Human accountability in AI? Great idea. Important idea.

Best of all, we can have it – using methods that are centuries old.

In his Computer Weekly article entitled “Accountability is the key to ethical artificial intelligence, experts say”, Sebastian Klovig Skelton notes that “The inherent opacity of artificial neural networks means human accountability is needed to keep these systems in check, rather than increased transparency of its inner workings.”

While Skelton’s concern appears to be related to practical control over particular instances of artificial intelligence, Stephen Hawking presented a more ominous concern in a 2014 BBC interview:

"The development of full artificial intelligence could spell the end of the human race….It would take off on its own, and re-design itself at an ever-increasing rate. Humans, who are limited by slow biological evolution, couldn’t compete and would be superseded."

Professional licensing gives us a measure of assurance when we visit a doctor or lawyer. It doesn’t mean they’re so competent they never make mistakes. It does mean that if a mistake is big enough they can find themselves at a new job, flipping burgers.

That important point about professional licensing is often overlooked by those who write about it. It’s not so much that rigorous education and testing qualifications ensure a mistake-free practice. Partly, they serve as evidence that the practitioner is serious in their intent. But more important than either of those considerations is the threat of loss of livelihood and, likely even more significant, the public humiliation of loss of professional license (“Daddy has a new job...”

The AI Practitioner’s Professional License

As Skelton points out, AI sources can contain millions of lines of code. That can be a huge challenge for a professionally licensed code auditor, whose job is to examine the code and attest that they found no funny business in it; nothing that deviates from the stated intentions of its specification, and no egregious security vulerabilities.

Now of course if we are to have professional licensing we must ask what public authority issues the license. According to Marques Hayes of World Atlas, there are 195 sovereign countries in the world, 193 of which are members of the United Nations. UN agencies such as the International Telecommunication Union have done a respectable job of getting all those sovereign nations to yield a bit of their sovereignty in adopting necessary telecommunication stantards. Entities such as IANA have been able to impose order on Internet communication, largely because governments were very slow to understand what the internet is and how it affects them.

But that still leaves parts of this Earth that could refuse to accept the need for professional licensing of AI practitioners. Now picture a B movie villain, recruiting the power of an AI server to take over the internet from his lair on some small nation that refuses to join the UN and its IEU.

What’s needed is a source of duly constituted public authority (DCPA) that does not involve nations , national boundaries, or for that matter any kind of geographic distinction.

Check out the City of Osmio and its Professional Licensing Commission at https://osmio.ch . Its 26 (and counting) professional licenses consist of x.509 certificates that are appended to x.509 identity certificates that work like digital birth certificates.

Let’s heed the warnings of both  Sebastian Klovig Skelton and Stephen Hawking, and put human beings in control of artificial intelligence.