Saturday, July 10, 2021

A(rtificial) I(ntelligence) researched properly

TL;DR -- As we have mentioned in lots of post, New England has a long arm through space and time. Very long. We have looked at the interior example of this. We have a nautical one, to boot, that sits there awaiting attention. And, we need technology for various matters. Keeping it simple over the decade was one goal. Accomplished as we do not even have a smart phone. From the outside, apps look to be idiotic. I'll get one soon and be more specific. With regard to the purveyors of the clouds that are strangling the folks, we are immune and neutral. We'll tsk, tsk at any who is not behaving in a mature manner (did Zuck ever get over his bull-in-a-china-closet self?). Once we do a little look back which implies future happenings similarly (truth engineering is the focus), we look at DL and its huge impact of an unreasonable nature (hint for the brainy, think Wigner's comments). What is wrong here? Well, we 'deep' dive, proper sense, into DL and come up with the pearls of wisdom that seem to have been lost in the mud. 


An ongoing concern is that we need to have a technical focus for a while as we re-look at the computing landscape. This type of thing seems to have occurred on a periodic basis which we have covered somewhat as we proceeded. But, since technology will be a focus going forward, we will be more regular on looking from that angle. 

And, we will pull together several threads as we work this. Then, we need to assess where the world is on related topics. Warning, we're going to the core and back in an attempt to clarify what are the real issues and why so much hype and confusion. 

So, let's recap, somewhat. It does parallel other activities in the world of technology. 

  • We started this in 2010 using facilities of Microsoft (MS) as we were looking to be in the world of  ASP.NET for web apps with access to OfficeLive for the business focus. The first cut of the web site went quickly due to their tools. And, all was well, or so it seemed. 
  • So, in 2012, MS pulled the plug (leaving a bunch of little moms and pops scrambling as they lost the core of their business model). Rather than step up to Office365, we went looking. That was nice. As we got to see how the web was about seventeen years after its 'wild west' start. We tried several of the state-of-the-art approaches. Mind you, we are now facing the onslaught of change coming from Jobs' gift (in other words, the app world was dawning). However, since our purpose was content management in a totally new environment, we fell back to our roots which was riding the wave of the internet from the beginning via Unix and Linux. Remember, at one time, SUN noted that the network was the computer (early 1990 timeframe). So, we converted from 'asp' to html using little graphics with attached functions as the GUI. It was fun. The choice was to use an ISP who ran Linux and offered lots of software choices including a web design tool. Didn't like it, for many reasons. So we did our own thing. 
  • After two years (2014) of managing content with antiquated methods, it got to be wearing. So, at that time, we took the time, again, to reassess the state of the art with respect to web stuff. And, we looked, again, at CSS. Yes. So, that became the GUI provider (parameters and systems, just like we see with DL - more on that, lots more). I saw a 3D graphic engine implemented in CSS/JS but still had lots of content to cope with. I was relieved to have some progress improvement.  
  • Then, by 2016, Google was saying to those who went out pioneering earlier, respect the little guy or you don't get any attention (meaning, in their ways of pushing stuff). So, we became mobile friendly. The keyword was being responsive. This was done with even more CSS. At the same time, we picked up WordPress in addition to Blogger which became under the auspices of Google. In the meantime, we kept looking at new kids on the block. After all, some spend their lives playing with the stuff. So, alternatives abound. Guess what? I have lots and lots of examples of people being screwed (hence, this is one thing on our plate) because they didn't take the time to understand technology. At least, know more than how to push buttons on that thing that Jobs' gift gave the world (incidentally, I am about to get one - the smart phone - I call it a dumbing device and can show examples of that, to boot). Anyway, Google provided a little test facility to check out your site. We passed. That was enough, then. But, again, soap box: everyone seems to have descended to this one-eyed little thing where one cannot multi-task. Can I explain further? You bet. It'll be on our table for discussion for a long while. Also, later I used someone's Jobs' gift to look at the site. Not bad given that it's not using custom apps but a good browser (that will not go away, folks, for many reasons). 
  • Then, 2017/18 came along. We decided that was a bit to type and decided to go a little smaller. We asked for opinions and settled on which has been a blast. We still have the old site for older material that will be reorganized. On the new one, we are stressing the need for truth and its engineering. Frankly, the site is to be a test bed for ideas and discussion, eventually. More on that below. At the time, we relooked at the options. And, settled, again, on HTML/CSS but brought in JS. Now, there are several varieties of JavaScripting available. However, I just saw top-of-the-line bit of research in truth maintenance (older concept) which balances multiple worlds (essential) which was created with html/css/js. Yes, I have seen lots of topnotch sites using this. Topnotch? Not commercial flimflam, please. Not academic, either. 
  • Now, we're 2020/21 and have started to redo, again. Of course, it's time to reassess the world. And, knowing about New England and American History a little more now than in 2010, that whole framework has to come to become more visible in a coherent manner that will persist. Guess what? We will argue that there is research to support that will be as important as biomedical research. After all, we're talking the soul of the nation and of the world's people being entrapped via technology that is driven by motives are that not what we might want given what we know of the 400 years of effort on the part of thousands of families. But, that's for later (see below). Now, we will look at apps. Also, we will be looking at multi-media and the different ways to express some notion with both provenance known and veracity shown. Ever heard of AI? 
Okay, switching gears. One thing that I have noted is that people who are presenting tough stuff (say, as one expects with science) always use a mode that is as complex as the subject. Others present complex stuff using simple means; some people call this 'static' but do so erroneously. I will have examples, but many of these sites represent the latest thinking in fields such as mathematical physics which really is the driver of all of science. Too, the web has gone wild with approaches to presentation. Some of these will live on. Those are the ones that we want to pay attention to. We will have the more entertaining aspect, but hard-core issues area going to be a huge focus. 

