The third AI boom, which began in 2012, has been steadily progressing, and there have been recent reports of human-like output in natural language processing.
I even received a comment asking, “Should I move to a primary industry already?”, so I’d like to put the situation in perspective for a moment.
By the way, the original story is this article.
In conclusion, bloggers don’t need to think about changing jobs right now. However, I would recommend that those in the media, such as newspapers and news, change jobs as soon as possible. I don’t have anything against you… Actually, I have a lot of resentment.
In addition, although I am an IT engineer, I am not an AI engineer, and I have a simple water-cooled NVIDIA TITAX X installed in my house that can recognize the type of car I drive.
What is OpenAI?
Google’s successful recognition of cats in 2012 with a tour de force led to the start of today’s third AI boom. Nowadays, AI research is being carried out all over the world.
The article I mentioned at the beginning of this article is one of them. I learned about it a month ago in an IT industry magazine I subscribe to.
- [SiliconAngle magazine] OpenAI’s latest AI text generator GPT-3 amazes early adopters (JULY 19 2020)
At the heart of the current AI boom is an algorithm called Deep Learning. It prepares a large amount of sample data and allows the AI to learn from it. For example, a single image is divided into several thousand pieces and thousands of data are accumulated for each region to learn patterns.
As you can vaguely understand from this description, this is a single-precision, epic matrix calculation that cannot be handled by general-purpose chips like Intel, and uses NVIDIA’s GPUs. Yes, those GPUs (Graphic Processing Units) that gamers use to display high definition images.
However, even if a GPU is employed, a single GPU is not enough to provide sufficient learning. This makes it impossible to expect “inferential” results based on the learning results. This is why the general approach is to use thousands of machines for distributed learning.
Current AI technology requires “power and skill“.
I remember an experiment Google did in 2012 where they used about 2,000 machines and had it train for about a week. There’s a paper out there that you can read if you’re interested.
So, in short, AI research requires machine power, and there will be a battle for computational resources before a conference presentation or paper submission. Google is significant in this regard. It now develops and manufactures not only NVIDIA’s GPUs, but also its own TPU (Tensor Processing Unit).
On the other hand, they are also researching learning algorithms. A while ago, there was talk about how AlphaGo, which was developed by DeepMind, a company acquired by Google, defeated the World Go Championship, but recently, research on natural language processing has become a hot topic.
The GPT-3 is the result of research on OpenAI, which was proposed and organized by Elon Musk. At first, it was established with a noble mission, such as to prevent the monopolization of AI technology, and the leading AI vendors participated in it. I think Google was part of it.
(The mission is true, but the benefit is that it prevents Google and others from monopolizing the technology.)
The content of this presentation is just like OpenAI. There have been studies on having AI create sentences from scratch, but it seems to be able to make inferences from a small amount of data. I’m impressed.
Applicable scope of AI
The current AI is basically just pattern learning (ML: Machine Learning), but it has a wide range of applications. In the first place, babies are growing up using pattern learning, which is very close to real humans.
In Japan, the University of Tokyo student venture PFN and Fanuc are at the forefront of this technology. They are also working on fields such as deep reinforcement learning, with Professor Maruyama, formerly of IBM Research Labs, as a fellow. (They are famous for their demonstration of a robot arm holding a can and moving it.)
I’d like to explain the comment about primary industry at the beginning of the presentation, but nowadays, there is an investigation into the introduction of IoT using AI and 5G for primary industry. The part of the industry that currently relies on foreign workers and student part-timers is going to be left to robots (AI).
For the time being, there are some areas in the primary industry that AI cannot touch, but they are comparable to IT engineers who have been trained over several years. It also requires a high level of judgement, and I’m not impressed with the idea of tackling the primary industry simply because the service industry is tough.
(It’s not easy to succeed in a U-turn to a rural area. The same is true for tourism.)
Again, today’s AI is pattern learning. I think I remember that there was a place where they were verifying its application to taking minutes of meetings. By the way, Google predicts that AI will surpass human thinking ability by 2035.
