The questions about artificial intelligence are big, alarming and perhaps alarmist: Will A.I. systems soon move beyond human ability to control them? Could they run amok? Will A.I., in fact, kill us?
In recent weeks, these questions — and answers that many find unsatisfying — have prompted calls for government intervention and a global slowdown of A.I. development, with tech leaders like Dario Amodei, the chief executive of Anthropic, voicing concerns about the safety of the technology. I asked two tech reporters, Mike Isaac and Kalley Huang, and their editor, Pui-Wing Tam, to share how they approach providing readers with clarity on a topic that even its leaders are still figuring out. Read our edited and condensed conversations below.
Looking back, do you think mainstream media has appropriately calibrated A.I.’s possibility, powers and risks?
PUI-WING TAM: From the start of this A.I. boom in 2022, we wrote about what A.I. did well (generating text, for instance) and what it got wrong (A.I. can hallucinate) and how the technology has since evolved. There are a lot of misconceptions about the technology — sentience is probably the main one that keeps popping up (A.I. is not sentient!) — so we see it as our job to explain, disentangle and clarify.
MIKE ISAAC: What I think is happening right now is a broader recognition of theoretical safety issues that researchers have spent years talking about and, frankly, we’ve spent years writing about. It’s just that we’re in an environment where statements like those that came from Jacob Coxon, the researcher who quit Anthropic this month and sounded an alarm, are for whatever reason breaking through to the public.
I would guess a lot of that comes after the past year of pushback around data center building, economic anxiety around job loss and a sense of more negative sentiment around A.I. from the public.
How has our coverage evolved?
PUI-WING: At each turn, we want to understand all the advancements of the technology and help readers understand the progress, the risks and the flaws.
MIKE: I would argue it’s more necessary than ever to exercise caution when covering these risks. The environment is so charged and people who have not followed the technology closely for years are instantly jumping to the reference points they’re familiar with — a “Terminator” scenario, for instance. (It is not lost on me that last month was the 35th anniversary of “Terminator 2: Judgment Day,” which saw the film rereleased for a short theatrical run.)
We’re trying to cover what’s really happening, including worrying behaviors like the Hugging Face hack, while also not trying to prognosticate too far down the road.
While some tech executives and politicians are urging a slowdown to A.I.’s growth, other leaders — including President Trump — have rejected calls for regulatory intervention. What do readers need to know about industry guardrails or the lack of them?
KALLEY HUANG: The United States has few rules on developing A.I. Earlier this year, President Trump signed an executive order asking A.I. labs to voluntarily submit their models for review before public release. Some labs have done so but this is, of course, voluntary.
Most labs have safety teams that work on what the industry calls “alignment,” or making sure A.I. understands what humans want and acts accordingly.
Some of the regulatory naysayers have said the labs can slow down development on their own, which is not untrue. Mark Zuckerberg recently said Meta delayed one of its products for months to work on safety and security, for example. But the supporters of regulatory intervention — chiefly Anthropic — have been asking for permission to coordinate on a slowdown without getting in trouble with antitrust enforcers.
How do you go about assessing threats posed by A.I.?
PUI-WING: We talk to the A.I. companies, we talk to third parties who run tests on the technology — really anyone who is involved in the creation and advancement of it — to get as much of a 360 view as possible.
MIKE: This sounds simple, but it really does boil down to reporting. There are differing views on this, especially in this field. Some people in the A.I. community tend to use what is called a “Bayesian” approach to wrestling with these existential problems, which is essentially a line of thinking that continuously adjusts the probability of a doomsday scenario based on past and new evidence.
I feel like that comes with pitfalls, especially when it freaks everyone out by putting percentages on an outcome that seems difficult to pin down. For our purposes, we tend to stick to the reporting we can observe.
KALLEY: I ask about what A.I. has actually done, and how it would actually do some of the things that A.I. researchers warn about. It’s easy to talk about threats in the abstract, but hard to assess them without at least some understanding of the nuts and bolts. I’m also curious about whether threats are because of A.I. misbehaving, or humans misbehaving, so I ask about that.
Most of all, though, I try to understand people’s and companies’ motivations. On both sides of the debate, there are incentives — often financial — to shape how the public and politicians think about A.I. It is important to factor that into assessing threats.