Let's look at one, for now. Of late, there is this thing called Deep Learning which has gotten a lot of press and generated both hype and fear. The former? Oh yes, AI, as in artificial intelligence, has finally shown its head. Well, not. We will explain.  The latter? Again, AI. Old Musk (pusher of the electric car and its illusions) says, it's summoning the beasts. Other brains are bewailing the future and it perils. 

Well, in short, AI ain't. We have spent the past year looking at the guts in terms of the underlying mathematical and computational approaches. And, we are ready to discuss where there is more hype than not. We use that phrasing as, of course, it's not all for naught. There is some good stuff there. On the other hand, business managers are buying into this stuff to the detriment of their customers and their employers and the rest of us. Won't name names, yet, but some billionaires are exhibiting classic idiocy. 

To close this, let me point to two papers that were in the recent ACM Communications. They are both open to public view (thanks, ACM) so that we can access content of the papers without going into a closed environment. The ACM is the Association for Computing Machinery and has been around since the advent of computing. Earlier, it had an academic flavor but is now balanced with people from the industry. The Communications is their flagship publication that is printed monthly. This recent issue is just full of good stuff as we find with these two. 
  • Deep learning for AI -- The ACM provides a Turing Award to folks who contribute to the discipline in an extraordinary manner. This paper is based upon a Lecture by three researchers who shared a Turing Award: Bengio, Lecun, and Hinton. Recognize those names? These guys were of neural nets all along and so can be seen to have sloughed through the years. They are not recent wonders exploiting graphic cards nor are they pushing notational magic such as tensors. Of late, news reports have mentioned these three talking (Hinton is whom I have seen referenced the most) and saying, wait, neural approaches have issues. We have to back up and relook at the situation. Well, this article expresses those issues and discusses some work that needs to be done. Great stuff. The article is open for reading (thanks to ACM). They mention symbolic approaches, the old stuff to some, and the later modes that are heavily numeric - exploiting age-old ideas about transforms, optimization and more. But, it’s in their ‘Recent advances’ and ‘The Future of Deep Learning’ where we see discussion of other things that are being pursued plus some suggestions and notions that need attention, like attention (now, a soft type). I liked the higher-level cognition little section. Yes, the old ‘planning and reasoning’ methods ought to be more than dead meat to those numeric overlays that are so insidious. I can talk that, too. BTW, Minsky’s ideas on the society of mind are still apropos, to boot. But, there is a lot more to put on the table. Next bullet.
  • Deriving equations from sensor data using dimensional function synthesis — ACM, thanks, this is open, too. Gosh, in the last bullet, B&L&H were musing about higher-order (which even tensor notations aren’t going to give to the neural approach) and its use. After all, B&L&H are academics and know the mathematical stuff. So, Cambridge (Tsoutsouras, Willis, and Stanley-Marbell) to the rescue. This paper reminds me of a lot of things that I saw and dabbled in. To me, what was the conclusion was having a trained person in the loop. Trained? Yes, thoroughly comfortable with the type of thinking that a good physicist would follow while modeling - used since model-based approaches are still germane to the subject, perhaps even more so now with the flood of data and improper analysis that is killing us - in order to solve some problem. Good engineers fit this bill. 
