AI Safety Expert: 99% Chance We All Die
Happy Monday... :)
I went into this interview expecting to talk about jobs disappearing...
Roman Yampolskiy is worried about something much bigger: humanity disappearing. He believes AI could soon begin improving itself, become vastly more intelligent than us and pass a point where nobody - not governments, tech companies or its own creators - can control what happens next.
Three things I learned:
The real danger begins when AI can improve itself. Roman believes AI research could be automated within two years, potentially starting a rapid cycle of machines designing ever-smarter machines.
We can have the benefits without building a superintelligence. Narrow AI could help cure diseases, solve scientific problems and create abundance without becoming better than humans at everything.
“Just program it to protect us” isn’t a solution. Humans cannot agree on one set of values - and researchers still cannot guarantee that an advanced AI would follow our instructions once it became more intelligent than us.
Do you get value from this channel? If so, please become a paid member.
The biggest risk AI poses will be its effect on the stock market when the reality turns out to be quite a lot less than the hype.
AI will certainly have an impact however from what I can see, it is most useful in turbo charging the productivity of people who are already highly skilled by taking care of donkey work and brining together information in novel ways. It may also end up replacing a lot of driving jobs and low skill jobs jobs that involve collating information.
Where it falls short is a conspicuous lack of common sense. It hallucinates because it's not actually intelligent it's essentially a sophisticated predictive text algorithm which really doesn't have anything in common with what is conventionally understood as 'intelligence'.
The idea that it will take over the world just seems like one more silly thing to get hysterical over. The fact is 100% we are all going to die - just not from AI.