The Everything About AI Report

May 7, 2024

dinosaur30a_thumbThis essay is the work of a dinobaby. Unlike some folks, no smart software improved my native ineptness.

I read the Stanford Artificial Intelligence Report. If you have have not seen the 500 page document, click here.  I spotted an interesting summary of the document. “Things Everyone Should Understand About the Stanford AI Index Report” is the work of Logan Thorneloe, an author previously unknown to me. I want to highlight three points I carried away from Mr. Thorneloe’s essay. These may make more sense after you have worked through the beefy Stanford document, which, due to its size, makes clear that Stanford wants to be linked to the the AI spaceship. (Does Stanford’s AI effort look like Mr. Musk’s or Mr. Bezos’ rocket? I am leaning toward the Bezos design.)

image

An amazed student absorbs information about the Stanford AI Index Report. Thanks, MSFT. Good enough.

The summary of the 500 page document makes clear that Stanford wants to track the progress of smart software, provide a policy document so that Stanford can obviously influence policy decisions made by people who are not AI experts, and then “highlight ethical considerations.” The assumption by Mr. Thorneloe and by the AI report itself is that Stanford is equipped to make ethical anything. The president of Stanford departed under a cloud for acting in an unethical manner. Plus some of the AI firms have a number of Stanford graduates on their AI teams. Are those teams responsible for depictions of inaccurate historical personages? Okay, that’s enough about ethics. My hunch is that Stanford wants to be perceived as a leader. Mr. Thorneloe seems to accept this idea as a-okay.

The second point for me in the summary is that Mr. Thorneloe goes along with the idea that the Stanford report is unbiased. Writing about AI is, in my opinion of course, inherently biased. That’s’ the reason there are AI cheerleaders and AI doomsayers. AI is probability. How the software gets smart is biased by [a] how the thresholds are rigged up when a smart system is built, [b] the humans who do the training of the system and then “fine tune” or “calibrate” the smart software to produce acceptable results, and [b] the information used to train the system. More recently, human developers have been creating wrappers which effectively prevent the smart software from generating pornography or other “improper” or “unacceptable” outputs. I think the “bias” angle needs some critical thinking. Stanford’s report wants to cover the AI waterfront as Stanford maps and presents the geography of AI.

The final point is the rundown of Mr. Thorneloe’s take-aways from the report. He presents ten. I think there may just be three. First, the AI work is very expensive. That leads to the conclusion that only certain firms can be in the AI game and expect to win and win big. To me, this means that Stanford wants the good old days of Silicon Valley to come back again. I am not sure that this approach to an important, yet immature technology, is a particularly good idea. One does not fix up problems with technology. Technology creates some problems, and like social media, what AI generates may have a dark side. With big money controlling the game, what’s that mean? That’s a tough question to answer. The US wants China and Russia to promise not to use AI in their nuclear weapons system. Yeah, that will work.

Another take-away which seems important is the assumption that workers will be more productive. This is an interesting assertion. I understand that one can use AI to eliminate call centers. However, has Stanford made a case that the benefits outweigh the drawbacks of AI? Mr. Thorneloe seems to be okay with the assumption underlying the good old consultant-type of magic.

The general take-away from the list of ten take-aways is that AI is fueled by “industry.” What happened the Stanford Artificial Intelligence Lab, synthetic data, and the high-confidence outputs? Nothing has happened. AI hallucinates. AI gets facts wrong. AI is a collection of technologies looking for problems to solve.

Net net: Mr. Thorneloe’s summary is useful. The Stanford report is useful. Some AI is useful. Writing 500 pages about a fast moving collection of technologies is interesting. I cannot wait for the 2024 edition. I assume “everyone” will understand AI PR.

Stephen E Arnold, May 7, 2024

Comments

Comments are closed.

  • Archives

  • Recent Posts

  • Meta