Why artificial intelligence still seems to be in the dial-up era

Generative AI has an outdated interface, is slow, can be expensive, and still needs to evolve a lot.

If you lived through the dial-up internet era, you probably remember what it was like to access the world wide web back in the day. Your computer would dial a phone number and make beeping sounds.

Then we would hear a hissing sound as it tried to connect to the server on the other end of the line. Only then could we access the wonders of the online world. And when we disconnected, everything disappeared.

I’m not saying this out of pure nostalgia—I have a good reason. The dial-up era feels a lot like the current era of generative AI. The more I think about the similarities, the clearer it becomes that AI still has a long way to go before it can live up to all the hype .

The most obvious similarity is in the look and feel. From the first time I used ChatGPT, I couldn’t help but think of BBS, an old system that allowed people to communicate via text messages before the internet as we know it. 

The user interfaces are nearly identical: you type something into an empty window, press Enter, and get a text response. You can even see the AI ​​chatbot’s responses coming in character by character, as if you were using an old-fashioned modem. In an age where most computing experiences offer instant feedback to our clicks and taps, this seems like a pretty rudimentary way to interact with software.

However, the slowness of AI has nothing to do with the speed of your internet connection, but rather with the fact that the technology is performing impressive “computational magic” that overloads even the supercomputers on which it runs.

This delay reminds me of the many times I gave up on doing something on the internet, like downloading a program of several megabytes, simply because the wait wasn't worth it.

Generative AI features often feel this same isolated and measured feeling. The simple fact that apps often use a shiny icon to represent it is a sign that the technology hasn’t yet been fully integrated.

Rather than treating it as a standard feature, many software companies are seeing it as an upsell opportunity. For example, Microsoft 365’s Copilot costs an extra $30 per month, which effectively forces organizations to be more selective about who gets access. It’s as if employers have to decide which employees need internet access and which don’t.

People are certainly embracing AI much faster than they embraced cyberspace. The first BBSs and mass-market online services appeared in the late 1970s. Even 24 years later, only 6 percent of Americans had access to the internet.

Today, according to a Reuters Institute study , 32% of people have used ChatGPT. However, only 18% use it at least once a week — suggesting that the service has yet to become a necessity like the internet, even for a significant portion of the people who have tried it.

The industry can’t make everyone use AI overnight. A recent Bloomberg article points out that only 3% of computers sold this year will be “AI PCs,” and that major software companies are wary of adapting their applications for such a small slice of the market.

Even with a careful strategy, Apple’s AI won’t be ready in time for the release of iOS 18 and other planned updates — and it won’t work on millions of older devices still in use. At this rate, how much longer will it be before every desktop, laptop, smartphone, and tablet is AI-ready?

The Internet only reached its potential after the end of dial-up. According to the Pew Research Center , in 2007 alone, broadband was able to reach 51% of American homes, which means that 49% of them still did not have Internet access or relied on dial-up.

AI needs time. One day, it will make today’s ChatGPT look as antiquated as 1998’s AOL looks now. Just don’t expect it to happen in a year or two.