Sunday, December 17, 2023

What do we know?

TL;DR -- So, what happened in 2023? Lots. The world has changed, and it has not. People showed their seemingly limitless capacity to bow to the machine. 


This post serves two purposes. For one, it looks at AI (usual sense) over the past year and collects material pertinent to ongoing discussion and work. The first list contains links with some commentary for each of item. This collection points to different periods over the last year. OpenAI made their announcement on 30 Nov 2022. At that time, we had these two posts which pertained to our continuing topics of research which includes technology.

  • Family Search (29 Nov 2022) - we have used this site form the beginning. It represents the efforts of descendants of those who went west. Over the past few years, our focus was on the Interior of the country for several reasons.   
  • Perspective, a necessity (1 Dec 2022) - the post looked at veterans from Ipswich, MA (after the Revolution) who went out to Ohio and created a settlement. The trip was well-documented and even repeated later in the 1900s. At the time, that was Michigan Territory. It was only 40 years later that the Santa Fe Trail was formed out of St. Louis, MO. A mere half-century after that, the railroad allowed easier coast to coast travelling. 

The image shows a copy of the "All time, popular" count by post. The topmost, related to the marriage of Thomas and Margaret, comes from the 2014 timeframe. The next one is 2021 as is the one titled "St Louis MO to San Francisco CA" which has spawned several of posts and discussions. The "San Francisco" post is from 2023. 

So, we will continue this area of research as there is plenty questions that are open. Our FAQ will be updated in 2023 to account for new information. In the post titled "A(rtificial) I(Intelligence) researched properly", we go through our own work over the past decade and make some predictions about the future. Given that this work carries on prior research allows a longer time frame that seems to be normal now. So, we will continue to describe things from the historical viewpoint, both technolically and phenomenonally (which is how we view the U.S.). 

Disclosure: We did not get involved with the xnn/LMM, generative material until after a couple of months had elapsed.  As noted above, our timeframe is much longer than normal; and, we need to slow things down in order to establish a little bit of maturity. All of this will be discussed. 

So, going forward, the second list deals with the major change in 2023. We will also use this to start a summarization of our positions on the subjects. The list follows a sequential timeline and uses posts from this blog and related material. 

  • Introducing ChatGPT (30 Nov 2022) - there is a thorough overview of the motivation and what was being offered. But, access to the OpenAI system was announced. It did not take long for millions of people to sign up. We did see references to this activity on sites like Quora (who offered POE, later; the head of this organization is on the Board of OpenAI - mentioned them, in yesterday's post - Little Exeter NH).
  • ChatGPT (2 Feb 2022) - we heard of the system from a colleague (a linguist) in late Jan and finally got around to doing a query on the 2nd (Groundhog Day). We asked about Thomas Gardner and thought that we were enlightening the thing with a suggestion. In this post, we mentioned that the output did not read correctly. That is a result of the techniques used and can be controlled, somewhat. On the other hand, there is a "omni" and "know-it-all" nuance that is obvious. The New York Times review, How ChatGPT Kicked Off an A.I. Arms Race (3 Feb 2023), looked at some of the issues which were to get increasing attention over the year.  
  • Square one? (19 Mar 2023) - Gloucester MA had kicked off their 400th. We were supporting from a distance without bringing in the crew as that would be a 2024 issue, we thought. Our thought was, too, to emphasize Margaret (Fryer) Gardner's contribution. We had not been paying attention to the feeds from WikiTree, however a researcher in Feb of 2023 went through the load of images from the digitized files of Sherborne, Dorset. In doing so, he discovered birth records for about all of the Gardner children, as having been born in the old country. Seeth was born in Salem. So, the family did come over. Did Thomas come over to support the Cape Ann effort? Rev. Hubbard suggested such. Remember, he talked to the principals. But, one thing noticed a while ago was that the Conant family's book did not mention Gardner. Nor did the Woodbury writeup. This feat of technology is interesting in itself, for various reasons. The current state of affairs? There are two Thomas Gardner profiles on WikiTree. One is our Thomas; the other has no information about him. So, one or two? It's like being back at the questions of two or three wives. We settled that in 2012 using the same records. So, Gardner Research opened the door. In the meantime, we did research on California families (the long reach of New England). 
  • Research notes (2 Apr 2023) - using our new WP site, we summarized the situation and provided a copy of the list of names and births. There is a gap between the birth of John and Samuel which might show Thomas being away. John was born in December. Thomas could have left without knowing that Margaret was with child. Supported conjecturing will be allowed and supported by technology. Note: at some point, let's discuss how an imaginary basis (of several types) supports science.  
  • Technology, computing (May 2023) - after some reading and discussion, I contacted the head of the former Sperry Univac Knowledge Systems Center. He wrote a brief article on how Univac got involved with the effort in the mid-1980s. Lately, one hears more of this, but our contention then and now was that the machine learning solely approach is unbalanced. Hence, we see that the output is not to be taken without some bit of salt (perhaps, a whole lot). But, that is not how it ought to be. Stay tuned. There are several other approaches that can bring maturity. A recent survey points to the diminishing in trust which has been ongoing for a while. But, ChatGPT and its ilk are accelerating the downward slide. Fact? Not there. Fiction? Have at it. Except, one caution (rule) would suggest that "fake" results be tagged as such (analog, truth in advertising). 
  • Intro to Bard (7 Jul 2023) - after some time with ChatGPT and looking at issues all around, we finally tried Bard. That activity was all research including things like evaluating how the generative approach performed with mathematical problems. One thing to note is that Wolfram tied this approach to his mathematics system early one, mostly to have more full interchange with users. We need to see how this work is progressing. But, there is a lot more. Bard, basically, was more palatable as it did not try to be argumentative nor did it spout out just to hear itself speaking, so to speak. The post, To Bard or not, summarized some of the issues. For instance, we ran into a PhD thesis that looked at concepts from Kant (the philosopher) being brought bear during training. A known downfall is that the xNN, as being used, squashed nuance. One cannot expect to retract either input or its influence; definitely, there is a need to honor the requirements for footnotes and other referencing modes. In Dec of 2023, Bard was updated significantly. We tried it, but the issue of more fiction than fact remained. One noted improvement was that Bard now stresses teaming, human and machine (rather the "omni" status mentioned above). That was significiant as it changes the tone of the interchange. 
  •  ChatGPT a year on: 3 ways the AI chatbot has completely changed the world in 12 months (30 Nov 2023) - we don't agree with the list, as a huge result is trust and its diminishing. What to believe? All of these purveyours suggest checking the results give by their system. Oh, we all have time and energy for that? 
  • AI not solely ML (4 Dec 2023) - this is our retort and focus for the future. We need to bring to awareness the knowledge work done in the industrial world. There are many varieties. But, the focus needs to be on humans and their abilities. Enhancements would not be of the machine and its dominance; no, human would obtain an extremely useful tool if this were to be done right.  
So, themes abound. But, what is America about? The dream? Technology can be both a boon and a bane. Why has the latter been more prevalent? We still say AIn't about the emergence of "intelligence" of note. On the other hand, phenomenal mathematical processing has been done by the computer. That could be and ought to be our focus going forward. 

We have a super abacus and slide rule at our disposal now. That is, if only we can understand the need for maturity and the capabilities that it brings. 

Remarks: Modified: 12/22/2023

12/22/2023 -- Dartmouth '99; New Frontier, 21st Century Type. ... THE FUTURE OF AI IN SCIENCE AND MEDICINE, talk at Gairdner Foundation, Oct 25, 2023. 

No comments:

Post a Comment