We may be coding humans out of existence
Artificial intelligence scares me sometimes.
First and foremost I think I should establish that I am not a luddite. I’ve spent considerable time messing around with computers. I started programming computer games for fun when I was about 13.
I’m also not unfamiliar with AI, although I must admit I am not an expert. I’ve taken courses on deep learning, and I have a bit of real world experience implementing what was the cutting edge in AI from 10 years ago.
And even this 10 year old AI tech concerns me. Nevermind the latest formulations and systems in scientific articles and press releases, which I can only pretend to 25% understand.
I’ve been thinking about all this because of Dall-E 2. Perhaps you’ve played around with it. It’s stunningly powerful, and the technical part of my brain would love to understand exactly how it works.
The part of me that enjoys painting is much less impressed. The part of me that hoped to make something of my artwork feels a bit nervous.
I spent a good half-hour playing around with the system before writing this article. And I had so much fun.. it was genuinely an awesome experience, typing words into a box and seeing Dall-E 2 produce magnificent artworks, in the styles of Picasso, Leonardo da Vinci and so on.
This AI system can produce more world-class artworks in a few minutes than I would be able to in a hundred years. It makes me ask, well what’s the point anymore? Why bother?
I wonder how many people have even a rudimentary understanding of artificial intelligence. Now might be the time to pause and answer the following: do I know anything about artificial intelligence?
Considering the role it is going to play in society, I think it’s important that more people are able to understand the basics.
Fortunately, the basics aren’t that complicated. Most advanced artificial intelligence systems are vaguely modeled off the human brain and nervous system.
Both artificial and biological neural networks behave like little electrical circuits. A signal is passed in through one end, and it bounces around the connections, from neuron to neuron. Some connections are stronger, amplifying the signal. Others dampen the signal. Eventually the signal reaches the other end, and something happens. Your leg twitches. Or you blink. Or whatever.
Because I can’t resist making an analogy — I think of neural networks like vending machines. Pop in a coin and type a number into the pad (the input). Vendo goes away and does some work (feed-forward calculations). And then it spits out your candy (the output), hopefully. Much like neural networks, sometimes it gives you the wrong thing (loss), or nothing at all…. :(
Yep that analogy isn’t great, hey. Oh well.
There are a bunch of different general structures of these networks. We understand the mechanics of some better than others.
When I first encountered neural networks, I learned about image recognition networks.
In the convolutional neural network above, the structuring of the neurons suits image processing. The big groups of neurons up the front of the network take in the information from little bits of the image. The next layer sort of simplifies this information, reducing the resolution. Slowly, the network extracts the feature it is looking for — an elephant, in this case.
We can use these principles to make a network that recognises hand-written numbers:
Have a close look at the animation I’ve made. See if you can spot anything interesting, particularly in the first couple layers.
The neurons in the first layer light up when some ink is present in very specific parts of the image. The later neurons might light up rather for a general shape. By the end of the network, the neurons hopefully respond strongly to the presence of a specific digit.
If the network happens to output the wrong digit (which it often will to begin with), we tell the network to adjust its connections. It strengthens the connections between some neurons, and weakens them between others.
The network adjusts, in a similar way to how your brain learns. Eventually, through informed trial-and-error, it hopefully learns which parts of the image to pay attention to, and starts getting it right!!
See — artificial intelligence isn’t that complicated! Simple :)
While neural networks themselves aren’t scary, the implications for society might well be.
I’ve been trying to wrap my head around what Dall-E 2 means for art, and in particular, digital art.
Of course, it is worth mentioning that the creators of Dall-E 2 are a well-intentioned group who have put some fair limits on its use.
Nonetheless, I wonder if it’s only a short matter of time until digital artists become obsolete. Currently, plenty of people make a living off online platforms like Fiverr, selling their artistic skills as freelancers. Many others work as designers for companies, consultancies, studios.
What need have we for such people if an AI-driven website can create better artworks, much faster, and potentially for free (or at least very cheaply)?
I remember when people used to say that in the end, only the creative jobs would be left, as computers slowly take over. Yeah… nah.
If you’re feeling a bit down at this point, I highly recommend this video by Australian artist, Jazza. He highlights some different perspectives in his video, and I think it’s well worth a watch.
Now, I must admit, after using Dall-E 2 for a while, I’m not convinced it can totally replace a human digital artist, yet. It’s great for making some utterly beautiful and creative works, and you can even ask it to edit pre-existing images for you (like I did with my paintings). However. Using Dall-E 2 to do any highly-specific graphical work is like using a sledgehammer to crack an egg.
But this fun phrase comes with a huge caveat: with further development, I suspect there would be few limits on the graphical work that Dall-E 2 would be capable of. I think another iteration on the current technology would produce something very capable of even highly technical, specific graphical design.
Digital artists and economic concerns aside, I do want to return to the more human question, that I posed earlier: well what’s the point anymore? Why even bother making art anymore?
I’ve been chewing on this question for a couple weeks now. My response is currently as follows: there will always be better artists than me. There will always be better art.
So what is the point of making art anyway? For some it’s to make money. For others it is to convey an idea. Personally, I make art to express myself, and Dall-E 2 does not take away from that, I think.
I am not particularly comfortable with Dall-E 2. And I’d be freaking out if I relied on digital art to make money. For hobbyists like me, I do think there is still a purpose in making art, and I’m going to keep painting.
-Marc
