make a conclusion, if some AI system comes to the level when its result in complex tasks is undistinguishable from human or even better, that could be potential mark of success.

Okay, so you wanna say, that humanity somewhere created a machine or algorithm that understands things like we do?

The answer is: No, but sometimes yes. Depends on point of view. Let me introduce you another experiment.

The Chinese Room Argument

Chinese Room Argument was first published in a 1980 article by American philosopher John Searle. I will try explain it in 3 points:

1. Imagine a room with an English speaker (im gonna call him the “operator”) who doesn’t know any Chinese. Inside the room are large books with instructions in English. Outside the room, native Chinese speakers send Chinese characters (questions) into the room through a some sort of a window.
2. Using the books, the operator can look up the input characters, follow the instructions, and send back the appropriate Chinese characters (answers) without understanding the meaning.
3. To an external observer, it appears as if the room understands Chinese because it can produce correct answers. However, the operator inside doesn’t understand Chinese at all, he’s merely following the instructions.

Searle’s argument is that, similar to the operator, a computer running a program might appear to understand a language or exhibit intelligence. However, it doesn’t genuinely “understand” in the same way humans do. It’s merely processing symbols.

Of course that was a decades ago, nowadays a lot had changed in the field of AI, but still that experiment is one of the my most favorite ones.

But what about ChatGPT and AI inside Skype and I think Google had one?

Yes, this is something completely new and in 1980s it would sound as sci-fi novels.

Large Language Models, like ChatGPT, are a type of AI system designed to understand and generate human-like text based on the patterns they’ve learned from insanely high amounts of data.

If try to make complex stuff more simple, those systems, after bieng trained and fine-tuned, designed to interact with text input from user. You do a question — LLM considers the sequence of words and applies patterns it had learned before for generation of contextually relevant response to the user. I don’t go deeper with transformer architecture and attention mechanisms. I do believe its more or less clear now.

Google Search and the AI competition

Imagine a person who have outerworld lust for some delicious Ukrainian Borsch. So person hop onto Google, type in “How to cook Borsch,” and bam! There are million links. Where to start?

Now, let’s switch gears. You ask the same question to a Language Model like ChatGPT or Microsoft’s Bing. Instead of getting a flurry of links, you get the actual recipe, well… at least it looks like so, but about that a bit later.

Here’s how it breaks down:

• Google Search: A list of Links! Click, click, click until you find what you’re looking for.
• LLMs: Here’s your recipe, friend! No need to surf through countless pages. The answer is served up hot and ready, just like Borsch.

No wonder about the news from Google and all fuss there the after ChatGPT’s “supernova”.

AI systems, in their current implementation, can be applied in a multitude of ways. While some of us see infinite opportunities, others fear the implications. Debates about AI development and its regulation are everywhere. From my perspective, countries are in a race for the AI supremacy.

References

1. The Guardian. (2014, June 9). Eugene the ‘person’ takes Turing Test prize. Retrieved from https://www.theguardian.com/technology/2014/jun/09/eugene-person-human-computer-robot-chat-turing-test.