Rebecca Gorman | This House Believes Artificial Intelligence Is An Existential Threat | CUS

Her points: 1. AI does not currently always do what we intend it to do. Her example is that AI applications meant to add entertainment to social media, instead cause death in teenagers. 2. Ai systems have been deployed at scale for 20 years. Some people have benefitted, some have been harmed. 3. Machine learning. Doesn't work in edge cases. 4. Example if we allow AI to decide when to fire nuclear missiles. 5. It is not built for the purposes to which it is being deployed. 6. The key is to build it such that it follows our instructions and then give it good instructions.

Can someone please explain how these systems kill teenagers? Hello?

Now her we can listen to!

No, the point Is not that AI is safe if it does exactly what we asked it to do. The point is that we do not know what AI understood about our values and intentions. So AI, if instructed to do so, could solve the world hunger, but at the cost of something else we did not expect.
The problem is that we can't know "the reasoning" behind the AI choices. So we can't know if the reasoning of the AI is flawed when AI is more intelligent than us.
Example? In chess, you will see AI doing very bad moves. But you consider them bad just because you're not smart enough to see the full picture, while AI is.
In a similar way, we will give autonomous powers to AI, but we will not be able to be sure that the final results of what we will ask will not carry a threat for humanity.

Can't say she put my mind at ease.

9 billion people on the planet, it only takes 1 person to make a mistake, at best..guaranteed someone will use it for their own goals, and within hours of being released

Poor