Or: can machines make art?
To be able to answer this question we first need to agree on a simple model which is equally applicable to a human and an artificial intelligence. Only then we can compare the two. I will spend most of your reading time explaining this model and the difference between man and machine.
Leading A.I. experts believe that AGI (and the subsequent ASI) will be developed on the basis of reinforcement learning (RL). In RL, we typically are dealing with a system composed of three components: the agent, the environment and a reward function. The agent observes the environment, takes actions and gets rewarded (or punished by negative reward) depending on the action it takes. The agent learns to maximize the cumulative reward over time.
For machines, the goal is typically relatively simple (winning a game, walking from A to B etc.). Accordingly the reward function is rather simple too (e.g. give a reward when the game is won, give a reward proportional to the distance walked towards B etc.).
How can we apply this model to humans? For us there is no external instance giving us rewards. Even if we get promoted by our superior the reward happens inside our mind. We get rewarded by our own brain with good (in this case proud) feelings for doing a good job. As we all know, we can get good feelings in a thousand different events or experiences. Often the reward for an event even depends in a very complex way on the circumstances of the event. For instance we experience sex to be much better when we are in love with a partner.
So why is the human reward function so enormously complex? As we will soon see the reason is that it is rooted in an almost infinite history and looking forward into infinity!
The goal of life cannot be described with a single event. From a biological perspective it is not enough to simply reproduce. We unconsciously want to create successful offspring which again produces successful offspring (and so on into infinity). Unconsciously means that the strategy to achieve this goal is „hard coded“ in our reward function and cannot be analyzed by our introspective capabilities. A simple description of a single event which will lead to this success does not exist. Accordingly myriads of intermediate goals are rewarded instead. They have been developed by evolution over millions of generations over a timespan of 3.5 billion years. The range goes from very basic things like thirst (resp. „enjoying to drink“) to the most complex ones like social interactions (e.g. friendship) and love. Following all these rewards in the right moment and state of the environment has proven to increase our success regarding the biological goal stated at the beginning of this paragraph.
If you want to know more about this topic: a more detailed discussion of the human reward function can be found in my 2018 blog post „Why the creation of human-like A.I. might require godlike powers“.
This history of millions of generations is what machines will not have for a long long (long!) time. Therefore (and most surprisingly!) our true value lies in our wishes, desires, pleasures, fears and our pain (and not our rather low intelligence). We are very much attached to the idea that our value comes from what we can do. Instead in the future our value will come from what we want!
This allows us to answer the second question posed at the beginning of this article: intelligent machines can be a very powerful tool to create art, but they cannot be artists. The artist is the one telling the machine what do draw as you need to be able to want a thousand complex things. A machine always wants the same simple thing it was told to want. Therefore it would, without receiving wishes from a human, always draw the same picture (an electric outlet or an oil gun? Sometimes it is hard to resist making a stupid joke).
This is admittedly a very radical definition of the artist, as it does not encompass any skills. The German aphorism „Kunst kommt von Können“ (art comes from ability) loses its validity (it has already suffered a lot in the past few decades!). But wait: will future artists only talk to intelligent machines about their ideas? No! These most powerful tools will always only be one of many ways to create art. The use of limitation is a fundamental principle for the creation of art. So there will be always people drawing, sculpting, singing and painting by themselves. Simply because this techniques limit the way we can express ourselves and lead to interesting results. But it is possible, that the results of these efforts will be regularly amplified and further explored using AI tools. It is also very important that we continue to produce art, as A.I. can learn only on existing data (= our past!) and will therefore always lag behind our cultural development. To keep it up to speed we have to feed the A.I. continuously with fresh art.
This brings us to the last question (posed in my last blog post): is it desirable for an individual to win over all others in the game of competition? Is it fun to become the singleton only entertained by A.I.? The answer is a clear no: to enjoy our future we need a billion human artists. Each equipped with unique cultural background and experiences which lead to a million wishes, fears, desires etc.. For the reason given above they cannot be replaced by A.I.. If there were only A.I.'s left to entertain us (or the only one left of us) we would stagnate forever. It would mean cultural death. In a future cooperative society powered by advanced A.I., people are not a burden anymore but a most valuable resource. This is the most compelling reason why building a cooperative society is in everybody's interest.
 |
 |
Animated GIFs „Machine Learning“ by GUM
Image: Shutterstock / Oksana Nazarova
Follow me on X to get informed about new content on this blog.
