Tech is always both good and bad. But we live in a time when everything gets weaponized—ideas, images, ancient texts, biases, and even people. And technology provides the tools to

Human biases have already been trained into neural networks, but that might seem trivial compared to what could happen. A computer scientist with bad intentions could introduce dangerous possibilities. According to data scientist and founder Rand Hindi, it might be possible for a bad actor to insert images into the training data used for autonomous driving systems—which could lead, for instance, to the AI deciding a crowded sidewalk is a good place to drive.

The bigger fear is that neural networks, given enough compute power, can learn from data far faster than humans can. Not only can they make inferences faster than the human brain, but they’re far more scalable. Hundreds of machines can work together on the same complex problem. By comparison, the way humans share information with each other is woefully slow and bandwidth-constrained. Big tech companies are already working on “generative” neural networks that process mountains of data to create completely new and novel outputs, like chatbots that can carry on conversations with humans, or original musical compositions.

Where this is all leading, and whether humans can keep up, is a subject for debate. Musk believes that as AIs begin to learn and reason at larger and larger scale, an “intelligence” may develop somewhere deep within the layers of the network. “The thing that is the most dangerous—and it is the hardest to . . . get your arms around because it is not a physical thing—is a deep intelligence in the network,” Musk said during a July speech to the National Governors Association.

The kind of sentience that Musk describes does not presently exist, and we’re probably decades away from it. But most experts believe it’s coming in this century. According to the aggregate response of 352 AI researchers in a 2016 survey, AI is projected to have a 50% chance of exceeding human capability in all tasks in 45 years.

Just the beginning

These examples are just the most sensational of the tech threats facing us today and in the future. There are many other near-term threats to worry about. In many ways, our technology, and our technology companies, are still a threat to the environment. Some of our biggest tech companies, like Seagate, Intel, and the Chinese company Hikvision, the world’s largest surveillance camera vendor, are enabling a growing tide of surveillance around the world. The ad-tech industry has normalized the destruction of personal privacy online. The U.S. government is sitting on its hands when it comes to securing the voting technology that will be used in the 2020 election.

It’s going to take a much improved partnership between the tech community and government regulators to ensure we stay on the good side of our most promising technology.