Kids are already getting As on essays that are GPT-3 output they skim through.
I had not heard of this, but I had also not looked. I would hope this is not happening since that seems ridiculous.
People in 100 or 500 years would probably laugh at us having a discussion board worrying about DALL-E and its implications, I imagine. Because I think if we focus on one application of the underlying advanced learning algorithms, and not the underlying algorithms themselves and the full scope of other possible use cases for them, then we would overlook the way those algorithms might affect society at large in the future. As I understand it, DALL-E and GitHub Copilot and similar technologies are primarily based on the "Transformers" as far as their implementation. I'm a bit ignorant on how exactly those are written, but it appears that they use learning algorithms to find parallels between two corresponding interpretations of data, like the association between an image and the caption describing it, or the association between a code black and the comment describing it.
In general, being able to computationally understand arbitrary inputs (or even the association between them) is just one step closer to the theoretically all-encompassing topic of artificial general intelligence. Obviously it might take us a long time to get to artificial general intelligence, but it may be worth considering that in the long-term that is the likely outcome of this kind of research.
Nevertheless, I guess you raise an interesting point; in my own life, when I was initially concerned not about some theoretical future artificial general intelligence but rather about the problem of excessive use of data mining even in the present using only the techniques that humans currently know -- and I was concerned that that excessive data mining appeared to already be happening to the detriment of my own life and mental well-being -- it seemed that if I would search related topics on YouTube or online, the internet tended to push back at that and instead end up feeding me content about the long term problem of artificial general intelligence in the future and its ability to destroy all human life if it wanted to, in theory. It would appear that with social media today, excellent data mining was used to push me more towards spending time on Elon Musk-flavored clickbait nonsense and away from solving current problems in our society.
Does that then suggest that we already have a pretty big problem, or is that just me and my own particular personal data that was getting used?
I do not claim to know the exact answer, but as someone who had "data mining" and "machine learning" related courses in college, and who struggled not to sleep through them because of what a past generation of humans might have called my "YouTube addiction," I do have a sense that something is wrong or is going wrong.
And, I think that I have been personally incredibly contradictory on this topic, because back in the early 2010s I was really into the idea of empowerment through better technology in my own life and especially in my Warcraft III modding. Improving my own modding technology workflow was an incredibly enjoyable and enriching experience. It seemed that it should be limitless. So, how can better and better technology be a problem?
Yet I think almost everyone today may have a sense that something is wrong or is going wrong. Most governments and established systems are under some kind of attack, whether visible or invisible. It might be the entropy of the universe and machine-learning-boosted, rage-inducing clickbait. Or it might be something far more sinister where some intelligence (whether digital/artificial or even just human) wants to improve its own control over everyone's thoughts and governmental systems. But I think especially for systems that were previously democratic, it can sometimes be apparent that the people who are the most involved in government appear to be given the most targeted false and misleading information. Conversely, for systems of government that are more top-down and driven by an all-powerful type of fascist or totalitarian regime, the humans at the top may be the most targeted in ways that lead them to make what would appear to be almost hilariously bad and uninformed decisions.
Government is relevant to machine learning because as far as I know, at present human lives still generally control the flow of societies and activities on most of this planet, or at least the parts that they can reach or touch. We know this was not historically always the case, and in the future there may come a time when it is again not the case. In light of that, what is so critically important to us as a people would seem to not only be whether artists have jobs, but additionally how we use and interact with our data processing technology as it becomes increasingly powerful.
