Susan Athey: How AI Can Aid in U.S. Economic Recovery

On September 10, Susan Athey, Stanford GSB Economics of Technology Professor and Institute for Human-Centered AI associate director, testified before the U.S. Congress’s House Budget Committee as an expert witness for a hearing titled “Machines, Artificial Intelligence, & the Workforce: Recovering & Readying Our Economy for the Future.”

Athey helped House members understand how to help the U.S. economy recover from the ongoing COVID pandemic and prepare for a brighter future and the key role of AI in this effort.

“AI has enormous positive potential for society,” the professor said in her opening remarks, noting the value of AI-driven solutions related to education, training, remote work, government services, medicine, and other areas.

Here are several key takeaways from her testimony (or read her full report to Congress). 

Digitization Means Cheaper AI

It is becoming cheaper to provide AI-based solutions in multiple domains, from supporting caregivers to helping rural residents find employment opportunities beyond their local communities.

“Digitization of everything from consumer interactions to supply chains has enabled creation of data that optimizes how AI is implemented,” Athey said. “Cloud computing now allows companies to rent computing as they need it. Software-as-a-service lets companies subscribe to the best products for their needs on a case-by-case basis.”

The proliferation of AI and machine-learning innovations, along with expansion of open-source software and free data-analytics tools, means firms don’t have to create technology solutions in-house through R&D, freeing their resources for other value-generating efforts.

Beware the “Black Box”

Athey noted that much of the AI-driven technology deployed over the past 15 years could be thought of as “automation on steroids,” where software follows pre-specified rules (created by humans) without human direction, such as chatbot-based phone systems.

But more cutting-edge machine-learning innovations generate decision rules learned from past data. “Analysts can just feed in raw data and the algorithm decides what’s important for the task,” she said.

While that might mean a wider array of general-purpose applications, from retail recommendations to medical-outcome predictions, it also means more black-box processes that even the engineers building them might not fully understand. For example, factors predicting loan default may shift due to COVID, and the algorithm may not take this into account. We need more research into best practices to ensure safe, fair implementation of such technology and safeguard against unintended consequences.

Think Beyond the Bottom Line

“Businesses may be indifferent between a worker and a machine from a cost perspective,” Athey said. “And if they’re indifferent, they’ll go with the machine” to protect their bottom line.

She cautioned that as a society we “can’t always count on companies to take the longer-term perspective.”

The potential solution lies in developing a national innovation and R&D strategy that focuses more on augmentation of human workers over replacement. “If universities or the government invest in AI that augments humans,” Athey said, “we can diffuse that more broadly across sectors — but we have to be intentional about it.” 

Where to Place Displaced Workers

Historically, we haven’t done a great job of dealing with displaced workers, Athey noted — a critical issue as automation proliferates. The number of bank-teller jobs dropped 26 percent in the past decade, for example.

“We have a lot more tools now to use data to figure out what the next best step for a worker is,” said Athey. “Like what types of upskilling will work for a person in a given circumstance.”

Greater confidence about future employment prospects, in turn, will enhance people’s motivation to retrain for more sustainable, fulfilling work. Athey is working with Rhode Island’s government, for example, to improve data to evaluate job-training programs and provide workers better information about training opportunities.

Toward a Longer, Happier (Work) Life

With mounting questions about the long-term viability of social-entitlement programs like Medicare, it’s critical to provide for the U.S.’s senior population, whether working or not.

“AI and automation can help people live independent, fulfilling lives,” Athey said. “Physical and cognitive challenges, including on the job, can be alleviated through augmenting AI or physical robots,” for example. That might help people work longer.

Moreover, she noted that much of current service work may not be replaced by automation, sustaining employment opportunities for seniors and enabling them to contribute to society in meaningful ways should they choose to.

The Promise of AI in Medicine

The U.S.’s early handling of the COVID crisis made clear we can do better in harnessing AI-driven solutions for medicine and health care more broadly, as Athey noted: “Using AI and machine learning to illustrate what treatments work best was very limited in the U.S. due to our disjointed medical system.” For example, leaders didn’t incorporate data from multiple sources: insurance companies didn’t get patient-related data until bills were sent out, there was little analysis of information spanning multiple medical centers, and we didn’t take a coordinated, data-driven approach to early clinical trials.

“AI and machine learning can only do their work if they're given the opportunity to access data and really influence decisions,” said Athey.

The good news is that AI, used strategically, can help us take a more effective approach to current and future medical challenges, including the ongoing COVID battle.

Interested in HAI policy reports and information? Sign up for our policy email newsletter.  

Stanford HAI's mission is to advance AI research, education, policy and practice to improve the human condition. Learn more.