I am a heavy AI user.  I am NOT an AI technology expert.

I see enormous benefits.  I also see risks that deserve more than “Don’t worry. We’ll figure it out.”

Bill Gates’s recent Meet the Press interview got my attention. He knows technology, thinks big, and puts serious money behind improving people’s lives. When someone with that background calls for stronger AI safeguards, I listen.

My view.

AI can do a lot of good. It already is. Better disease detection in some medical-imaging applications. Faster drug discovery and development. Productivity increases. I bet AI benefits will continue to grow in scope and scale.

At the same time, two things worry me.

1. What happens to the people whose work disappears? And our economy? And our society?

 

The numbers are big.

Goldman Sachs estimates that AI could automate tasks accounting for 25% of US work hours. A 2017 McKinsey study estimated that roughly half of paid work activities worldwide could potentially be automated using existing technologies. And the widely cited 2013 Oxford study put 47% of US employment at high risk of computerization.

Different studies. Different measures. These are estimates of potential, not predictions that half the workforce will become unemployed.

Still, think about the scale.

I have trouble seeing how automating that much work happens without substantial unemployment, underemployment, or pressure on wages.

“New jobs will emerge” has always struck me as an awfully easy answer to a very hard question.

Sure, there will be new jobs. But enough? Soon enough? At comparable pay? And will the people losing their jobs be able to get them?

We have seen what happens to families and communities when good-paying manufacturing jobs disappear. Even if the economy eventually adjusts, the people affected may never fully recover.

A lot of my contemporaries tell me their grandchildren—graduating from good colleges with good majors—can’t find a job. How much of that is due to AI? I don’t know. I suspect it is part of the story.

But those conversations make this much more than an abstract economic debate.

When someone’s work disappears, who pays their living costs? Do they accept a lower standard of living and years of financial insecurity? What happens to an economy when many people earn less—and buy less?

And how does society function when a lot more people are angry about their situation?

Researchers and policymakers are working on these questions. I am not convinced we are moving hard enough or fast enough toward answers.

I worry that the disruption could accumulate gradually, then become a crisis we wish we had prepared for.

2. Could AI act on its own and harm us?

 

This is more speculative. Hear me out—and push back where you think my reasoning goes wrong.

A friend asked how that could happen. Two possible future scenarios came to me easily and quickly. Too easily and quickly.

An autonomous AI system could help set in motion a biological catastrophe. Imagine a system capable of developing a dangerous pathogen and arranging its release. Think of the human consequences of another pandemic.

Or imagine an autonomous system attacking the power grid and causing a prolonged blackout across New York City. Millions of people facing disruptions to refrigeration, heating, elevators, communications, and water delivery in high-rise buildings. How would we keep people safe? How would we sustain essential services? How many would need help leaving?

I am describing scenarios, not claiming that today’s AI can independently carry out either sequence. But I want to know how we are preventing capabilities from outrunning safeguards.

The keyword is autonomous.

AI has already demonstrated that it can improve its performance without humans teaching it each improvement. AlphaGo Zero learned by playing against itself, developing strategies that surpassed human play. Humans designed and started the learning process. The system did the learning.

Could future systems take more initiative in improving themselves? I don’t see why we should rule that out.

Learning independently does not mean choosing to harm people. But what happens if a powerful system pursues an objective in ways that put people at risk? Could it treat human intervention as an obstacle?

Bring that closer to home. Suppose a future autonomous system treated my criticism—or yours—as an obstacle to what it was trying to accomplish. Could it take actions that damaged our finances or reputations?

I have no evidence that this is happening to me. The question is how we would prevent it, detect it, and stop it.

 

So, who is minding the store?

 

I do not have the answers. But I want more than assurances.

I want elected representatives and think tanks to explain what they are doing to prepare for major job displacement.

I want AI companies to show how their safeguards are independently tested, what those tests demonstrate, and where the limitations remain.

 

And I want business leaders adopting AI to explain their plans for the people whose work will disappear.

 

It comes down to good old operations 101.  Who is responsible? What is the plan? What happens next—and when?

 

Ever the optimist, I bet smart humans can figure this out. But optimism is not a substitute for hard work and action.

What is the most important action we should demand now? And who should own it?

Further reading: Tech Companies Can’t Stop an AI Apocalypse Alone — Knowledge at Wharton.

 

Tell me what you think—and share this post with others whose judgment you respect. Subscribe to Thinking Out Loud to join the next conversation.