Bladerunner will continue to be fiction.

Note: This article is a little janky, and I am not an expert on how the current technology works. This article is more of looking into the definition of the words "Artificial Intelligence".

Tech companies and politicians are known for creating worthless buzz words and phrases that in reality mean nothing, but people take them seriously. Think "[Insert Tech Company Here] isn't a monopoly", "Windows is a bad operating system", "Windows is a good operating system", "Apple products are better", "Androids are faster", "Linux runs the internet", "We can just not use Amazon", "The Social Network is a good movie", and worst of all "Artificial Intelligence". 

I want to focus on that last one but for those curious: Tech companies are acting like they are ran by Rockefeller clones. Windows can be really good and actually be innovative, even with its trillions of problems. Apple makes good products - that are over priced, sure - but they can make crap no one wants or likes. Linux doesn't run the internet more than FreeBSD, Illumos, or Windows Millennium Edition (there is a very painful story behind this being included) do, just that Linux is the more popular option at the moment. Amazon Web Services make all online activities Amazon-usage, and Amazon is so widespread that 100% removal of them in your life isn't even possible. The Social Network is an over rated movie.

The last one, "Artificial Intelligence", is probably the worst of all. AI, or Artificial Intelligence, is the worst thing imaginable - a marketing gimmick. While by technicality, AI exists, it also doesn't. The rules for what is an isn't AI isn't clear. It could be as simple as "Turing complete", or as complicated as an arbitrary set of rules of what is and isn't an AI. The problem is, most of what is considered "AI" isn't really what it might seem.

To figure out why we consider AI to be a lie, let's look at our definition. Artificial Intelligence needs to be two things: Artificial, Intelligence. Seems easy enough? Wrong. Let's actually look into what each word means.

Artificial is easy enough in our case, as what we mean is human-made (sorry, but even I have to say rats and foxes - no matter how much I want them to - will not be building anything artificial intelligence in the near future). Intelligence on the other hand, needs some things.

There are three levels to being "Intelligent" - the baseline definition, Sternberg intelligence, and human-like - Post-human is currently technologically impossible. The baseline definition is "The ability to independently obtain and apply information." Keyword: Independently. What we mean is that if someone tells you the information directly, or tells you how to apply it, it isn't intelligence. AI doesn't meet this level - yet.

Let's look at human civilization. We think we are pretty smart and powerful, right? We aren't even a Type One civilization yet, but we are really close. As of 2019, we are a 0.73 on the Kardashev scale. AI, by the given definition, isn't fully independent, therefore not real AI. But that doesn't mean it isn't getting closer.

Let's take Generic Example Co.'s algorithm for an example. I will be borrowing heavily from the explanation given by GPG Grey in his video "How Machines Learn", but modified to the tasks of Generic Example Co., a fictional company founded just now. Generic Example Co. needs an AI that can define whether an image contains a face, a pencil, or a rat. Why they need this? No one really knows, but it might have something to do with world domination using these items. Defining the difference between a face, a pencil, or a rat might seem simple, but in reality it isn't. You cant just do a simple script like say...

case image.jpeg{
    "face": this_is_a_face();
    "pencil": this_is_a_pencil();
    "rat": omg_i_love_rats_uwu();
}

Our brains are already intelligent, but making what is essentially a complicated rock do the same things as our brains is actually a lot easier said than done. AI is usually made in three stages: Build the bot, test the bot, repeat. If you have never seen the channel Code Bullet already, you should. He does amazing videos that actually help explain AI a lot better than they realistically should watch. I recommend watching "I Created a PERFECT SNAKE A.I." (which still has the same problems of AI as explained here, but it is actually one of his many perfect examples of what this next part will mean).  You start by defining what is the end goal, then building a more simple program that does a lot of complicated wiring and definitions - this stuff is interesting but really complicated to explain in what is supposed to be a short article. This program - which is written directly by the human as it is far less complicated - builds a set of "test AI programs" to be tested. John Phantom Doe, an intern on "Codename: Faces, Pencils, and Rats" by Generic Example Co. then uses an existing sorted set of pictures of faces, pencils, and rats to a testing program. The testing program then takes the set made by the AI program to see if they can properly define these images. The first generation will do terrible as the builds are a lot more random. But once the best selection are found, they go back to be refined, while the worst preforming are scrapped. Over time, they can better tell what is a rat, what is a face, and what is a pencil.

 So AI can take the images, tell whether or not they are face, pencil, or rat, and accurately say if they are. So is it real artificial intelligence? Not quite. The problem comes from the independence. While sure, you could put an infinite amount of work and information for them to base off of, there is one major problem - do they really know what a face, a pencil, or a rat is? Give them a strange angle of a face, rotate the image, or invert the colors, and things for that AI then become a lot more difficult. It's still human dependent. The problem is when it comes to the testing. The building is bog-standard all around, a brain is a brain is a brain. AI is still mostly missing one key component of that definition. Acquiring. These programs need to be 100% self sufficient in gaining this information, and cannot rely on being told what is a bee by a human.

This doesn't mean every company branded with "AI" is scamming you, or that real AI can't happen, it is just limited to being more artificial, less intelligence.

Some further resources for people interested in how current AI works, how it can be applied, or other general tech stuff if you are interested (some of my favorite channels on YouTube, and the ones that inspire me to make Potabi, The Potabi Foundation, and everything TPF is working on):
CaryKH: https://www.youtube.com/user/carykh
CodeBullet: https://www.youtube.com/channel/UC0e3QhIYukixgh5VVpKHH9Q 
Jabrils: https://www.youtube.com/c/Jabrils
Linus Tech Tips: https://www.youtube.com/c/LinusTechTips
Techquickie: https://www.youtube.com/user/Techquickie

Comments

Popular posts from this blog

Potabi's Biggest Misconception: It isn't BSD.

We have an estimate for FreeBSD 13.1-RELEASE (and an answer to why we care)

Official Apology: The Social Media Manager Incident