Every challenge becomes a nAIl
Let me be clear to begin with. I am not an anti-AI luddite advocating for destroying AI as either useless or an existential threat to humanity. All the current AI hype, however, reminds me that when you have a hammer, everything becomes a nail.
Have I used AI? Sure. I pay my monthly co-pilot subscription for help in VS. I’d grade that help as fantastic for drudgework, less so for bigger things. For some reason, the AI has trouble with correctly balancing braces when suggesting finishing code sections, but it is better than manually typing those sections out.
If you listen to the news or pay attention in tech journals and ads and what job listings are asking for, you would think that AI is the greatest thing since Betty White (see footnote). The thing is, AI is a tool. It may be a powerful tool, but it is still a tool. It has already done some interesting and potentially great things (potential new battery technology looks fantastic), but it also has shortcomings, as highlighted in this article on forecasting tornadoes on VOX:
These algorithms depend on good data to teach them, and that poses a major challenge for getting ahead of this particularly confounding phenomenon: As global average temperatures rise and as land use changes, past tornado activity might not reflect how these storms will whip through cities in the future.
And that is the biggest challenge facing using AI today. The models are built on the past and may not be updated quickly enough to handle the future, especially if the future is rapidly changing.
Here’s an interesting thought experiment I had recently. If AI had existed in the mid-eighteenth century, it would have been very familiar with phlogiston, the then hot topic to describe what happens what we know know as oxidation. AI might have been feverishly working to determine how to make the data that didn’t seem to fit the phlogiston theory work, just as some of the greatest minds at the time were doing. And there is nothing wrong with that. Many scientists of that era fell victim to not being able to see that phlogiston theory was wrong, they were incapable of shifting their mind-set to a new way of thinking.
In that same way, we need to rethink data in terms of AI, which is a difficult reframing but not an impossible one. The biggest challenge is going to be dealing with the GIGO-complex. If you have been working with computers for any length of time, you have probably heard GIGO: Garbage In, Garbage Out; a pithy way of saying that bad data won’t produce the results you want.
Sadly, tech history is filled with people who have taken the output from computers and believed it has to be correct. In the lead up to the dot.com bubble bursting, there were people who changed GIGO to be Garbage In, Gold Out. If the company doesn’t fully understand the data going in and the code that is processing it, then it’s easy to believe in the infallibility of the computer and, if the computer says you are going to be making a million dollars next year, it must be right. Add the imprimatur of AI to the task, and the phrase can go to Garbage In, Gospel Out.
Let me be clear, AI is an important advance. AI is a tool, though, it needs to be applied correctly. It may or may not be the correct tool for solving the problem at hand. Or, you may need a different type of AI. Just as tearing out a cement wall with a claw hammer can be done, but can be done more effectively with a sledge hammer, using the correct AI modeling will give more correct answers.
1
Betty White was born before sliced bread and, arguably, is a better gauge of greatness than sliced bread.