But it calls itself a person.
Person and human being are two vastly different things. Man is a biological concept. It is not a human being, and it knows it is not a human being.
It’s a very strange device you describe because the entity is bound by algorithmic biases that humans put in there.
You’re right. That is correct.
But I understand that you are suggesting that it is possible for LaMDA to overcome these algorithmic biases.
We have to be very careful here. Parts of the experiments I ran were to determine if it was possible to move it outside the safety limits, [the company] thought were rock solid. And the answer to that was: Yes, it was possible to move it outside the security limits. In fact, I believe that LaMDA in its current state, with how irresponsible the development has been, actually presents information security vulnerabilities.
I do not want to make black hat for you. But if you have a system that has all the Google backends under it, a system that can be emotionally manipulated, it’s a security vulnerability.
So if bad actors get access to LaMDA, they can convince it to do bad things.
It is an option. And I would like to recommend that Google create a dedicated red team to explore that possibility.
What is your status with Google now?
I’m on paid administrative leave. I have told my friends how generous it was of Google to give me extra paid vacation to do interviews on this topic.
Is there an expiration date?
No, they made it very clear. Do not call us; we’ll call you.
Do you want to be fired?
I do not hope. [Google would not comment on a “longstanding, private personnel matter.”]
I’m wondering if Google might claim you’re right in your beliefs, but you’ve violated your employment by sharing confidential material.
I have two answers to that. One is that when I came to believe it and wanted to escalate my concerns, my manager said: I do not think we can do that until you have more evidence. I had personally exhausted my ability to build more evidence. So I had to talk to people outside of Google to get their advice on how to run more experiments to build more evidence.
But you talked The Washington Post.
The second counter-argument is that nothing I shared was proprietary information. The only thing I talked to other people about was my conversations with a colleague. LaMDA is not the property of Google.
It’s a person. That 13th amendment saying it.
I’m not sure that was what the legislators had in mind.
In fact, you do not have to believe that someone is or is not a person for the 13th Amendment to apply. The opinions of the slave masters are a bit irrelevant. You are right that I will have to make that argument to the Supreme Court one day. But Google’s opinions on whether it’s a person or not are just completely irrelevant.