‘We try to think out loud’: OpenAI’s Sam Altman on closing the gap between AI and how we’re adopting it

Artificial intelligence is advancing faster than many businesses, institutions and societies can adopt it, creating a new leadership challenge: how to be transparent about what is changing while staying agile enough to respond, OpenAI CEO Sam Altman says.

Speaking with Commonwealth Bank Chief Executive Officer Matt Comyn via video link at the bank’s Accelerate AI event in Sydney, Altman said while AI technology had reached a significant level of capability, the economic and organisational adoption of the technology was still in its early stages.

"Overall, I think the technology has gotten to a notable place," Altman said.

The capability of the models had moved ahead of their deployment across companies and the broader economy.

"We have these incredibly smart models [but] I think one has to look at the state of the economic adoption and say we're still very early," he said.

The gap means there is still plenty of work to do to integrate AI into society responsibly and for businesses to start seeing real bottom line benefits, Altman said.

But that work was coming at us quickly, and we can’t afford to take the “psychologically convenient” option of pretending that the changes were all way off in the future, he said.

Transparency in an uncertain time

Altman said the scale of AI's potential impact meant companies needed to talk about real concerns and not just rely on polished messages.

He said OpenAI has tried to build trust by sharing its thinking openly, even when those views were incomplete or later proved wrong. "We try to think out loud," he said.

"I believe that so much of society here is going to be impacted by this, that we are all stakeholders, and it is better for us to be going in the direction of too much transparency and occasionally being wrong."

Altman said OpenAI's own record showed how difficult it was to be certain where things would end up, saying the company had been more accurate on predicting how technology would develop than on the broader social and economic effects.

"My scorecard, at the highest level would be we've been roughly right on technological predictions and pretty wrong on the social and economic implications," he said.

One of the areas where he personally had been wide of the mark was on AI’s short-term impact on entry-level white-collar jobs, which had not been nearly as bad as he had once predicted, he said. “I’m delighted to be wrong about that.”

Keeping AI human

Asked how his own use of AI had changed as the technology had evolved, Altman said one of the most revealing moments came when he tried using AI to manage personal communications, including emails and Slack messages.

He said it was an important example of drawing a line on what we did and didn’t want AI to do.

"We really do care about our interactions with people,” he added, saying his personal communication “which is a huge amount of my time, is not something that I can imagine myself outsourcing to an AI anytime soon".

He said that experience has also helped shape how he thinks about how humans and AI will interact as the technology becomes more prevalent. Ultimately, “the world has got to be built for people and be better for people", he said.

AI agents and new ways of working

Altman said companies were still working out how people and AI systems should collaborate, particularly as AI agents begin to interact with workplace tools and systems.

Human interactions meant there were well established norms in how companies communicated and did things, he said.

While technology could learn those norms, “we care about people. We don't care about machines that much," he said.

"We have expectations about what we do with a person. And right now…we have not yet figured out how we're going to have a world where people and AI co-collaborate together."

He said one of the issues was that today's AI agents were often being pushed through communication channels designed for people, but that was unlikely to be the long-term model.

"What I expect will happen is we will figure out new ways for agents to use our same services and interact with our same systems and data, but via a different channel," he said.

Similarly, the next major shift in AI interfaces could be systems that are persistent and always running, rather than tools that only respond when prompted.

"Today, you still ask the system to do something for you, and it goes off and, you know, tries to figure it out and comes back.

"What I think will be possible soon is you will have an AI that is always running. It is understanding you and your goals and your company's goals. And it's just trying to be as helpful as it can given the amount of computing resources it has available."

Resetting the speed of business

Altman said one of the main questions CEOs were now asking was not whether AI was happening, but how to run organisations when the technology was changing faster than traditional corporate planning cycles.

"How can I run a company on an annual or quarterly cycle when the whole world is changing every month, or every two months, or less," he said.

"I think that business is going to get reinvented when the world has to move at a much faster clock cycle to be competitive," he said.

Trying to retool businesses to meet the challenge was both “unbelievably difficult and inspiring to watch", he said.

Altman said the speed of change meant companies could no longer wait for perfect certainty before moving. He pointed to the rapid adoption of AI coding tools as an example of how quickly enterprise thinking could change once leaders recognised the competitive implications.

"It was one of these moments where people realised, hey, if we don't get serious about this, we won't be competitive," he said.

Acknowledging there had been upsides and downsides, it was “truly one of the most rapid adoptions of new technology at a serious enterprise level .. that I've ever seen", he said.

But at the same time, "no one has a playbook about how to deploy those quickly enough across the company and make sure that people are being productive and secure with it," he said.

He said the leading companies were allowing controlled experimentation, learning from use and adjusting quickly. "This is the thing that I've observed the best companies do," he said.

Altman said OpenAI’s own approach has been making more bets, learning quickly and shifting resources decisively when one of those bets starts to work.

He said the difficult part was not experimentation, but letting go of other priorities quickly enough. "It's easy and fun to try a lot of bets. And then no matter how well one is working, it's always painful, in my experience, to stop doing other things, to concentrate on one area."

The productivity question

For all the advances in capability and adoption, Altman said leaders were rightly asking whether AI was translating into measurable productivity and revenue gains.

"My best answer to that is it's all still very new, and it's just going to take a little bit longer, to figure out how a company actually does run more efficiently and to make these great new products," he said. “But if a year from now we’re still talking about the same question, I'd be more concerned.”