These days I’m puzzled as to why, among all the intelligent & alert experts I’m been hearing speak about Large Language Models, almost no-one is pointing out that LLMs consist of extremely-high-level statistical processing of patterns rather than anything like what would conventionally be called “intelligence.” But now ChatGPT and all of its 3rd & 4th cousins are called “AIs.” The production of impressive “natural language” is part of what makes LLMs so convincing to people as intelligence, but this “natural language” is accomplished by a most unnatural means.
Of course, one can argue: how is it that human children develop language capacity but by absorbing patterns and drawing from this vast store of patterns? Isn’t this what LLMs do? Ok, yes – but there is something else vast that is missing from this picture: the embodied experience that is attached to every language pattern, including, especially, a huge spectrum of emotions, plus all aspects of perception, proprioception, etc. Intelligence is not just verbal language. Memory itself is not just patterns of verbal language – it is episodic, experiential, saturated with emotion.
I am not a computer scientist, but isn’t this plain enough? What am I missing?
(And: there is something else vast that is being added to this picture: the astronomical amount of energy that LLMs require to accomplish what they do. The idea of brute force seems to take on new meaning here – brute force as contrasted with real intelligence.)
Moreover, statistical analysis of patterns, with extremely large amounts of data so as to be able to predict how humans would most typically string words and progressions of argument together, is not the same as employing logic or inference or intuition, or having confidence in a proposition based on a human connection with the person who made the assertion. At best, LLMs are merely an imitation, a mimicry of what human written responses look like. In my view, LLMs are not only far from “general AI” but are far from intelligence itself.
It seems to me that the real intelligence represented by LLMs is the long-accumulated technical expertise and creativity of human computer scientists who have built on each other’s accomplishments and figured out how to write code to produce such astonishing and powerfully-useful results. For good or ill, they have changed the world.
Meantime, I’ve discovered that my intuition about intelligence is shared by Sara Imari Walker, theoretical physicist and astrobiologist, who helps run a research center at Arizona State University studying the origin and nature of life. She was interviewed by Sam Harris in his Making Sense podcast posted 10/22/2024. When Sam asserted that nobody doubts we’ve achieved a narrow kind of artificial intelligence (as opposed to general AI), Walker pushed back: “I probably don’t ascribe the level of intelligence to current algorithms that other people do, so I maybe don’t agree that we actually have narrow forms of intelligence – I think it’s very easy for us to recognize things we think of as intelligence because we see ourselves mirrored in our technology. But the technology has been selected to mirror us, so it’s, it’s very unclear to me how much the substrates themselves are embodying things that we might view as intelligence versus being kind of a false mimic to us.” Yes, exactly.
One real shock was to hear Douglas Hofstadter, in a recent YouTube video (about his breathtakingly creative and highly acclaimed book Gödel, Escher, Bach, 1979, 1999), claim that LLMs might be conscious. Not just intelligent, but maybe with consciousness, which is an enormous leap. How can a person with Hofstadter’s level of genius think something this dumb? He says that a lot of what is being produced here is totally coherent and believable, and sensible, and thus “I have to start assigning meaning to the symbols they are using, and saying, if there is meaning here, then there are ideas here. And if there are ideas here, there is thinking here, and if there is thinking, then there is some degree of consciousness.” (https://www.youtube.com/watch?v=JYZcHSqqxtg).
But Hofstadter has failed (at least here) to adequately define “meaning” – in what sense is there meaning? Meaning is an aspect of the experience of conscious biological beings. Meaning is related to self. But the self of an LLM is a gigantic computing center, not a human with a head, legs, arms, emotions, bodily sensations, etc.
Thus LLMs cannot experience human meaning – they can only produce a simulation of what is considered meaningful by humans. None of their “meaning” is related to an actual self, so there is no “knower” inside, no real meaning and no consciousness (cf. Antonio Damasio, Self Comes to Mind, 2010).
(My tentative conclusion about Hofstadter is that he has dreamed of machine consciousness for so long – to be accomplished via recursive, self-referential patterns – and he has contributed so profoundly to what has been achieved so far toward that goal, that he finds confirmation bias to be too ready at hand. We are all, from highest to lowest, myself included, susceptible to believing what we want to believe.)
