You say you use Artificial Intelligence (AI) a dozen times a day? A dozen times an hour? Unless you are using it almost constantly, you aren’t using it enough. The current state of AI is like a well-educated, super quick intern who is willing to help or give an opinion about anything you are working on. While you can’t currently count on it to do everything just the way you want, it is nevertheless an invaluable tool for almost any task. Although I have long been skeptical of the panic surrounding AI’s potential to threaten white-collar jobs, I am increasingly in awe of AI’s current capabilities and I can’t even imagine what the future is like.

AI helps me with almost everything I do. I have a number of different ventures going on, of which this blog is a minor one. Among these are being on the Board of Directors of a tech product startup, coding for the major open-source AI project: LlamaIndex, consulting on blockchain technology for a large brokerage, and creating an AI to collect and answer questions about colleges and universities.

AI helps me with all of these, and with this blog. For this blog, I don’t allow the AI to write for me. This is not out of any objection to the idea of it writing my content. It would save me a lot of time to have AI write this blog for me. It’s just that I don’t like its writing. I’ve gone through a fair amount of effort trying to get AI to write like me and while the results are good enough to fool the ChatGPT detectors, I don’t feel the writing represents my style correctly.

I do, however, use AI in the form of ChatGPT, constantly in my writing. I ask it to define terms or suggest different wording if I’m having trouble with a paragraph. I ask it to proofread, but not rewrite, all of my blogs. It does a great first pass, but it always misses some typos and grammatical issues. I still need an experienced human editor to ensure that it’s all correct.

As a specific example, yesterday my main computer system was being used to go to college websites and gather information about each school, a very time and compute-intensive task. As my other projects were in a brief lull, I found myself with a rare free afternoon. Having been looking at some of the data about colleges for the data gathering project, I got to wondering about where in the country the highest concentration of colleges is. I always hear that Boston has more colleges than anywhere else in the USA, but the Philadelphia area has a lot as well. I thought it’d be a fun project to put together a quick heatmap to show the location of the colleges.

A few decades ago, I put together a heatmap showing casino visitation rates for people across the USA, and I thought maybe I could do something similar to see where the most colleges are located. Putting together that casino heatmap years ago involved gathering data, finding a suitable map, and researching data representation methods. It took a month’s worth of work. The result was very impressive, but quite time-consuming.

With some of the newer tools, I knew it would be easier, but I still thought it’d be enough work that I might not have time to finish what is really just a fun little side project. What I thought might be complicated was remembering how to represent latitude and longitude on a map. I remembered that it involved calculus and converting things to radians and the like, but I had long since forgotten the specifics of how it was done.

I went to ChatGPT to see if it could give me a primer. First, however, I explained to it what I was trying to do and gave it a few possible approaches about how to display this data– a process called ‘clustering’– and asked if it had any other ideas. It did, in fact, have many ideas about how to go about it.

Then, to my surprise, it went about doing almost all of the work. It encountered many errors along the way and made many mistakes, but it kept recognizing these mistakes and attempting to fix them, sometimes without my even asking.

Like a knowledgeable intern, we worked through the process together, each contributing observations. I did have frustrations, such as the AI repeatedly asking me to upload a file it already had in memory, but overall I managed in a few hours what took me a month to do just a few decades ago.

I have long worked closely with other programmers, especially when doing data analytics tasks. It’s complicated work and having different perspectives on how to handle it is important. AI now provides that other perspective, and in a way that’s as quick as any person, and quicker than most.

More to the point, the AI is close to not needing my expertise at all. If you even glance through this transcript of me and GPT4 working to create this heatmap, you’ll see that it is like me providing overall direction and only occasional specific guidance to a very capable assistant.

Just over the past year, AI has gone from being helpful on specific tasks, such as understanding how to handle radians, to general tasks, such as taking this data and making a heat map from it.

A friend of mine recently commented that software engineers are often wary of learning AI for fear that it will take their jobs. I see things very differently. AI is already at a point where it can greatly enhance very skilled humans and teach those not as skilled. AI’s capabilities continue to expand at a dizzying pace. From quickly generating heatmaps to streamlining data analysis, the practical benefits are amazing. AI is quickly becoming an integral part of many jobs and, like software engineers, those who embrace it will prosper while those who resist it will ultimately be replaced by those who do.