To me, sometimes the conversations around A.I. feel similar to climate change. This consequential thing is happening at a global scale, and everyone disagrees on how — even if — to handle it. Do you see any similarities or learnings from climate change policy that could lend itself to the A.I. space?
MIKE: I like this comparison and have used it when talking about the issue with folks in the field. I think it’s apt particularly because of how enormous the consequences are. So much so that it can be stifling to think of what to even do.
The heartening thing, I would argue, is that like climate change (at least historically), artificial intelligence is now seen as an issue to be taken seriously and at least try to address at an international level. That’s more than we can say about how regulators have looked at A.I. even just a few months ago.
KALLEY: I get the impression from people who work on A.I. and A.I. policy that that’s how they’re thinking about it too. This is a slightly more cynical perspective than Mike’s, but I’m also struck by how much talk can happen with so little action.
Part of covering A.I. means occupying a space where journalists don’t want to be: as part of the story. How do you ensure fairness when we are adversaries in a major lawsuit against Microsoft and OpenAI?
PUI-WING: The lawsuit is rarely in our minds when we report and write. That’s a business-side effort and we, as journalists, are not involved in the legal work. Our job is to report, inform and reveal what is happening at Microsoft and OpenAI. We will keep doing that.
Is it personally hard to cover an industry that is creating technology that does some of what many of us do as writers and editors?
PUI-WING: No, it’s fascinating to cover the A.I. industry. I’ve been covering or editing tech for more than two decades, and the scale and speed of change in Silicon Valley is the fastest and biggest that I’ve ever seen. Yes, A.I. can generate writing, but journalism is so much more than just text (or photos or video). A.I. cannot obtain original reporting and it cannot make the millions of unseen and tricky decisions that are often required in putting together a story.
KALLEY: Agreed. These are exciting, weird times. I like learning about A.I. and helping people understand the technology, the people and companies who make it and how to think about their place in our world. And I also don’t think A.I. can do original reporting either. People trust other people, and people like talking to other people.
What has surprised you about your work?
MIKE: How genuinely excited and interested I am covering this stuff. It’s corny, but my favorite part of the job is talking to people steeped in the work and learning more about it.
KALLEY: How contained of a story A.I. was. We in San Francisco are swimming in A.I. We see A.I. on our billboards and buses. We, unfortunately, hear about it in our gyms and bars. And much of the Bay Area works in or adjacent to technology.
But for a while, I don’t think that my friends and family in New York and North Carolina thought about A.I. nearly as much, or at all. That’s changing now, though I think there’s still a gulf between San Francisco and the rest of the country.
The A.I. space is packed with what-if questions around the future of this technology. What if chatbots take our jobs? What if A.I. is used to create a new wave of biological weapons? What “What if” scenario do you worry about?
MIKE: I find spending too much time worrying about the future paralyzing. There is certainly value in projecting outward, but playing the tape all the way out to the end of the world seems like it’s not that fruitful.
I do think it can be helpful for us to see what products are being released and how they can be used for good or ill and write about those. And again, it’s most useful in my mind when we follow the reporting on how people are doing those things.
Looking at Meta could be instructive. There’s a discussion online about how the internet might be made unusable by bots that pollute it with bad information. Is that a reasonable concern? I would argue yes, and we don’t have to look far to find it; take a look at how much A.I.-generated “slop” is already filling Facebook and Instagram feeds.
So … are we appropriately scared? What keeps you up at night?
PUI-WING: Whether Mike and Kalley are working too hard and need a break because the A.I. news flow is so nonstop!
MIKE: I’m more worried about bills and whether I locked the back door of my house than whether I’m going to get taken out by a robot tomorrow.
KALLEY: Sometimes, the mosquitoes in my apartment.
Do you use A.I. in your own work?
PUI-WING: Nope!
MIKE: Nope!
KALLEY: I sometimes use apps that record and transcribe my interviews, but I inevitably end up listening through those recordings anyway. Otherwise, no.
PUI-WING: Kalley broke our streak.
The post How Three Times Tech Journalists Make Sense of A.I.’s Promise and Peril appeared first on New York Times.