The above considerations remind me of a pithy statement I came across as a graduate student more than twenty years ago: “Lurking behind predictive scenarios of a computer-driven society is an emaciated view of what it means to be human” (Charles Jonscher, 1999, The Evolution of Wired Life, p.248). Yes, we are still doing the same thing we’ve been doing for many decades now, ascribing far too much humanness to our present and future clever creations, whose mimicry of some human behaviors, while impressive with each new breakthrough, is still profoundly limited.
Do you have expertise, such that I might want to pay you to use that expertise for a need I have?
Can you write coherently about your area of expertise? Are you able to think and communicate convincingly about the fuzzy boundary between theory and practice in your area of expertise?
These capacities will be of great value to me, whether your expertise (and my need for your services) is in law, in architecture, in environmental engineering, in software design, in music production, in medicine, etc. I will want access to someone who has real contextual understanding of the important issues that shape that field and that inform decision-making.
With the advent of the newest family of AI tools (large language models like chatGPT), a far wider circle of people can pretend convincingly, in writing, to have your expertise. Not only can they do so, but they can ostensibly perform the same tasks that only someone with your specialization can do, and do this compellingly, much faster and for much less money. With only minimal knowledge, they can learn to prompt the technology to produce reports, proposals, contracts, theoretical speculations, etc. at a level far above their own understanding. The fruits of your long years of labor, and of millions of other professionals like yourself, have been extracted endlessly, without constraint and without your involvement, and are now being given to others who have not earned them.
Thus, at some point, it will no longer make good economic sense for me to pay you for your expertise, since I can now get it elsewhere for next to nothing.
I should hasten to acknowledge: this is the same kind of thing that has been happening for 25 years and longer to many, many kinds of human expertise, including that of professional musicians (especially drummers). And I am certainly among those guilty of taking advantage of these trends in my own small way as a musician.
So I must ask myself, “Why does this newest phenomenon feel so different? Is it really different at all – that is, is it really a cause for great concern?”
As a productivity tool for real professionals, who have enough understanding and know-how to check behind chatGPT and correct its errors – this kind of tool can present an overwhelming argument for improving one’s efficiency and one’s overall work life. What a time-saver it has already become for many, many people!
But what percentage of the factual information contained in the output of chatGPT4 is accurate? Perhaps 85%? The rest is simply false, including citations and references to material that doesn’t even exist (see Just Because ChatBots Can’t Think Doesn’t Mean They Can’t Lie).
So, by and large, what percentage of users of this technology are going to bother checking behind such a great digital assistant? Even for professionals who have enough expertise to do so, how many will be conscientious enough to go through this tedious task for every document that chatGPT has helped them create? Humans are humans, after all.
I suspect that, as with so many other internet-enabled phenomena, the problem is partly a matter of scale. It is also a matter of access and social reach. Both of these topics bear much more scrutiny and discussion.
Meantime, the harms enabled by this kind of technology, both intentional (e.g., new kinds of fake news) and unintentional (who knows?) will be at least as large as the enormous potential benefits. So far, society has barely scratched the surface of either the positive or the negative (and this little essay omits a great deal of what’s already being tried out there).
Meantime, the salience of the problem, for me, is partly that it touches my family. Not directly me, nor my wife – we are both happily retired and not in danger of starving or losing our home. But our son does PhD-level work for the state where he lives. He is, for now, the sole breadwinner for a family of five. What will this phenomenon do to his field of expertise over the next 10 years?
[No, I don’t recommend going back to quill pens! But I do recommend real human fine art, like this elegant photo by Lisbet Sjoberg. ]
I am not accustomed to writing “this is a lie,” saying it directly and without hesitation. (I’m not even accustomed to posting anything at all, lately, in this blog.)
But the other day (May 19, 2021), in my Gmail interface on my laptop, Google presented the shocking message below:
This was shocking because it is patently untrue [Correction: it was patently untrue until 2017, when apparently Google quietly announced that it would no longer mine Gmail contents for advertising purposes. I’d been immensely preoccupied with work in 2017, so I missed it somehow]. When Google first introduced Gmail, there was debate about whether users would accept the fact that Google would mine all email content to feed their algorithms for advertising (as declared in the terms of service at the time). This was in the DNA of Gmail from the beginning (and in the DNA of the company as a whole), as is well documented in sources such as the thoroughly researched The Age of Surveillance Capitalism by Shoshanna Zuboff.
We’ve come to expect all manner of evasive and self-justifying language from Google about their surveillance-based business model, but I never expected to see such a [statement]. … (And if they’ve suddenly stopped using email content to feed their algorithms, it should be announced as a major change, not a blanket statement implying that it’s always been this way – and this major development should be all over the technology news realm.)
