A few days ago, Bill Gates published a blog post about his concerns with artificial intelligence (AI).  He singles out three big risks: job losses, AI as a tool for harm, and the possibility that AI will stunt kids’ mental abilities.

Gates makes the assumption that software developers will be able to solve AI’s unreliability problems. I don’t share that view, at least not as long as AI continues to depend on “large language models”—see my previous coverage of the topic. But Gates believes that “the reliability problem is being fixed quickly, as researchers create models that can check their own work and improve themselves.”

Even today’s approach to AI, unreliable as it is, could eliminate some jobs. And Gates’ other two big worries—AI use for intentional harm and its impact on growing brains—are already starting to be visible.

At some point AI, perhaps in a new form, probably will become reliable, and then it will be able to eliminate far more jobs. So I don’t fundamentally disagree with Gates’ concerns.

Read more: Bill Gates is getting alarmed about artificial intelligence

Loss of jobs. Gates lists law, customer service, medicine, software, and manufacturing as industries that will see job losses, especially for entry-level workers. And the rapid advances in the capacity of AI may mean that this change happens very quickly (unlike the changes brought about by technologies such as the automobile and the computer, which took many years to actually impact employment). Gates also thinks robots will take over many routine manual jobs (more slowly, but beginning “by the end of the decade”).

Information for harm. AI makes it easier to find out how to make bombs, bioweapons, and computer viruses. Gates also lists fraud, disinformation, deepfakes, and surveillance as additional problems that AI can aggravate. Attacks on infrastructure, such as the computer systems of hospitals and power grids, are far easier when AI is used. And AI in the hands of the powerful can make them even more powerful than they already are.

Stunting educational and emotional development. A supportive AI companion can make it unnecessary for a child to deal with real people in the real world. Gates himself lacked social abilities as a child, and if AI had been available he might never have developed them. In education, AI can lead children to assume that all the answers they ever need can be gotten from the computer. There are signs that critical thinking is negatively impacted by heavy AI use.

It’s not all bad news, but we need a plan. Gates is optimistic that AI will provide positive outcomes in some situations. He sees important benefits in medical care. He describes how farmers around the world will be able to benefit from AI, increasing the amount of food they can produce. He sees important opportunities in government and in education. But if AI is to be a net benefit, steps need to be taken to direct and restrict its use.

The companies involved in AI can’t be counted on to implement the necessary restrictions. Rather, “solutions should be developed through a public democratic process that includes elected officials, policymakers, educators, health workers, local officials, and community leaders.” Both national and international organizations will be needed to manage our transition to AI. “Building the framework I’m talking about will take years”, Gates writes, “which is why we need to start now.”

Gates also notes that we will probably want to establish certain “human reserved” categories of activity that AI and robots should not handle. This might include some kinds of caregiving, perhaps some aspects of education and medical interactions, and so on.

How should AI be taxed? Gates thinks the current way that investments in AI and robotics are counted as capital investments can be unfair to the people who are currently doing the jobs that will be eliminated. Because of the difference between how capital is taxed and how labor is taxed, our system favors capital investments over labor.  For years, Gates has been a proponent of a “robot tax” whose revenues would be used to cushion the impact on workers.

Gates (along with the Gates Foundation) is working on developing these ideas leading to a universal plan for the implementation of AI. It is an almost overwhelmingly difficult project, but a necessary one. It needs our support.

In the future, Gates promises to write in more detail on all these issues. If you want to read his posts when they come out, follow the link in the first paragraph of this blog post to his current newsletter article, scroll to the bottom, and click on “Get my newsletter”. You’ll get an email each time he posts.