Most economists oppose President Trump’s trade policies. As I recently argued on this page, it’s not just that he won’t help those hurt by trade. Now, with his sweeping tariffs and escalating trade war, he’s also going to hurt those helped by it.

Just because he’s got the solution wrong doesn’t mean there isn’t a problem. Under the assumption that the political pendulum will eventually swing back and very different policymakers will be in charge, what is the policy set that would tap the benefits of globalization and ameliorate its downside?

One key plank in that platform should be to help American manufacturing. I’ve explained why I don’t think the tariffs will help, but I also reject the pervasive argument that automation is just too powerful a job-killing force such that there’s no point in trying to help our factory sector. In fact, according to an important new paper by economist Susan N. Houseman, the argument that “labor-displacing technology,” i.e., robots/AI/etc., has killed factory employment is not supported by the data.

Most people think American manufacturing employment has been falling for generations, but that’s only true as a share of employment. In absolute numbers, the figure below shows that jobs in the factory sector held steady at around 17 million from around 1970 to 2000. They then cratered, falling by about 5 million jobs, peak to trough.

The share of manufacturing jobs has been falling for decades, as government and private service employment have flourished. But what happened in 2000 that took such a hard whack at the absolute number of manufacturing jobs?

The evidence points more toward trade than technology. China joined the World Trade Organization in 2001, and its exports to us sharply increased (trade researchers label this period the “China shock.”) Our trade deficit as a share of GDP averaged minus-4.3 percent of GDP in that decade, a bigger negative than in any other decade on record, with data back to the 1930s.

Yet most economists argue that what’s really killed jobs in the factory sector is not global competition with low-wage, “mercantilist” countries pursuing export-driven growth. It’s automation-driven productivity.

This assertion is based on the view that output in the sector has grown quickly while employment has tanked, implying faster-than-average growing productivity, or output-per-hour. But Houseman shows that when you pull out one unique sector — computers and electronic products — which accounts for just 13 percent of manufacturing output, the sector has significantly lagged behind the rest of the economy. Without computer production, real “GDP growth in manufacturing was less than half that of the private sector average from 1979 to 2000, and only 12 percent in the 2000s.”

The reason computers, a relatively small subsector, make such an outsized contribution to overall manufacturing output is because our statistical agencies credit computer production with large quality improvements, which show up as lower prices and more real output. Such adjustments are justified, to be sure — this isn’t a mismeasurement argument. But these price declines have everything to do with ever-faster semiconductors and little to do with automating the production process.

That is, it doesn’t take fewer workers to install faster chips in a new laptop. Automation in computer production drives up GDP growth in the factory sector, but it doesn’t have much impact on jobs. The rest of manufacturing, as Houseman’s key figure shows below, has lagged behind the rest of the economy. It’s a picture that stands in stark contrast to the myth of a cutting-edge factory sector generating tons of output with ever-fewer workers.

Stepping back from these sectoral nuances, it never made much sense to blame productivity for manufacturing job losses, at least not without invoking other factors in play, most notably globalization and import substitution. After all, a more productive domestic factory sector should lead to lower prices and greater demand for its output. In fact, it’s the intervening variable of demand that explains why productivity that consistently grows over time can coexist with low unemployment. But by increasingly meeting our demands for manufactured goods through imports, we’ve jammed that channel.

To be clear, that’s neither a terrible nor unusual outcome — it’s one of the benefits of expanded trade and a common dynamic in advanced economies, as they shift from goods to service-based industries. But a) it’s far from costless, and b) it’s not a function of automation. Moreover, U.S. manufacturing job losses have been a lot larger (adjusting for the size and age of the workforce) than those of many of our competitors.

But if tariffs won’t help, what will?

For one, getting the exchange rates right. Our trade deals feature chapter upon chapter protecting investors and patents but have never produced rules against our competitors artificially suppressing the value of their currency to make their exports cheaper.

Regional economist Tim Bartik looked for common practice across communities that achieved better-than-average manufacturing job growth. He identified customized services to help smaller manufacturers overcome financing and information barriers, improve their technology and product design, and link up to global supply chains; infrastructure and land-use investment to make areas more attractive for business development; and life-cycle skill development, including high-quality child care, high-quality preschool, K-12 education, college scholarships, and adult job training.

Finally, I agree with where I think Larry Summers is coming from here. You want to go after China’s forced technology transfers, fine, but I doubt that does much for U.S. workers or consumers. What’s more worrisome is the extent to which they’re investing in AI, robotics, STEM education, and cutting-edge consumer technologies, while we squander our money and time on wasteful tax cuts and highly dysfunctional politics.

I’d put these investments high on the policy to-do list. Not only does manufacturing provide high value-added jobs, often in places that need them, but it punches above its weight in R&D and has large positive spillovers to other sectors.

If some robot tries to tell you that’s a waste of resources, ignore her.