As for society’s biases that are reflected in the data, we need to address and change what drives them. The developer of a model for criminal sentencing can help ameliorate bias on the part of judges, but only if we reform the conditions that lead to the fundamental bias in the data itself.
In addition to a more rigorous, purposeful testing and training, O’Reilly suggests developing mechanisms for evaluating AI algorithms. He calls for public involvement in the process of improving AI models, not just relying on public outrage to punish the developers whose first iterations got it wrong. For example, having a mandatory period of public evaluation and comment before models are put into use by government agencies or regulated enterprises.
As for specific instances of AI programs that need to be regulated, O’Reilly says “I think it’s clear that if we impersonate someone and spread lies about them, that’s a clear harm. Pick up a few clear harms and use them as test cases.” Sometimes companies do the right thing for consumers without any regulation, but not always. “If companies don’t have the backbone to do it maybe they should be forced to do it.
O’Reilly tells me he thinks it’s amazing that he just created a painting in thirty seconds: “We are empowering people to do things that were previously impossible. People are going to be creative with that in ways we don’t expect.” Moreover, with the democratization of AI, “we get a reset of the competitive landscape.”
AI or more accurately General AI is the creation of a competing lifeform. Like a chocolate dipped shotgun nobody should be biting on this remarkably bad idea.