What to make of this – ?
I should not have allowed several days to pass before posting this. And it’s unlikely much of anyone will notice my post. But I haven’t yet noticed other media covering it. I hope journalists somewhere are picking it up.
[I don’t remember it being treated as big news in 2017, but, once again, I was very busy that year.
Meantime, what Google hasn’t done, apparently, is to discontinue using the profile data of users that it had already obtained from Gmail for over a decade.]
If impeachment is inherently political, as pundits keep telling us, then it is wrong for the voices of attorneys to be included, for either side, in impeachment proceedings, or at least in the trial stage. The reason this is wrong is that attorneys, by law, have no allegiance to truth or to what is best for the country – their only obligation is to win the case for their clients. But in an impeachment trial, the voces of attorneys only take us further away from any pursuit of what is best for the country, what is honorable, what is true. The nature of politics is already problematic enough without adding the amoral rhetorical skills of attorneys, whose objective is simply to win.
Having attorneys advising during the preparation stages, before the impeachment trial, seems fair and appropriate enough. But in the actual trial, the body conducting the proceedings and reaching a verdict is composed of elected public officials whose obligation is to the people, to the country as a whole. All of the speaking and listening in an impeachment trial, other than that of witnesses called to testify, should be done in this public-service capacity, despite the tendency for party loyalty to overwhelm it.
To preserve the hope that party loyalty can occasionally give way to a greater loyalty, keep the voices of attorneys out of impeachment trial proceedings.
I haven’t posted anything here in a long time, but today I thought I should put down an anchor in regard to something I don’t hear anyone addressing yet.
(First: I am an American who votes as often as possible but has never joined a political party. I am an independent voter, neither Republican nor Democrat. My voting history, were it to be made public, would bear this out. BUT neither am I a supporter of the current occupant of the White House.)
Given what we now know about the huge topic of social media manipulation, populism, Russian (and other) interference in American elections (and elections elsewhere), etc., and given the uncommonly large number of Democrat candidates who’ve thrown their hats in the ring for the American 2020 presidential election …
My sense is that there has been a subtle and sophisticated effort, still going on now, toward persuading as many Democrat politicians as possible to run for president. What better way to support Trump’s re-election? The profusion of Democrat candidates helps to make sure 1) they will attack each other enough to do real damage, and 2) the last one left standing will be a caricature candidate whose platform has been boxed in too far toward extremes to make for a viable presidency (or a viable-enough candidate to win).
One can easily imagine this: artificially manufactured indicators of extra support can give an energy boost to multiple less-well-known Democrat candidates, so as to convince them that they have a chance. I fully expect this to be uncovered as one of the strategies being pursued by those who want Trump to be re-elected, whether they be here or abroad.
Let’s see if my theory bears out in the coming months.
I’ve had respect for LinkedIn among social media sites because their privacy policy has been relatively good and because of their professional focus.
And this week I have been once again recommending to our students that they join our program’s LinkedIn group, one of only two groups I’m a member of.
[I still recommend this. The group’s name is long – Learning, Design, and Technology (formerly Instructional Technology) at the University of Georgia – but you can easily find us by searching for Learning, Design, and Technology.]
Lately though, some strange things are happening – will LinkedIn continue to have clear value for professionals – especially, our students when they are in the job market?
Yesterday I went to my main account page and then clicked into the menu – Interests>Groups. Though it’s by far the group I’m most active in, including a post two weeks ago, LinkedIn’s algorithms couldn’t find the group at all – “Let’s find you some amazing groups to join!”
Even more strange are the “Top Job Picks for You” emails. Among the positions I should consider applying for, according to recent emails from LinkedIn, are:
Latin Teacher for Gwinnett County Public Schools
Store Manager for Victoria’s Secret
Electrical Technology Engineer for Goodyear Tire & Rubber Co.
Food Technologist for Zaxby’s
Assistant Director of Housekeeping for Ritz-Carlton Hotel Company
Warehouse Manager for UberPrints.com
Sales Team Manager for VoxPro
Sales Manager for Macy’s
Say what? Where did all this come from? Anyone who looks at my LinkedIn profile would see how disconnected these are from any of the qualifications and job history I’ve provided. Some of these are, perhaps, very remotely conceivable; but – food technologist? Housekeeping? Victoria’s Secret?
