Nur, you study how firms can adopt AI and quantum in a socially responsible way.   Can you tell us what you are finding about the adoption of responsible AI?  Im thinking of your paper with Nan Jia of USC on this topic.

There’s good news and bad news here.   First, Nur and his co-authors have studied the 2000 companies in the US that hold AI patents.  Just 125 of them have published AI principles that guide their work.   Why so few?   AI principles can slow you down because they require process, people, internal inspection to prevent things like hallucinations, biases, and scandals.   Organizations have to spend time and money to follow principles.  And often they don’t have that.  To Anita, this sounded like some bad news.

Second, those AI companies that are adopting principles – or something akin to principles – are doing so to attract talent.   It turns out that engineers and other AI experts who become employees demand good behavior.   So access to talent can come at the price of enacting principles.  To Anita, this sounds like bad news is the sense that AI companies are only pursuing principles to get access to key resources, but there’s some good news is that employees are demanding accountability, transparency, and the like.

Third, there’s good news in that many of the employees that push for principled AI come from University environments where they are trained to make these demands.   The University of California at Berkeley stands out as a place that has published a lot of responsible research, and that graduates engineers and other AI experts that push their employers to adopt AI principles.    Anita is relieved that this is possible, but of course worried that it’s not systematic.

All in all, principled AI seems underdeveloped.   There’s a lot of work to do to get this right.

What kinds of AI principles matter most to employees?

Nur’s research with Neil Thompson and others suggests that these are very complex issues.  It’s difficult technically to overcome hallucinations, protect privacy, and advance transparency (for examples).   Many of the companies – including Google and Anthropic – have done a great of work on this.  But of course there is much more to do.

Nur, I know you are especially interested in AI adoption, but you have also done fascinating related work on geothermal energy adoption.  It strikes me that the unifying thread is at least in part on the role of regulatory ambiguity.  When does ambiguity help and when does it hinder tech adoption?

Nur’s work in this area focuses on geothermal, but the issues are also relevant for work on AI and quantum.   The authors are seeking to point to regulatory ambiguity as a new construct that we can discuss as central to the evolution of these new technologies.

Think of this example:  California has clear regulatory steps that geothermal energy generators must undertake to get permits to build generation capacity.  It’s neighbor, Nevada, does not.   In Nevada, there are multiple agencies, questions about authority, unclear steps, and imprecise language.

What does this mean for startups vs. established companies that want to build generation capacity?    Established companies might be attracted to California because the risks of the regulatory ambiguity in Nevada are too great.   A large company wants a clear process, minimal reputational risk, and protections for large-scale investment through clear permitting processes.  In contrast, a start-up with less to lose may prefer Nevada, where the permitting process is cheaper and more flexible and expansive.  

Over time, California may lose ground to Nevada if innovative startups use the regulatory ambiguity in Nevada to build out their facilities.  Once they are built, then Nevada may not want to lose their comparative advantage in energy generation.  This is critical particularly for AI and Quantum computing, which will require massive amounts pf power in the coming years.

Will California win as well?  It’s hard to know, but perhaps California retains its advantage in the conceptual and software sides of AI, and Nevada excels in power generation over time.  But there are too many wildcards to make a prediction, since there’s no guarantee of investments that will translate into the outcomes that we’re envisioning.  And there are many other states seeking to get onto these playing fields.

The take-away is the idea that there will be jurisdictional competition, and that regulatory ambiguity will play an important role in the investment calculations that companies make.

 What makes you most optimistic?

Nur comes from Bangladesh, a poor country in which barely 5% of the population had electricity in the year that he was born.  Now 99% of the population has electricity.   And other accomplishments are significant, such as that now Bangladesh is polio free.   These things are accomplished when people take the reigns to make things better.  This is what Nur wants to do:  To make a difference by stepping up and contributing – even in a small way – to improving lives.