Time is running out to tackle the dangers posed by AI

00:00 06:48

Mitchell Reiss has narrated this article for you to listen to.

Is this what it felt like in the months before August 1914? Or during the years leading up to September 1939? The discussion around artificial intelligence produces a deep foreboding that we are in the grip of forces largely beyond our control. Are we sleepwalking towards disaster?

That is the feeling I have after reading Genesis, a collaboration by Eric Schmidt, the former CEO of Google, Craig Mundie, the former chief research and strategy officer at Microsoft, and Henry Kissinger, who died, aged 100, soon after completing this book. They have crafted a holistic analysis of the social, political, psychological and even spiritual impacts that a superior machine intelligence would have for humanity. 

We are broadly familiar with AI’s current and future benefits. These machine tools can process massive amounts of data at unnerving speeds. They can select their own goals, learn from their errors, upgrade their algorithms and design things that no human has ever previously imagined. Some experts predict the machines may soon achieve sentience, demonstrating the elements of human consciousness: memory, imagination and self-awareness. 

We already see AI’s impact across business and in medicine, especially with cancer screenings, drug development and clinical trials. AI is performing human tasks, such as booking holidays, deciding mortgage eligibility and helping determine criminal justice decisions. Enthusiastic techno-optimists gush that it may even help find solutions for intractable global problems such as climate change, the transition to clean energy, global poverty and conflict between nations. 

While the three authors celebrate these developments, they emphasise that there is no instruction manual telling us how to develop AI safely – one that ensures that it serves humanity and does not subordinate it. Many downsides are already evident. AI will disrupt job markets, causing unemployment especially among less skilled white- collar workers. It is also turbo-charging the spread of disinformation, with more than 70 countries already using this technology to undermine democratic institutions and civic cohesion. 

But, Genesis argues, we are running far larger risks that speak to who we are as a species and what it means to be human. If we are approaching the biological limit of our intelligence and are about to be outpaced by AI, what will it mean to share the planet with more intelligent beings? Would we forfeit control over economic decision making?  Over the exercise of the political process? Over the choice to wage war or negotiate peace? And would AI even give us the options? 

The authors realistically accept that research on AI will not stop, given the outsized financial rewards, ego and power at stake. Speed and secrecy are being privileged over safety. There is no consensus on what constitutes unacceptable risk. Laws and regulations cannot keep up with technological advances. Accidents, errors and unintended consequences in developing and applying the technology seem inevitable. There is currently no way to ensure that AI, especially when married to advances in quantum computing and synthetic biology, will not be developed for malevolent ends such as cyber attacks, automated war or engineered pandemics. Further, the competitive nature of the international system means that states will speed ahead as fast as possible, driven by the fear that second place could mean perpetual servitude. 

Genesis is at its most interesting when it imagines how states might overcome their mutual hostility and suspicion and co-operate to control this new force. One proposal is to create a supranational entity that could license AI production facilities, refine datasets and regulate operations. But even if this organisation could somehow be created, would members automatically receive the benefits of AI if they wanted to use the technology to suppress dissent and deny human rights? How would the organisation ensure compliance and punish violators? 

Speed and secrecy are being privileged over safety. There is no consensus on what constitutes unacceptable risk

A second proposal is for the United States and China to collaborate in forging a protocol for jointly managing AI’s perils. The two sides held AI meetings this past year, all unproductive. In November, President Biden and President Xi Jinping agreed that there must always be a human in the nuclear chain of command, meaning that decision-making should not be delegated to an AI machine. 

These efforts fall miles short of what is needed to address the catastrophic risks AI poses; yet it is difficult to envision much more progress in the coming years. The US and China are locked in strategic competition across multiple domains, including AI; indeed, the US is trying to slow China’s development by denying it the most advanced AI chips. And even if a bilateral condominium could miraculously be achieved, would other states agree to accept a subordinate position in perpetuity? 

The authors see humanity being at a ‘hinge point of history’, hence the title Genesis, but they don’t seem optimistic. They warn that we have a decade ‘at most’ to get things right and concede that creating a balance and equilibrium among competing states would require ‘a Herculean effort’. 

An abiding focus of Kissinger’s life’s work was the intersection between technology and public policy, starting with his writings in the 1950s on nuclear weapons. He knew that technology alone could not overcome mistrust among states, curb human ambition or eliminate bad actors. Yet, through a combination of thoughtful action and luck, a nuclear holocaust was averted.  Kissinger devoted the final years of his life to bringing China and America together to negotiate a pathway forward on AI. His last book stands as both an impassioned warning and an urgent challenge.