Over the past years, droves of people have sung praises to AI as a technology. They say it is the future, something that will change the world like fire or electricity did. That the very concept of "work" will be revolutionized. Naturally, such these are mostly exaggerations.
In truth, AI is algorithms relating to a long string of calculus equations. It, as well as all software, can be thought of a list of automated instruction on how to turn a series of 0s and 1s into another series of 0s and 1s in ways that are useful to users. AI is no different. The way they are built is by training the intelligence to create an equation that will take in a prompt and output a result based on the most probable (read average) prediction on what the right answer might be. The AI is adjusted and trained better match outputs to desired inputs leading to programs like Leonardo.Ai or ChatGPT.
What these sites often fail to mention is the limitations of AI. A computer can only compute arithmetic sequences and use algorithms to compute algebraic and more complex mathematical concepts. Because of this, everything we must ask a computer to do must be first abstracted in numerical terms, and the code written to output a result that can then be converted back into a useful result. Image generating AIs are given massive libraries of art works in order to create their images. However, unlike inspiration in which there is lateral interpretations and extrapolation, an AI will respond to the prompt, "girl with blue hair" and create an image of a girl with blue hair, not because it understands what "girl", "blue", or "hair" is, but because in its library, there is an image that is label with those words, so that must be the most likely correct answer. This often means that the AI will simply return one of the works in its database, bit rendered as it was passed through a filter. This is, by even the standards of the US courts, not an original work but instead theft. Take into consideration that most artists did not consent to have their artwork used in these programs, and you have an industry bankrupt of innovation.
You might have noticed there have been a lack of talk about AI innovations. In fact, I would wager there is more work on the internet about the anxieties of AI rather than leaps in AI development. There is a reason for this. Because AI is biased towards average based results, most of the things done by AI trend towards the generic. In addition, AI does not understand anything, so when it is asked by, say edit a paper, it will more often than not alter in ways that make sense to a statistical system, but is bizarre and unnatural for a human to say or read, due to the idiosyncratic nature of human language. More relevantly, AI can be outright disastrous when it comes to legal documentation. We have talked before about the court case in which lawyers used ChatGPT to write their arguments and the software outputted fake cases, which ended up being used in an actual trial. You can imagine the judge's reaction when he found out. Because AI has no concept of "reality", it might quote information that is completely fictitious because it is statistically likely information like that might exist, which to the computer is the same as it existing.
People in the AI industry are people who benefit from people believing in the infinite potential of AI. But if AI was so revolutionary, why would these companies not change the world themselves but instead ask their clients to use the technology for them? Ultimately AI might not be a passing trend, but its hype will be.
