Thursday, June 19

Forget superintelligence – we need to tackle ‘stupid’ AI first

New Scientist. Science news and long reads from expert journalists, covering developments in science, technology, health and the environment on the website and the magazine.

Should politicians ensure that AI helps us colonise the galaxy, or protect people from the overreach of big tech? The former sounds more fun, but it shouldn’t be the priority.

Among the Silicon Valley set, superintelligent AI is viewed as a rapidly approaching inevitability, with tech CEOs promising that the 2030s will see a golden era of progress. That attitude has reached Westminster and Washington, with think tanks telling politicians to be ready to harness the power of incoming AI and the Trump administration backing OpenAI’s $500 billion initiative for ultrapowerful AI data centres.

It all sounds exciting, but as the great and the good dream of superintelligence, what we might call “stupid intelligence” is causing problems in the here and now. One of the questions facing the AI sector is whether hoovering up vast swathes of the internet – a necessary part of training AI – is copyright infringement.

There are reasonable arguments on both sides. Proponents say that, just as you aren’t infringing New Scientist‘s copyright by merely reading these words, AI learning should be treated the same. Detractors, meanwhile, now include entertainment giants Disney and Universal, which are suing the AI firm Midjourney for reproducing images of everything from Darth Vader to the Minions. Only legislation can settle the matter.

We are heading towards a world in which machines could kill with little human oversight

The battlefields of Ukraine pose another thorny AI problem. While OpenAI’s Sam Altman has said he fears a superintelligent AI may one day kill us all, deadly stupid intelligence is already here. The Russia-Ukraine war is driving us towards a world in which, very soon, machines could kill with little human oversight.

Politicians have entirely failed to get to grips with this threat. The United Nations held its first meeting on regulating “killer robots” in 2014. A decade later, we are no closer to restricting their use. If our leaders are biding their time in the hope that a superintelligence will eventually solve their problems for them, they are very much mistaken.

Topics:

Article source: https://www.newscientist.com/article/mg26635482-700-forget-superintelligence-we-need-to-tackle-stupid-ai-first/?utm_campaign=RSS%7CNSNS&utm_source=NSNS&utm_medium=RSS&utm_content=technology