That said, this is just a comparison between budget-conscious AI resources and ordinary humans. In terms of supply and demand, and cost performance, I think the situation will continue to be quite troubling for the foreseeable future.
Bloggers and media reporters
Now, even this article I am currently writing may be able to be created using GPT-3. After all, anyone can write the same thing, just by extracting existing information.
An article that anyone can write is, in other words, a sentence (voice) that is easy for an AI to create. And unlike humans, AI processes data accurately, so if it’s the right data, it will extract it appropriately.
This means that AI is sufficient for newspapers and news for simply compiling information. Looking at the recent series of problems with misinformation on the road, I’d rather promote the use of AI actively.
As you can see from the company’s press releases, there are a number of releases scattered throughout the country stating that the content of the media release is wrong. As long as reporters have a strange agenda, the information they provide will be distorted.
If it is left alone, Google will get rid of the bad money. In fact, Google is a commercial company, so it may end up destroying the good ones as well.
It’s not just Google, but many business models these days offer services for free and make their sales from advertising revenue. Naturally, they have a lot of basic data, so they are in a good position to apply AI technology.
It’s pattern learning, so it can be a big miss, but the know-how seems to have grown recently. I feel like I’m getting better at using the ad section at the beginning of the article.
Incidentally, the reason I can comment like this as if I were someone else is because bloggers have three advantages.
- Primary Information
- Review cost
Even though it’s not a copyright issue, my blog often employs images that I have taken myself. If someone uses these images on another blog without asking, it’s a copyright issue, and Google, or any other company for that matter, is no exception.
Bloggers are quite a hobbyist. That’s why you can send a complaint to a copy site on copyright grounds, and in some cases, you can settle it with a lawsuit or other action.
So if you provide original articles, your blog posts are automatically protected by copyright.
However, Google is aiming to digitize all the data in the world with Google Books and others. If you don’t say anything, not all of your books will be digitized without your permission.
Google may have a noble mission, but it’s essentially a commercial enterprise. They need to make sales by placing ads on the side of public information and doing business that way.
We have a group of super bright minds, and you should not be caught off guard in this regard.
As you may already know, Google is currently energetic in protecting weak media. Summary articles are hard to rank high in Google searches.
Even with newspapers and television, it’s rare to find an entity that goes to the field to get the latest information. This comes with risks and costs. This is a risk and an expense, and this is how major media use the information they get.
They may be more comprehensive and sophisticated, but if you rank them in the top ranks of search, you won’t be able to survive with locally sourced, weak media.
That’s why Google ranks primary information, even if it’s a little crude, in the top search results. Nowadays, if you don’t work hard and add value, it is becoming difficult to compete with secondary information.
That’s why information obtained at conferences or articles made up in your own way, like mine, are more likely to rank higher in Google searches.
Now, like the opening article, this time the AI creation article was the most popular. Yes, it’s true that it’s not the top Google search, but it’s still an accomplishment.
However… Will my competitors really adopt AI? First of all, there is the mighty GAFA, and you don’t know when they are going to come in. As an added bonus, the cost of adopting AI is substantial.
On top of that, you also need engineers to make the AI learn. At least I have my hands full just teaching my kids. That’s where the time and cost comes in as well.
As a further tome, there is a “screening cost”.
In this case, they posted the AI created article as it was, for the sake of experimentation. However, AI’s learning is not as accurate as that of humans, and it rarely gets away with it.
That’s why we can’t leave it all up to the AI, we need to at least check the finished blog draft. Naturally, you’ll need experts to check them.
As Google claims, by 2035, AI’s ability to think may surpass that of humans. But at least for the foreseeable future, there will be costs associated with the above three processes.
So for now, I’ll settle for the conclusion that it’s not worth the cost of having AI write articles.
(As someone who handles internal public relations, it would be really helpful to me to be AI-enabled. But it’s still going to be a struggle for a while.)
With the above, it’s going to be a while before bloggers are scared of AI. And the threat will probably be more than AI, starting with the parts of the world where labor costs are low and originality is not necessary, such as how to create a WordPress site.
However, it seems like it’s time for this to happen, but with the start of automated news distribution sites in Japan, the need for media is becoming less and less.