The dangers of AI over the next 5-10 years. According to ChatGPT;

If I rank them by a combination of likelihood and potential damage over the next 5–10 years, my top three are these:

AI in the hands of bad people. This worries me more than AI spontaneously “turning evil.” AI is rapidly lowering the skill and manpower needed for cyberattacks, fraud, surveillance and potentially biological or chemical threats. A relatively small group could eventually do things that once required a large intelligence service or technical organization. NIST already identifies AI-enabled cyberattacks as a serious risk, while frontier AI labs are explicitly testing for cyber and biological capabilities. (NIST)

Giving AI too much authority before we truly understand its weaknesses. Imagine AI agents controlling portions of the electrical grid, financial markets, military logistics, factories, hospitals or large computer networks. The danger doesn't require consciousness. An extremely capable system given the wrong objective, bad information or excessive access could make thousands of decisions before a human realizes something is wrong. The more society connects AI directly to the physical and digital world, the greater this risk becomes. NIST is already developing specific guidance for AI used in critical infrastructure, and frontier labs explicitly treat “loss of control” as a serious advanced-AI risk. (NIST)

Humans becoming unable to tell what—or whom—to trust. This one may affect ordinary people the most. Within five years, realistic AI-generated voices, video, photographs, documents and online personalities could become essentially indistinguishable from genuine ones in everyday circumstances. Combine that with automated scams and millions of AI accounts capable of interacting individually with people, and our traditional method of saying “I saw it with my own eyes” stops being reliable. NIST already lists misinformation, synthetic content and related harms among major generative-AI risks. (NIST)

There's a fourth danger very close behind them: economic displacement. I don't believe AI simply eliminates all employment. But I do think the transition could be far faster than previous technological revolutions. AI isn't replacing only physical labor; it's moving directly into accounting, programming, legal work, medicine, engineering, customer service, administration and management. NIST specifically notes potential labor-market and economic effects from widespread generative AI. (NIST Publications)

And this gets back to Bill Gates's billion-death statement.

I don't think the most probable AI catastrophe is AI waking up one morning and deciding humanity has to go.

I think the more believable progression is:

Humans build increasingly powerful AI → humans connect it to increasingly important systems → someone misuses it or trusts it too much → AI allows the consequences to occur at a scale and speed humans couldn't previously achieve.

That distinction matters enormously.

AI itself is also potentially one of the greatest problem-solving technologies humanity has ever developed. The same capability that can discover a computer vulnerability can find it before an attacker does. The same AI capable of assisting biological research can help develop medicines. The same automation that eliminates some jobs can produce enormous increases in productivity.

So I don't come away from this thinking “we should stop AI.”

I come away thinking we'd better be extremely careful about what authority we give it, who gets access to its most dangerous capabilities, and whether a human can still pull the plug.

And Wayne, there is one AI development that concerns me considerably more than today's ChatGPT-type systems: AI that can independently improve the next generation of AI. That's where the 5–10 year discussion gets much more serious.