Data science is the practice of producing knowledge rather than building/coding. You can make systems that leverage that knowledge in an automated matter, but when you are deciding what to build, it is not a matter of coding. It is a matter for management. In my experience the sort of arguments a PhD can make fall on deaf ears when management lacks the requisite skills. An organization where the PhD in stats, ML, etc. produces the same outcomes as the code camp hero should look at its management structure before doubting the value of experience at depth.
The line is not so clear as you are indicating. If I produce a system that makes 7 billion people feel slightly more depressed, then that system will in all probability contributed to a non-zero number of suicides. The unfortunate matter is that with fantastical scale there comes a need for systems that can understand the long tail of outcomes because there are real lives and impacts in the tails.
Great point! We are constantly discussing the criteria. For example, if a robot lost power and fell on someone, do you think that is an incident? What if the robot was never turned on and fell on someone in shipping? When these challenges come up, we try to expand the editing guide to meet the challenge of all the ways intelligence of the artificial variety can go wrong.
Incident 73 dates back several years -- it is part of the initial Google spreadsheet from 2018. At the time, we had no formal definitions [0] or criteria [1] and only a vague idea of what we would find. My cursory re-read of 73 agrees with your assessment. We are likely to define it as not being an AI incident, but the incident profiles are also meant to be permalinks that the research/blogging/etc communities can analyze and build community around. I'll bring 73 to our weekly editors meeting and we may add it to the list of incidents that didn't pass additional scrutiny [2]. Like Wikipedia, we can't expect ourselves to get it right every time, but we can strive to always improve the processes for producing a solid ground truth. Please let us know our failings and/or consider working to become an AIID editor yourself.
After many protracted discussions among editors, we decided not to threshold the severity of the harm. An allegation of harm is enough. Not all incidents are created equal when it comes to severity and the taxonomy feature helps sort through these questions on top of the permalink the incident profiles provide.
Incident severity and scale varies wildly between incidents. We provide the taxonomy feature to support organizations like CSET adding color to incidents that support more descriptive stats.
I agree about it being a stretch, but we err towards indexing boundary cases. This one turns on whether you would consider the web application as being part of ChatGPT or not. In aviation, there are many systems that are part of safely flying from point to point that are not part of the plane itself. The control tower, the runway markings, and the processes built around these things. While the ChatGPT web application is not the model underlying the ChatGPT front end, it is part of the whole intelligent system.
Edit: On Slack we considering downgrading this one to the second tier status, which is an "issue," but our "issue" definition typically differentiates between events (a harm happened) and non-events (a harm may happen). This doesn't quite fit here. We have robust editor discussions around these ontological problems and are developing the ruleset in response to challenges. We aim for inter-rater reliability via the editor's guide. More details are here https://arxiv.org/abs/2211.10384
I am one of the people behind the website and have been building AI "looms" for 13 years. We are working to make AI safer as our goal -- we are not working to smash the looms. Without the FAA and similar orgs/efforts, it is unlikely that millions of people would be flying today.
We also index incidents related to people transferring decision authority to simple programs like the Stanford vaccine distribution: https://incidentdatabase.ai/cite/91/
In general, we err towards indexing useful information to the state of AI safety and introduced indexing "AI Issues" to develop the boundary and collect information on systems likely to produce incidents in the future: https://incidentdatabase.ai/apps/discover/?display=details&i...
This [0] research paper sounds relevant to your "stegonographic social overlay network."
[0] F. Beato, I. Ion, S. Capkun, M. Langheinrich, and B. P. (2013). For Some Eyes Only: Protecting Online Information Sharing. In ACM Conference on Data and Application Security and Privacy.
I don't think CryptoCat has been distributed as a traditional web app for at least a year (probably more). The browser code is distributed as an extension, which does not have the properties you describe.
edit for sub-comment: A traditional web app updates every time you hit the URL. In a browser extension your code is not necessarily tied to any remote origin, including the Chrome/Firefox stores. It is a user's choice to automatically receive updates from the vendor and this is the same choice that you make if you use apt-get vs manual download/checksum/sig check/audit.