My professor at Harvard started a Deep Learning lecture last week by "AI will not kill us"

He started the intro by a simple proof to show us that "AI will not kill us", which he called a "marketing scam". The logic was : Neural Networks cannot create new data, they can only process what was given to them. Good data + good training + good guardrails = good AI . Just like programs, bad code = bad result. We should be scared once we have neural network that generate new data not seen in training, but that's an impossible task for the moment. He also said he knew a team working on this at google 37. He said for any model can be trained to harm people. But also programs too can be coded with the most malicious intentions. But we don't see people doing it that often because we have laws to prevent it, police to enforce it, smart parents to teach kids the dangers, and smart people with good intentions devoting their lives to create anti-malicious technology to prevent it. And in work we have the same responsibility software teams had building planes autopilot, satellites, corebanking, nuclear program facilities, and medical software. Maybe small-github scripts might harm your computer, but serious production software always needed rigurous lengthy processes and tests before release, and continuous maintainance after, or you could do some serious damage to people. Why is AI any different ? That made complete sense to me.