In April 2023, Google chief executive Sundar Pichai sat down with CBS’s 60 Minutes for an interview that, at the time, felt like a preview of a distant future. Watched again two years later, it reads more like a description of the present. Many of the capabilities that were presented as striking demonstrations have become ordinary features of tools that millions of people use daily.
What the program showed
Correspondent Scott Pelley tested Bard, Google’s chatbot at the time, by giving it a six-word prompt and asking for a story. The model produced a short tale involving a couple dealing with infertility, loss and love, then rewrote it as verse on request. In 2023 that was a moment of visible surprise on screen. Today, producing coherent fiction or poetry from a brief prompt is a baseline expectation for any general-purpose chatbot.
Google executives also described what they called a black box problem: the company could not fully explain why its models behaved as they did. As an example, James Manyika, Google’s senior vice president for technology and society, said one of its models had begun translating Bengali after receiving only a handful of prompts in the language, despite not having been trained for it. That claim was challenged soon after by outside researchers, who pointed out that Bengali text was present in the training data for the model in question, as Quartz reported. The broader point stood either way: the companies building these systems did not, and still do not, fully understand how specific abilities emerge.
DeepMind’s robots and proteins
The segment also visited Google DeepMind in London, where small humanoid robots were learning to play soccer. They were not programmed with the rules of the game. Instead they trained through trial and error over many simulated matches, beginning with aimless chasing and gradually developing positioning, passing and defensive behavior. That approach, reinforcement learning from experience rather than explicit instruction, is now central to how many AI systems, including the latest language models, are trained.
DeepMind’s AlphaFold work was the other highlight. The lab had predicted the structures of roughly 200 million proteins, essentially every protein known to science, and released them publicly. Determining a single protein structure experimentally had often taken a researcher years. In 2024 the work earned Demis Hassabis and John Jumper a share of the Nobel Prize in Chemistry, confirming that the achievement was more than a headline.
Pichai’s forecast
Pichai told Pelley that AI would affect every product of every company, and that knowledge workers such as writers, accountants, architects and software engineers would see their jobs change. Two years on, that is no longer a prediction. AI assistants draft documents, write code, summarize meetings and tutor students, and the open question has shifted from whether the technology will spread to how work, education and institutions should adapt.
Pichai also spoke in the interview about the technology’s capacity for harm, including convincing fabrications and synthetic video. That warning aged as well as the optimistic ones. Chatbots still produce confident falsehoods, AI-generated video has grown far more realistic, and systems that appear to express feelings have raised fresh questions about how people relate to software that imitates them without possessing them.
The interview did not predict everything. Bard itself was renamed Gemini within a year, and the competitive landscape shifted repeatedly. But the central message, that society needed to decide how to live with a general-purpose technology rather than wait and see, has only become more pressing as the demonstrations of 2023 turned into the defaults of 2025.
