Generative Data Intelligence

Software is listening for the options you want it to offer

Date:

Comment A fortnight ago my Apple Watch automatically updated to WatchOS 10 and ever since it’s taken me twice as many taps to perform basic tasks like telling the device to stop tracking exercise sessions.

I’ve grown used to the redesign but I can also feel in my gut that Steve Jobs would have critiqued it by pointedly asking “Why is this getting harder to use?”

My smartwatch hassles are a symptom of a more generalized syndrome. As our computers have grown more capable, we’ve loaded them with features and options that are hard to manage. More than half the time, users have no understanding of the consequences of an option – they’re using the minimal feature set to get them exactly the results they want, and no more.

When users do get confused, their problems lands on the desk of an administrator who is constantly bombarded by problems created by option management. A blanket policy which automates the option into meaninglessness is a common response.

An automated option isn’t an option at all.

When it feels divorced from the natural relationship that emerges from a user and their hardware, the poor design of software can be sensed.

Those failures can be corrected. Yet these subtle, invisible failures – where options multiply so furiously that no one has the time to properly tune them to the task at hand – mean that software never rises to its full potential. It could, but who has time for that?

If time poverty and technical debt mean that we can never get our heads around all the options software presents to us, why have any options at all? Couldn’t the software itself be able to work out the correct settings from an observation of how it’s being used?

That may sound a bit far-fetched, but it’s exactly the line of development being pursued by OpenAI with its new “GPTs” – custom chatbots. The entire process for creating a GPT is conversational; exactly what you’d expect when interacting with an AI chatbot that also happens to be the GPT design tool.

The chatbot asks a series of questions, using those answers to generate the necessary text prompts to give the GPT its specific qualities, propensities, tone, knowledge set, and guardrails. The whole process takes just a few minutes to spin up a GPT. Although the interface provides a tab with a few basic configuration options, they’re never necessary to the process. Most GPT creators will never touch them, nor need to. Ever.

Whatever your opinion about AI, it’s obvious that using a linguistic interface to configure something as complex as a chatbot represents a triumph of usability.

Rather than struggling with dozens of configuration options, OpenAI presents a simple conversation as the main interface – effectively removing the bar between an idea for an AI chatbot and its realization as a GPT. That means we’re going to see exponential growth in the number of GPTs over the next few weeks – and that we’ll see this same interface adopted by many other types of software.

Given that Apple is now reported to be spending $1 billion a year on AI research, and iOS 18 promises to be “AI infused,” we can begin to imagine how we will soon manage the increasingly complex edifice of software – conversationally. Windows Copilot can already adjust system settings to suit the needs of individual users, MacOS, iOS, iPadOS will work out how I want to work by observing me at work, and WatchOS will finally respond to me screaming “just quit that exercise, WILL YOU??!”

That world will not be perfect. We don’t vanquish complexity by hiding it behind a linguistic interface; instead, we move from complexity into ambiguity. “Did my smartwatch do what I meant – or do what I said?” That’s the next hill to climb. ®

spot_img

Latest Intelligence

spot_img

Chat with us

Hi there! How can I help you?