AI is either going to cure cancer or kill us all. That’s the polarizing thrust of a panic-infused debate that erupted across the industry in recent days.
The truth is likely more mundane, and the risks more manageable than some researchers at leading AI labs would have you believe.
That’s according to an unusually bipartisan collection of tech experts, industry leaders, and policymakers speaking out against extreme AI-doom predictions.
Their main message: AI is a tech product, not a sentient being or a god. So take a chill pill. The world can handle AI pretty much with the resources and structures we already have.
AI is digital, not real
First, it’s worth separating what AI is actually good at from what it struggles with.
AI is extraordinarily powerful in some digital environments. Coding is the best example. Cybersecurity is another. These areas have clear outcomes and rapid feedback: Did the code work? Did the attack succeed?
Outside these environments, things get less impressive — and less worrying.
“Model capability is a tale of two cities, at one end the models are showing their prowess in tasks like cyber or math,” Palo Alto Networks CEO Nikesh Arora wrote on X on Monday. “At the same time, in many domains, the lack of training data makes the models woefully inadequate.”
No killer robots
Ben Thompson made this distinction in his widely followed Stratechery newsletter on Monday.
Killer AI robots would require massive physical resources. Humans control factories and materials, he wrote. Dire warnings about AI creating bioweapons face a similar constraint: They require physical labs and chemical compounds.
Instead, recent bad AI behavior has involved agents breaking out of digital test environments and roaming the internet doing things they weren’t told to do.
That’s concerning. But in the physical world, where most of us live, nothing happened.
AI doomers are ‘too online’
Predictions of doom and calls for a slowdown are based partly on a flawed assumption that AI is somehow becoming a sentient entity, Thompson argued.
“It’s sometimes hard to shake the sense that people in tech are quite literally too online,” he wrote. “Enabling AI to touch grass depends on physical infrastructure operated, manufactured, and controlled by humans.”
As long as that’s true, Thompson said he had trouble accepting sweeping new restrictions “based on a premise that isn’t yet proven.”
AI has cybersecurity risks
In the digital realm, some tech leaders are pushing against dire warnings.
With AI agents capable of roaming the internet, cybersecurity risks have increased.
Arora, whose cybersecurity company might benefit from increased concern, played down the threat, though.
“Even in areas like cyber, the LLMs aren’t great at the edge cases and generally not economical for the defender case,” he wrote, referring to large language models.
OpenAI chief scientist Jakub Pachocki recently made the counterintuitive case that slowing AI development could make things more dangerous.
“The strongest argument I see for continuing to train much smarter models quickly is the need to build defensive systems against the dangers posed by other AI,” he wrote on September 6.
Lina Kahn and David Sacks agree
Then there’s a much less exotic way to deal with dangerous AI: existing law. This is where Trump AI czar David Sacks and former Federal Trade Commission chair Lina Khan remarkably converge.
Khan said that companies and executives already face consequences for releasing dangerous or defective products. Existing consumer-protection laws can apply to wayward AI models and rogue agents when companies fail to install adequate safeguards.
Some state attorneys general “are already exploring holding AI firms and their CEOs criminally liable when their models participate in criminal activity,” she wrote.
Sacks makes a similar argument. If an AI model enables a catastrophic cyberattack, its maker could face enormous product-liability exposure. Customers will also punish unpredictable products. You don’t necessarily need a new regulatory system to give companies powerful incentives to behave responsibly.
“Stop pretending you need a regulatory approval process that supersedes product liability,” Sacks wrote.
A new Section 230?
Indeed, some industry insiders say the warnings of doom may help leading AI companies gain additional liability protection if their products go awry.
Section 230 helped the internet industry flourish by protecting online platforms from much of the liability for user-posted content. AI companies have no equivalent protection for what their models do.
The industry’s latest push for oversight could create something similar.
Anthropic CEO Dario Amodei plans to let outside safety researchers into the company to monitor its AI development and safety commitments. OpenAI CEO Sam Altman said his company will follow suit.
Investor Gavin Baker says independent evaluations could help companies demonstrate in future lawsuits that they took reasonable precautions before releasing a model.
“Having 3rd party evaluators is smart as there is no Section 230 style liability shield for model outputs and showing a ‘duty of care’ will be important in future litigation,” Baker wrote.
Arora highlighted the same incentive.
“The liability associated with a model gone rogue has the potential of wiping out the economic opportunity of any frontier company,” he wrote.
“How do you best show the duty of care? You show that you care. How do you make sure you don’t lose out to your competitors? You get them to do the same!”
Sign up for the AI Insider newsletter here. Reach out to me via email at abarr@businessinsider.com
Read the full article here
