No uprising of machines on the horizon?

Could artificial intelligence spin out of control?

Artificial intelligence is being integrated into the legal framework in an increasingly systematic way. A comprehensive law of this kind has already been adopted in Italy, and many other countries are following suit. Active work is underway in this area in Belarus as well. Some of our strategic and programmatic acts state that the use and development of AI technologies in various fields is in the national interest of our country and is, overall, one of the priorities of innovative development. So, in the very near future, we too will have to regulate the use of neural networks by law.

Word on subjectivity 

First of all, let’s define the basic terms of this article: no artificial intelligence exists in nature, nor will it appear in the foreseeable future. It is just a fancy name dreamed up by people with a penchant for sensationalism. 

In our objective reality, there is machine learning and neural networks, and all of this has absolutely nothing to do with intelligence. A neural network is simply a highly specialised calculator for solving intellectual tasks. 

Models are designed and trained to obtain specific results, but they cannot think. This is not because evil people prevent machines from doing it independently, fearing their uprising — we simply do not know how to do it. A neural network cannot be declared the author of the content it generates, simply because a neural network does not possess a goal-setting skill or self-awareness; it is incapable of deciding to do something independently. It is merely a set of mathematical algorithms, a tool that possesses no subjectivity. Importantly, it is a tool that can be copied and reproduced, meaning it is not unique. 

Ode to greed

To explain the core conflict between neural networks and copyright, it is necessary to briefly explain how they are trained. To put it simply, neural networks designed for image generation are trained by being given pictures with labels indicating what they show: person, cat, dog, tree, and so on. During training, the neural network takes each image, breaks it down into ‘noise’, similar to scattered confetti, and then tries to reassemble everything, comparing the result with the original. When the first neural networks were trained just a few years ago, no one really gave much thought to copyright, and many were not even sure if anything interesting would come of it. Yet, the impact was so great that it shook the entire planet.
It was precisely then that artists and photographers began filing lawsuits against the companies that created the first popular neural networks. The motivation was simple: my works are publicly available and could have been used to train your neural networks, which means you owe me money for copyright infringement. Later, actors also started worrying about the use of their images and voices.
The creators of ‘machine intelligence’ drew appropriate conclusions from all of this, and now new neural networks do not know the style of specific contemporary artists. However, if you need the style of a specific master, no one is stopping you from going to a specialised resource, downloading a pre-trained model with the skills of this creator and using it at your discretion, or training such a model yourself. After all, we are not talking about billions of images here, and a few dozen are easy to process at home.

Responsibility

So, we have arrived at the issue that concerns people far more than the authorship of the generated images — responsibility. When it comes to authorship, any user proudly declares that they used their own intellect, will, creative vision, knowledge and so on to achieve the result, and therefore they are the author. However, when it comes to responsibility, people usually start shouting that it was not them, the neural network did it all by itself, and they only showed it to others. Let’s break this down…
Now that AI technologies are becoming increasingly accessible and widespread, it is crucial to understand one essential thing: if you do something illegal using a neural network — no matter what it is — you will be punished.

There is a Criminal Code, and it contains perfectly clear and unambiguous formulations. Therefore, if you have done something that falls under these stipulations, you will be subject to punishment regardless of the tools used in the process.

If you have come up with something that causes doubts from a legal point of view, only a flick of your finger separates the result from returning to oblivion. After all, no one forces you to save it, let alone show it to anyone. If you do, you bear full responsibility for your actions, and the ease of obtaining a particular result will not mitigate the severity of responsibility for its use.

Let’s summarise

Over the past few years, neural networks have become so ingrained in our lives that governments have already begun to regulate them through national legislation. So far, the general consensus is that it is simply a tool, while the author is the one who used that tool to realise their creative vision.
Questions surrounding the training of neural networks provoke considerably more debate, but even here, the global community largely agrees regarding materials in the public domain. Of course, different countries have their own specific nuances, yet the trajectory of national legislation is the same for everyone. The emerging neural network race and attempts at unfair competition by individual nations aiming to take the lead are, perhaps, of greater concern.
Thus, the United States is trying to restrict the supply of certain equipment for training neural networks to China, and even high-end consumer graphics cards are only allowed to be supplied with reduced memory. This is simply because China is already overtaking the USA in this race.
As for Belarus, we certainly cannot stand aside from all these events. So we, too, are adapting our national legislation to the new realities, naturally, taking into account our national interests.

It may be necessary to address the legal regulation and licencing of data centres in advance. Energy is already becoming an extremely sought-after resource, required for the functioning of the infrastructure for training neural networks — and Belarus boasts an entire nuclear power plant, with another one coming soon. So, the development of the industry is only a matter of time.

WHERE DO YOU GET DATASETS?

Neural networks do not just create pictures and videos. They can also be used in engineering, medicine and even the military. The question of where to get data to train them is a pressing one for developers. You can simply download pictures from the internet, but where do you get data on cancer patients to create a new software product that will be able to detect cancer in its early stages?
Large arrays of structured information have suddenly become a very valuable commodity. Therefore, we will most likely have to legally regulate the issues of datasets that can be used in the training of neural networks, simply to avoid abuse in this area. At the very least, it is probably worth requiring developers to confirm the legality of the facts used to train specialised neural networks.

RESPONSIBILITY STILL APPLIES

It is now possible to copy both appearance and voice, place a person in any circumstances, and even make a video with them. But if you use these tools for fraud or libel, you will bear responsibility regardless of the effort expended.

By Yuri Terekh