So, we will need to have a technical focus, in part. One motivation is that DL, as automated mode, is gnawing at things textual and visual and creating what is considered knowledge views. Well, it is not. There are 'deep' issues at play here that are being ignored. These deal with what can be ungrounded domains, such as mathematics and philosophy. One might ask about 'grounded' which has been discussed forever. Well, humans will play a part in whatever dynamics are needed. 

Example of human deep learning
guiding DL

Yet, that complexity does not mean that things (are allowed to) just run a natural course with negative impacts upon humans and their culture and society because we shirk our responsibility to be mature. Run fast and break things? Heard that before? It's right out of Silicon Valley (to me, silly) during the past decade and one-half when things accelerated to more instability than many thought possible. Sure, the web/cloud is there. Well, the majority of that is something that we need to study which some have termed crapularity (really crapology - in response to the interest in the Singularity). 

We use the concept of research as being important. Science is one focus that needs to be there. Definitely, mathematics will be of concern. Now, people? What do they bring? That is something to learn more about in a way not done before. One suggestion is that the web/cloud offers lots of modes to look at in a new way. For starters, a survey of the past four or five decades with respect to how computing evolved and a consistent presentation of the results is imperative. I see now lots of older practitioners, who are retired, talking of their work. One theme, perhaps to be expected, is how the new modes merely rephrase what was done before with new gimmicks. Some sciences seem to redefine themselves from time to time. BTW, what will be re-evaluated, to boot, will be DNA from several perspectives that seem to have been swept off the table and under the rug as commercial interests took over the science. Wait, the old and the young? Generational dynamics? Actually, the topics have no limit. However, one important element? We are dealing with the American Dream. Who owns that? What is it, exactly? And, how might American Indian thoughts apply in modern contexts such as science? 

Remarks: Modified: 08/05/2023

07/11/2021 -- Review and edit; add image. 

07/12/2021 -- We are back in the saddle. Concerns abound. Such as, some have been taken to the cleaners due to unscrupulous developers and outright charlatans with modes that lurk and pilfer. Some have paid way too much for some capability since they were too old to know or to make the effort. Oh yes, youngsters not knowing? Yes, that is the case, too. Can we have a sustainable web for all? 

08/05/2023 -- I haven't been watching, but this post rose on the list of the most watched, recently. The theme is one reason. We will lift this up to further usage, as we are working on GB XIII, 1. One topic is year-to-date. The other deals with new directions. Wasn't there a boy band with this name? I know that there is a literary publishing company with this as a name. ... In Nov of 2022, OpenAI pushed out ChatGPT which is an example of the capabilities of LLM. We will have more on that, and related themes, soon.  

No comments:

Post a Comment