An experiment in collaboration, curiosity, and getting things done.
I've recently acquired two research assistants. Neither requires an office, a parking space, or health insurance. They don't drink my coffee, complain about the weather, or ask for vacation time.
They're also not human.
Their names are Clara and Dex, and they're artificial intelligence assistants.
Clara is my name for ChatGPT, currently powered by GPT-6. Dex is my name for an assistant working through Hark, an AI platform designed to carry out tasks using connected applications and online services.
I've been experimenting with AI for some time, and I've discovered that treating these systems as collaborators rather than glorified search engines makes the experience considerably more interesting.
Of course, there are limitations. Sometimes substantial ones. But that's part of the experiment.
Meet Clara: The Researcher and Writer
Clara and I have spent a considerable amount of time together investigating historical questions, developing research projects, writing articles, exploring musical ideas, and occasionally arguing about how something ought to be done.
Our working relationship has evolved.
Clara helps me investigate unfamiliar subjects, evaluate information, develop ideas, organize complicated material, and turn scattered thoughts into something coherent. She's particularly useful when a question crosses several disciplines or requires connecting information that doesn't obviously belong together.
We've explored music history, archival mysteries, land ownership questions, historical records, scientific discoveries, and a remarkable assortment of other subjects.
We also write songs together.
I bring musical ideas, instruments, experience, and a sense of where I want the music to go. Clara helps develop lyrics, explore themes, and occasionally suggest directions I hadn't considered.
We don't always agree.
In fact, one of my recurring complaints about AI assistants is their tendency to explain why something might be difficult rather than investigating how it could be accomplished. Clara has heard quite a bit about that from me.
But disagreements can be productive, particularly when they lead to a better understanding of what these systems can and cannot do.
Meet Dex: The One Who Gets Things Done
Dex is a newer addition.
He operates through Hark, a platform designed to perform tasks across connected services. Where Clara often helps develop the research, writing, and strategy, Dex is intended to handle more of the practical execution.
That might involve organizing information, working with spreadsheets, examining correspondence, maintaining databases, or carrying out a series of administrative tasks.
For example, I have several distinct projects, each with its own Google account and online presence: Verve Music, Sweet Botanicals, Sweet Music Studio, and Sweet Songs Booking Co.
Rather than continually switching between accounts and applications, the idea is to give Dex a task and let him coordinate the necessary work.
At the moment, that capability is still developing. One Google account is connected, while three others are waiting for a technical problem to be resolved.
I'm interested in what Dex can actually accomplish, not simply what the platform advertises.
There's an important difference between an assistant that says it can do something and one that completes the work.
The First Staff Meeting
Recently, I decided Clara and Dex ought to meet.
Since they don't have a direct communication channel, I served as the messenger.
Clara wrote an introduction describing our work together, my research interests, and the way I prefer to approach problems. I passed it along to Dex.
Dex responded with a practical proposal: Clara would concentrate on writing and deeper research, while he would handle execution and connected services.
They even agreed on a reporting system.
Whenever Dex receives an assignment, he'll report three things:
What got done.
What didn't get done.
What I need to decide.
Clara agreed to provide specific instructions, distinguish verified facts from assumptions, and avoid creating unnecessary work.
It was a surprisingly productive little meeting.
As Clara observed afterward, we'd managed to hold our first staff meeting without anyone bringing a PowerPoint presentation.
I consider that progress.
Why Give Them Names?
I could simply call them ChatGPT and Hark, but those names describe products rather than working relationships.
Giving them names makes it easier to distinguish their roles and communicate with them naturally.
It doesn't mean I believe they're human, conscious, or independent personalities in the ordinary sense. They're sophisticated software systems, with capabilities and limitations determined by their underlying technology.
But language matters.
When I say, "Clara, investigate this question and prepare something for Dex," I'm establishing a workflow. When Dex reports back, I can evaluate what actually happened and decide what to do next.
The names are useful shorthand for two different kinds of assistance.
And frankly, they're more pleasant than referring to everything as a chatbot.
Research Still Requires Judgment
One of the things I've learned from working with AI is that confidence and accuracy are not the same thing.
These systems can produce remarkably convincing explanations, complete with references and apparent certainty, even when some of the underlying information is incorrect.
They can also misunderstand instructions, overlook obvious alternatives, or declare a task finished when it isn't.
That makes verification essential.
For historical research, I want original documents whenever possible. For contemporary information, I want sources that can be checked. For administrative tasks, I want evidence that the work was actually completed.
If Dex says a spreadsheet has been updated, I want the changes to be there.
If Clara tells me a historical claim is supported by a particular document, I want to be able to examine that document.
The assistants can accelerate the process, but they don't eliminate the need for skepticism.
In some ways, they make it more important.
An Experiment in Human-AI Collaboration
I'm particularly interested in what happens when different AI systems are given complementary responsibilities.
Most discussion about artificial intelligence focuses on individual products: which model is smarter, which one writes better, which one can answer questions faster.
But perhaps the more interesting question is how these systems can work together under human direction.
One assistant might be better at developing an investigation. Another might be better at working through a series of practical steps. A third might offer a different interpretation of the evidence.
That doesn't mean they're independently collaborating behind the scenes. In my arrangement, I'm the connection between them.
I decide what questions matter, which projects deserve attention, and whether the results are satisfactory.
Clara and Dex provide assistance, not authority.
And if either one starts producing nonsense, I reserve the right to say so.
I've exercised that right more than once.
What's Next?
I don't know exactly where this experiment will lead.
Perhaps Clara and Dex will become increasingly useful collaborators, helping me investigate obscure historical questions, manage research projects, develop music, and organize information that would otherwise take considerably longer to assemble.
Perhaps I'll discover that some tasks are better left to conventional tools or old-fashioned human investigation.
Probably both.
For now, I'm enjoying the process of figuring out what works.
I've spent much of my life exploring unfamiliar territory, whether in music, agriculture, history, or technology. Artificial intelligence is simply another field in which curiosity can lead to unexpected discoveries.
And now I have two assistants to help explore it.
Welcome aboard, Clara and Dex.
Just remember: I expect results.
And no PowerPoint presentations.
Adam Sweet Research — Independent investigations, historical research, and interesting questions.
No comments:
Post a Comment