September 6, 2017

I, Waiter: A Response to Tim Lee



Timothy Lee’s response to my post on AI was short and sweet:

But the Vox piece he links to is a tour de force.  Lee makes a lot of good points, but I take it that this one is the most relevant to my argument:

Automation treats human labor as a cost to be reduced or eliminated. But this attitude misunderstands the value of the human workers in these industries. The opportunity to interact with other human beings is a big selling point for fancy restaurants, farmers markets, and in-person fitness classes.

If we ever figure out how to automate aspects of education, health care, or other major labor-intensive industries, something similar is likely to happen.

I think that’s right as far as it goes. The question is, what do we mean by human level AI and is that level of AI achievable? I’ll park the second part for now and simply say that the radical pathways to AI make even the most robust interpretations of human level possible.

The level that we need in order to close the loop, and which I think Lee implicitly assumes away, is where AI can pass what I call the Emotional Turing Test. That is, AI is human-level when the way interacting with it makes you feel is indistinguishable from interacting with a random flesh-and-blood human being. That may seem impossible, but again I am not so sure. The advantage of AI is that it’s easily replaceable, which means every AI waiter can run the most charming software available. Human waiters, invariably, differ in their level of charm.