Artificial intelligence is changing things at dizzying speed. About 18 months ago, the tech company OpenAI unleashed its AI chatbot, ChatGPT. Within a couple of months, 100 million users were regularly using the tool, making it the fastest-growing consumer app in history. While tech bubbles are always easy to slip into, many people argue the world can be divided into a pre- and post-ChatGPT world.
That interest wasn’t a blip. This week, the web traffic analysts Similarweb announced ChatGPT’s website hit new record highs of interest, with 83.5m visits on a single day in May. The premise and title of my book released last week, How AI Ate the World, appears to be true. AI is now basically inescapable.
Yet touring the country to talk about it, I still meet holdouts; people who don’t want to be part of the AI revolution, or haven’t yet seen the need to interact with a text-based chatbot. An announcement on Monday by OpenAI of a new model, GPT-4o, may change that.
For the technically minded, GPT-4o is a significant change. But for the general public, the important difference is how easy it is to interact with. Prior to GPT-4o, the primary way of interacting with ChatGPT was to type text-based questions and wait for text-based responses. A voice interface was available, but was clunky and slow. I have tried, in recent months, to get ChatGPT to help teach me German – to better interact with my partner’s Austrian family – but the agonising delays between me asking questions, and ChatGPT formulating a response and then synthetically vocalising German words, often in incomprehensible and unaccented American English, made it next to useless.
The tech demos shown by OpenAI earlier this week change that. In one section of the launch event, ChatGPT acted as a real-time interpreter for a conversation between English and Italian. In another, it laughed in response to a “top-tier dad joke”. And in another, it switched from a rote reading of a bedtime story to a dramatic reading that even Brian Blessed would blanch at, before concluding with a song.
According to OpenAI, this is the new normal: an AI model that can “reason across audio, vision and text in real time”. It appears, at first glance, to be another significant step towards turning science fiction into science fact. The always-helpful, always-on, human-like robot butler that we’ve seen and read about for decades is getting closer, OpenAI would suggest. And the impressive smoothness of the interaction might shunt a few naysaying holdouts towards being AI adopters. Making it free, as OpenAI has done, will also help.
However, it’s worth remembering AI’s original sin, dating back to 1956: its naming. “Artificial intelligence” is certainly artificial, but it’s not yet intelligent – and arguably never will be. The more that ChatGPT and other tools like it mimic human interaction, learning to act as witty, wisecracking raconteurs that can croon and swoon, the more likely we are to forget the “artificial” bit of the term.
The smooth interactivity that OpenAI has laboured hard to enable does well to paper over the cracks of the underlying technology. When ChatGPT first elbowed its way noisily into our lives in November 2022, those who had been following the technology for decades pointed out that AI in its current form was little more than snazzy pattern-matching technology – but they were drowned out by the excited masses. The next step towards human-like interaction is only going to amplify the din.
That’s great news for OpenAI, a company already valued at more than $80bn, and with investment from the likes of Microsoft. Its CEO, Sam Altman, tweeted last week that GPT-4o “feels like magic to me”. It’s also good news for others in the AI space, who are capitalising on the ubiquity of the technology and layering it into every aspect of our lives. Microsoft Word and PowerPoint now come with generative AI tools folded into them. Meta, the parent company of Facebook and Instagram, is putting its AI chatbot assistant into its apps in many countries, much to some users’ chagrin.
But it’s less good for ordinary users. Less friction between asking an AI system to do something and it actually completing the task is good for ease of use, but it also helps us forget that we’re not interacting with sentient beings. We need to remember that, because AI is not infallible; it comes with biases and environmental issues, and reflects the interests of its makers. These pressing issues are explored in my book, and the experts I spoke to tell me they represent significant concerns for the future.
So try ChatGPT by all means, and play about with its voice and video interactions. But bear in mind its limitations, and that this thing isn’t intelligent, but it certainly is artificial, no matter how much it pretends not to be.
Chris Stokel-Walker is the author of How AI Ate the World, which was published earlier this month
