Query Results

Make a list of brilliant thinkers and doers when it comes to ethics and artificial intelligence—as consulting group Lighthouse3 did—and you’ll find some Santa Clara people in the mix: Irina Raicu J.D. ’09, who directs the internet ethics program for the Markkula Center for Applied Ethics; and Shannon Vallor, the Regis and Dianne McKenna Professor in the Department of Philosophy. For a series of conversations about AI, we sat down with Raicu to ask what we should be asking.

What’s fair? And what’s human?

Some of the questions that have become more obvious—and that a lot of people are dealing with now—revolve around fairness and AI. A few years ago, only a handful of academic papers focused on AI and fairness. Now that’s grown exponentially. One of the questions that will come up for the foreseeable future is: What is it that makes us human? What do we mean by “artificial intelligence”? What is human intelligence? How is it different?

There’s something about real humans that’s messy and complicated. And most situations are more complex and nuanced than we would like them to be. That’s part of what’s interesting about the difference between humans and AI. Humans understand context and translate among concepts; AI is a long way from doing that.

Should there be AI-free zones?

One of the questions about artificial intelligence that I’ve been writing and talking and thinking about recently is: Should there be some AI-free zones? Are there areas in which AI would actually not do a better job than human decision-making? If we can delineate those, we can help prevent some harm and also increase trust in AI in the areas where it does do a better job.

Right now, part of the problem is that people are presenting AI as an improvement on everything. There are areas in which there are changing social norms and in which implementing AI/machine learning as we know it would actually be embedding the current or prior social norms—rather than allowing things to develop.

For example, there is a claim by one writer that I’ve been rebutting. He argues that, eventually, algorithms will make “better” decisions about whom we should marry. That’s exactly the kind of area in which there’s so much complexity, so much variety, so much personal preference, and so many changing societal norms that no, I don’t see how an algorithm is going to do a better job of deciding that for us. That’s true with other relationships, too. We need to allow society to continue to grow—morally as well.

I spoke about this recently at a meeting for the Partnership on AI, a consortium of businesses, civil society groups, and academic centers working on a global effort to share best practices, advance public understanding, and make sure we’re developing AI for good. The bottom line is that we can’t allow algorithms to decide societal norms.

How do we keep humans accountable?

We want to make sure that it’s still humans who are held accountable for what the AI does. Some developers have talked about “bias laundering”: the notion that you have all this potentially bias-embedding data, and then you have the biases of the people who are designing the algorithms, but somehow we would run the data through the algorithm and the outcome would be objective. How do we make sure that people understand that it’s humans all the way down, and that the accountability then stays with the humans?

In a lot of areas, I don’t think we ever want to live in a society where we say AI has gone through so many iterations that it’s no longer really the humans who are in charge. (But AI also plays very differently in different contexts, so broad generalizations are not really helpful.) The accountability has to be placed somewhere, and we can’t let go of it by just allowing the neural networks to do their thing.

What happens with AI discoveries that are too dangerous to release?

The research institute OpenAI came up with an artificial intelligence text generator that had the researchers worried enough about its capability that they decided not to release it to the general public. Like with the “deepfake” video clips, this tool could generate plausible-sounding fake news. Some have argued that holding back the code is an empty gesture.