Principle 4.5: Design things with a computer that would not be possible without!

Being at Stanford, it seems like AI is everywhere. In conversation, education, and pursuit of future occupation it always seems to get brought up. Sometimes it is positive, how we could extend our capacity through more intelligence, or outsourcing of necessary tasks. Sometimes it is negative, how it might take over the world, be biased, be exploitative. In these discussions it sometimes seems like computers are both incredibly smart and stupid at the same time. You often hear how AI is increasing in its scale and power at an alarming rate that may or may not be a threat to humanity itself. At the same time, vast amounts of research, time, and money continues to go into making computers see the way we do, hear the way we do, move the way we do without thinking twice about it. Sometimes this make sense, there are things we would rather not do, that it might be nice for a machine to do for us. But in reading this chapter I also wondered if sometimes, in forcing computers and our technology to mirror the way we think and act naturally, we may be missing out on alternative ways to leverage this new technology. One of the things that comes to mind for me is hallucinations in using large language models. Often when I hear people talking about LLMs they express concern that they do not have the same internal logic that we do. This is true, but perhaps it is a mark of a poorly designed tool to try to make AI mimic thinking like a human being. Could this kind of technology be improved if it better reflected the kind of logic it actually has? What might this look like?

Yet I also think that it can be true that in exploring something familiar in a new domain we can get a new perspective on the familiar, or even learn something very new about things we thought we understood. For example, in trying to mimic natural vibrato, John Chowning found a new utility in FM synthesis. Likewise, in trying to reconstruct our perceptions in computers maybe we have come to better appreciate how complex our own thought processes really are (and perhaps how they are greatly shaped by the data we’ve been exposed to?). I am curious to see whether or not this new age of technological revolution will show that we understand the medium of the computer or whether we have not designed well enough for the medium.