Computers today already know how to draw great paintings using artificial-intelligence (AI) algorithms, after analyzing many real-human paintings. A sales house just sold one machine-generated portrait painting for $540,000, and by now there are startups that produce AI-generated portraits for $40 a piece. On the musical front, there already are algorithms that, after analyzing compositions made by Bach, compose “Bach” symphonies that even avid listeners cannot tell apart from the real thing. This brings up the question of what’s in the future for artists, now that machines create art that is indistinguishable from that produced by humans.
The same question (at a lower scale) has also been asked about security professionals. Now that machine learning algorithms can tell good from bad when looking at any type of event data, what would human security analysts be left to do? Traditionally, machines used to only sort through records using rules that humans wrote for them, but as it seems, machines are constantly getting better at writing those rules for themselves as well.
So should both worry for their jobs? It is my stance that not at all, and for surprisingly similar reasons.
What is art anyway?
The reason we consider those portrait-generating algorithms to be doing “art” is just because we tend to define the term “art” by the properties of its current common artifacts, rather than by what it really is. Once we define what art is, we see that AI merely scratches off its outermost layer, rather than truly disrupt it.
Art can be seen as the material artifact that triggers our emotional reaction of awe, amazement, astonishment, or alike, without engaging our functional, rational, logic. We feel a sense of amazement or excitement when looking at fireworks, but this sense is immediately tied to our understanding of what fireworks are and how they work, and so our sense of amazement is rationally-justified, and hence not really art. We can also see a new piece of machinery which we have no clue how it works, but our astonishment is still rational; we have no idea how it operates, but we are astonished with what it does, so this is not typical art either.
When you hear a certain piece of music, or look at a special painting, you sense those or similar emotions without associating them with rational cause-and-effect logic. It is beautiful for what it is, not for how it was made nor for what it does.
When we say that an algorithm that draws a portrait, or any other painting for that matter, generates “art", it is just because we associate art with such paintings, after seeing many such paintings that we traditionally have considered to be art. When you see the first machine-generated portrait you are astonished; probably more by the concept than by the actual creation. Once you see a hundred such, they just resemble a schoolbook, no more. When you hear the first musical piece written by a machine, you appreciate the melody as much as you appreciate the novelty of machine-generated music. Once you hear a thousand such, this melody becomes just more “machine-generated music” for you. Our human senses that react to pretty paintings and good music will saturate. Those will always be pretty to look at and pleasant to listen to, but they will no longer cause that additional Wow emotion, and hence they will stop being considered as art.
Art is in the interaction between two humans: the creator and the consumer. The artist, i.e., the creator, will obviously have to run for his money now. The artist will no longer be able to draw “yet another” painting and gain the consumers appreciation. Yet the consumer is human, and the creator is human, and it is up to the creator to revive that sense of “Wow” with the consumer. I trust the human to do just that, as it has been done ample times throughout history. At the same time, I do not envision that a machine can reinvent itself with the same level of creativity and agility when interacting with a human consumer. It can keep imitating whatever it sees to the level of indistinguishability, but as an imitator in essence, its progress uncovering new turfs in the journey to the Wow reaction will be slower. That machine will have to pass what we may consider to be “the emotional version of the Turing test“. It will iterate and try, as it does best, and it will occasionally succeed, no doubt. However, for the most part with humans, it takes one to know one.
What will the security architect do once an AI algorithm is better than he is (and certainly faster and more reliable) in determining good versus bad when analyzing events? The answer follows a similar logic. Just like the goal of the artist is to cause an irrational “Wow” sensation with his human peer (the consumer), the role of the security architect is to respond to clever malicious attempts by his peer, in this case – the human adversary. As in the case of art today, where we too-strongly associate “art” with certain types of paintings, in the security case we associate “protection” (or even just “detection") with certain detection of malicious patterns in events. But just like with art, what we essentially deal with is an interaction between humans, which goes way beyond such shortcuts.
Computers will become increasingly better in reviewing event data and figuring out malicious anomalies, and they will get better in uncovering villains and their acts. However, villains will exist for as long as humanity exists, and whatever the scope of AI capabilities is to ever become, criminals will have to find ways around it. Just like artists, those adversaries will certainly succeed, as they forever will be in the forefront of agility and creativity, even if (one day) not in that of flat intelligence. And each time that happens, the only way for the defenders to tackle such creativity will be through the only antidote to the creativity of man – creativity of another man.