T.The past few months have been the most exciting in my 17 years working with artificial intelligence. Among many other advancements, OpenAI’s ChatGPT (a type of AI known as large-scale language models) broke his record in January, becoming the fastest-growing consumer application in history, with user numbers in two months. We have reached 100 million people.
No one knows what will happen next with AI. Too much is happening in the public eye, too much behind closed doors. But we do know that AI is now in the hands of the world, and that it is likely to change the world as a result.
Such transformative potential is due to the fact that AI is a general-purpose technology that is both adaptive and autonomous, containing some of the magic that led humanity to reshape the planet.
AI is one of the few practical technologies that has the potential to redesign entire economies to achieve net zero. For example, my collaborators and I are using AI to predict intermittent renewable energy sources (solar, tidal currents, wind, etc.) and to optimize the placement of electric vehicle chargers for equitable access. or better manage and control the battery.
But even if AI brings significant economic benefits, some people may suffer losses. AI is now being used to automate some of the jobs of copywriters, software engineers and even fashion models (economists Karl Frey and I said in 2013 that 98% of his I presumed it could be automated with probability).
An OpenAI paper estimates that nearly one in five U.S. workers could have half of their tasks automated by a large language model. Of course, AI is likely to create jobs as well, but many workers may still face instability and wage cuts. London taxi drivers, for example, experienced wage cuts of around 10% after the introduction of Uber.
AI also provides alarming new tools for propaganda. According to Amnesty International, Meta’s algorithm played a major role in the atrocities committed by the Myanmar military against the Rohingya people in 2017 by facilitating hate speech. Can our democracy resist a torrent of targeted disinformation?
Today, AI is mysterious, unreliable, difficult to maneuver, flawed to cause harm, and potentially harmful. AI is already responsible for wrongful arrests (such as the false accusation of Michael Williams by the AI policing program ShotSpotter), sexist hiring algorithms (Amazon was forced to make concessions in 2018), and the destruction of thousands of lives ( Dutch). Tax authorities falsely accused thousands of people, often ethnic minorities, of benefit fraud.)
Perhaps most concerning is that AI may threaten our survival as a species. In a 2022 survey (possibly with selection bias), 48% of AI researchers believed AI had a significant (10% or more) chance of wiping out humans. First, rapid and uncertain advances in AI could threaten the balance of world peace. For example, proving that AI-equipped underwater drones can locate nuclear submarines might lead military powers to believe they can launch a successful nuclear preemptive strike.
If you think AI can’t be smart enough to conquer the world, remember that the world was just conquered by the coronavirus. That is, enough people were sufficiently aligned with their concerns about the apparently harmful pathogen that caused Sars (e.g., “I have to go to work with this cough. , otherwise I would not be able to support my family”, etc.). -CoV-2 he kills 20 million and disables tens of millions more. This means that AIs that are considered alien species may initially oppress or even exclude humanity by working within existing institutions.
For example, an AI takeover begins with multinationals using their data and AI to find loopholes in rules, exploit workers, deceive consumers, gain political influence, and ultimately may come to such an extent that the whole world seems to be under the control of its bureaucratic machinery. -like power.
What can be done about all these risks? New and bold governance strategies are needed to address the risks and maximize the potential benefits of AI. For example, we want to make sure that not only the largest companies can bear the complex regulatory burden. Current efforts on AI governance are either too light (like the UK’s regulatory approach) or are already two years in the making (like the EU’s AI law, the time it took ChatGPT to reach 100 million users). 8 times) is too slow.
We need mechanisms for international cooperation to develop common principles and standards and prevent a ‘race to the bottom’. We need to recognize that AI involves different technologies and requires different rules. Above all, we may not know exactly what will happen next with AI, but we must start taking appropriate precautions now.
-
Michael Osborne is Professor of Machine Learning at the University of Oxford and co-founder of Mind Foundry.