Honestly, at times I struggle with it. I tried changing my desktop wallpaper on my computer for a while to a big alert message trying to warn me that my understanding of what a computer is and what it does was incorrectly trained growing up. I wanted to believe that when I use this kind of device, I click or command the device to take an action and then the action is taken. In the end, this is a misuse of something more valuable -- like if you beat your wife until she submitted to your every command and ignored that she was human. For a long time, I ignored the potential of the computer to be compounding future computations based on pre-programmed instructions that I could have written, so that it would be supportive towards my predicted future goals and personal interests. A 1 GHz processor can do 1 billion mathematical calculations per second! Even if we assume that the data fetch mechanisms cause delays so we lose efficiency and essentially lose 99 out of each 100 calculations to wasted effort, that would still mean we could do 10 million possibly meaningful calculations per second! That is 10,000,000 calculations. Your phone could probably solve every math homework problem you were ever given in school, and then perhaps also every calculation for how much physical change to receive from every financial transaction in your entire life, faster than you can finish reading this post. But it would have to have been given the correct instructions! As examples of how I could have used my physical devices better:
- I could have pre-programmed my devices to report their status and physical location (if available) to my personal web server any time that they were online, and always check for new instructions, so that all technology I own would at my command (when powered on) even if it was away from my person physically.
- I could have pre-programmed one of my physical devices to check my bank account every morning, check the public records of company stock values, and then update my personal investment portfolio each day based on some loose prediction for the future expected value of any given investment.
- I could have pre-programmed my devices to offer an alert at a consistent time each night indicating that I should sleep unless some critically important task needs my attention (emphasis on "critically important" -- this is rare!)
- I could have pre-programmed my personal devices to host some kind of automatically generated calendar that could advise me on fun ways to spend my time each day that had no bearing on any corporate advertising and were based only on my own personal recordings of how I liked to spend my time, which are not to be shared with any corporation
We may think that our "cloud" technology today was going the direction of what I have listed above, or better, but I think that in general for most people who use technology today even these basics are stored not on each our own personal private web servers, but rather on corporate-owned services where a third party reviews our records of our lives and processes how to use them to get an advantage over us each individually. I feel as though I ignored this problem until it was too late, and often even though I am subconsciously aware of possible solutions I find myself feeling that it is already too late and I will most likely not win this effort, and so frequently I forget to even try. Even though maybe I am a bit ashamed to admit to that directly.
So, when I think about AI generated artwork and its impact on society, I guess I am thinking this is only the beginning. A month or two ago when I was first given a link to a test version of a DALL-E knockoff, one of the first things that I asked it to draw was "A new version of Warcraft III that all players really enjoyed."
I look at these bizarre and almost dream-like renderings of something that appears to be, but isn't, some primordial version of the Warcraft III game and I cannot help but almost wish that i could play the versions of this game dreamed up by DALLE and its knockoffs. But I think this highlights why I brought society and government into the topic a bit more -- I personally find myself feeling like the ever-growing power-creep of the desire for better modding technologies on Warcraft III for me will end when society reaches a point where an artificial intelligence becomes sufficiently advanced to generate Warcraft 3 mods and spinoffs instantaneously based only on loose prompts like, "A version of Warcraft III where the Naga is a playable race" or "A version of Warcraft III where each race has five heroes."
Presently with the 2022 technology, if any company has artificial intelligence advanced enough to instantaneously produce Warcraft III mods then I am personally not aware of it. As far as I can see, the best we have yet would be these image generating transformers that can spit out a dreamlike render of a screenshot from a Warcraft mod that I never had. But if we think about how even before society has artificial general intelligence, they might some day have a deeply powerful code-generating transformer that could generate any Warcraft III mod based only upon its description, I realize something.
I realize that when I mod Warcraft III, in the end I am doing something futile. Like a monkey building a stone door by using a big rock for his cave before the first humans are born who create cave-drilling machines and concrete. When I create a Warcraft III mod, I create something that may be without a greater purpose and is far more temporary than I want to admit to myself. But if in truth everything is nothing, and all that I have ever created or ever will create inspired by Warcraft III is actually without value and will be created a thousand times over in the blink of an eye in the next age of this planet, at least I may take solace in believing that the digital representation of my experiences as it appears on Hive and my YouTube channel and other places might be a digital representation used as part of the inputs to help a learning algorithm to dream of its own Warcraft III mod some day.