Focusing on the risk of AI extinction fog becomes a dangerous distraction.
Last week, the Silicon Valley-funded Center for AI Safety said, “Reducing the risk of AI-induced extinction is a global priority, alongside other societal-scale risks such as pandemics and nuclear war. It should,” he issued a one-sentence statement. The document was signed by many industry leaders and respected AI researchers and received extensive press coverage.
The widespread reaction to this statement was harsh. Oxford University professor Sandra Wachter said it was just a publicity stunt. Some of the more sensible signatories, such as security expert Bruce Schneier, were quick to express their disappointment at the signatories. “I don’t really think AI poses an annihilation risk,” Schneier said.
Some thought the statement was really a fundraising ploy. Kieren Healy, a professor of sociology at Duke University, said, “My friends and I are committed to using truckloaded absolutes to mitigate the literal species-level survival threats associated with this we are arguing.” I’m going to ask for a subsidy,” he posted in a fake paraphrase.
Marieche Schaake, a former EU member of parliament and now of the Center for Cyber Policy at Stanford University, said the gist of the statement was that policymakers should address existential risks, and that business leaders should be aware of the practical rules for using AI. It was suggested that it is to set the AI may be new, but this tacit claim that AI industry leaders are “best placed to regulate the very technology they create” has been used in the past on social media and in cryptocurrency debates. It’s just a matter of reusing things, she said.
Apocalyptic warnings that AI will gain conscious and independent agency call for regulators to step up enforcement of existing laws and policymakers to consider reforms to address legal gaps In my view, it distracts from the real challenges of AI.
As Federal Trade Commission Chairman Rina Khan said, fraud conducted using AI is illegal. The agency has already warned against using AI to impersonate people to commit video and phone fraud. China has realized the same thing and is cracking down on AI-powered fraud. It’s not clear if there are any new legal issues here, but it will require significant law enforcement efforts to control the coming flood of AI scams. Fully funding the agency’s $590 million budget request would be a much more productive use of public funds than researching the existential risks of AI.
Everyone is understandably concerned about AI-generated misinformation, such as the recent “fake Putin” announcement that Russia is under attack. Labeling can go a long way in mitigating these risks. The Republican National Committee commercial, which aired following Biden’s presidential candidacy announcement, used AI to generate images of what would happen under the new president, labeled as such. The risk of false alarms has been reduced.
AI transparency is an easy policy outcome that policy makers should seize upon, as some do. This week, European Commission Vice-President Bela Djulova urged tech companies to label content generated by artificial intelligence. In the US, Rep. Richie Torres (DN.Y.) will soon introduce legislation that would require services like ChatGPT to disclose that their output was “generated by artificial intelligence.” .
Copyright and AI are also challenges. There needs to be some clarity about compensating copyright owners for the use of their material for AI training. Getty sued Stability AI in the US in February, accusing the company of unauthorized copying of 12 million Getty images to train its Stable Diffusion AI image generation software. The company last week asked a London court to block Stability AI in the UK, alleging that it infringed Getty’s copyright in training the system.
These cases will be thoroughly resolved in court. However, there is a sane argument that there is no need to compensate copyright holders at all because of fair use reasons, or because only unprotected facts and ideas are extracted for AI training. In addition, the European Union’s 2019 Digital Single Market Copyright Directive requires that online copyrights be protected unless copyright holders opt out using technical protections such as header blockers to prevent scanning. Exceptions are included to allow text and data mining of unsolicited material. This could cover AI training data.
The current draft of the European Union’s Artificial Intelligence Act mandates disclosure of copyrighted material used to train AI systems. This appears to be intended to allow copyright owners to exercise their right to opt out of text and data mining for AI training. But it could also be a step towards something more. This could lead to a compulsory licensing system that provides some compensation for the use of intellectual property while preventing copyright holders from blocking AI training. Addressing these copyright issues will require intensive attention from policy makers.
EU AI law also requires high-risk AI systems to undergo a certification process aimed at ensuring that risks have been properly assessed and reasonable mitigation measures employed. . The current draft treats underlying models like ChatGPT as dangerous systems to certify, which creates a potential burden, with Open AI chief Sam Altman urging ChatGPT to be a European It seems that he said that he would withdraw from He has since ignored the threat and said he had no intention of leaving.
But he has a point. Policy makers looking at specific problems wonder how a general-purpose AI system like ChatGPT can be qualified as “safe” when many of the risks only appear when it is applied in practice. You have to ask yourself if you can.
These are just some of the major AI issues that policymakers should be concerned about. Others include the employment impact of increasingly sophisticated AI systems, privacy concerns when training data contains personal information, the concentration tendencies created by the enormous cost and network effects of training AI software, This includes the application of Article 230 Liability Regulations and the enforcement of legal regulations. Laws prohibiting bias in lending, housing, and employment when AI drives eligibility assessments.
Policy makers need not and should not wander into foggy realms where autonomous AI programs spiral out of control and threaten human survival. They have a lot of work to do to meet the many challenges of AI in the real world.
follow me twitter Or LinkedIn.