I can perhaps account for some of this by surmising that while LinkedIn has pledged to do mostly reasonable things with the data they collect from me, they’re also mining data about me that they’ve purchased from other sources. For example, I studied Latin in high school (40+ years ago). Also, before I settled into an educational media/instructional technology career, I had several sales jobs. But they wouldn’t know these things from my voluntarily-completed profile.
Meantime, my profile does include a couple of past positions I’ve held that included the word “director” or “manager” in the title. So I guess this means I could be director or manager of anything!
Thus we have another fine example of technological affordances that are touted as “personalizing” features when they are actually impersonalizing. They reveal these algorithms to be just the mindless machine processes that they are. Nothing could be more impersonal.
On the other hand, why is the problem so egregious at the moment? Perhaps this is a temporary problem, and they’ll get closer to the mark again at some point. I hope so, because I want those “Top Job Picks” emails to be meaningful information for my students.
I have finally posted three videos, both to YouTube and to Facebook, from the Nov. 7 2015 performance of the Root Groove Quintet at Perk Avenue cafe in Madison, GA.
None of these are complete tunes, but each goes to the ending. Video taken by Sun on my phone. Because of the vertical orientation, VIEW THESE IN FULL SCREEN.
It’s only appropriate that I make these links available here … so proud to have had these extraordinary gentlemen with us!
Tom Kuhlmann, Chief Learning Architect for Articulate Global Facing 21st Century Design Challenges
August 22, 2015 https://youtu.be/UGnvxjY68eU
Dr. Keith Bailey, Director of the UGA Office of Online Learning Instructional Design: The Difference Between Good and Great
August 21, 2015 https://youtu.be/yrrVdw6UVe0
The 2nd annual IDD@UGA conference is exactly one month behind us now. I have been busy with catching my classes up and various other things, but now we have the evaluation data in hand!
For me, the feeling of energy and enthusiasm among the participants was palpable. We really seemed to be doing something meaningful together. And many sources of verbal and written feedback have been telling us we had an extraordinarily good event this year. Now, we have data as well to back this up. For purposes of this blog post, I’ll just share one number: on a scale of 1-5, with 5 being the highest rating, the mean of means from the evaluation survey is 4.52. As evaluations go, this is an excellent number!
Tom Kuhlmann, Chief Learning Architect for Articulate Global, delivering the Saturday morning keynote address at IDD@UGA 2015
We also increased our registration numbers over last year (from 106 to 114). This is more significant than the numbers imply, as several key parameters, including location, were very different this year. Instead of largely repeating what we did the first time around, we were in experimental mode again. We also extended our geographic reach, having Georgia, South Carolina, Florida, Minnesota, and Texas represented.
It was extra special to me to have my academic colleagues from other institutions get involved: a tip of the Clinton hat to Dr. Brad Hokanson of the University of Minnesota and Dr. Michael Grant of the University of South Carolina.
But what is most significant to me, as always, is seeing important connections established, connections that enhance our program and contribute to career development, as outcomes of the conference.
I owe very substantial thanks to all who helped “make it so,” especially event coordinators Melanie Baer & Beth Massey, my colleague and mentor Dr. Lloyd Rieber, graduate student David Squires, and COE Office of Advancement writer and designer Kristen Morales. These, along with graduate students Yinning Zhang and Hua Zheng, were indispensable. Also thanks to all our new, continuing, and just-graduated IDD students for diving in and embracing what the conference was all about. And to the IDD@UGA conference committee and IDD faculty for their ideas and assistance (Lloyd Rieber, Mike Orey, Ikseon Choi, TJ Kopcha, Yinning Zhang; Rob Branch, Chanmin Kim, and Janette Hill; Mark Schwartz of Articulate Global and Robb Bingham of Converging Solutions). And once again to our proposers/concurrent session presenters & panelists added rich content that was consistently well-received. And to volunteer session facilitators! And finally, and especially, to Articulate Global for sharing the talents and wisdom of Tom Kuhlmann & Mark Schwartz; the UGA Office of Online Learning for sharing the talents and wisdom of Keith Bailey; and to workshop presenters Rob Branch, Robb Bingham, and Ron Braxley (the “three Rs”).
AGAIN, THANK YOU!!!
Dr. Brad Hokanson of the University of Minnesota led a portion of our wrap-up (“warp-up”!) session with a discussion of takeaways. Learning, Design, and Technology doctoral student David Squires took extensive notes, and a brief version of these are given below.
Best to everyone & see you next year!
Greg
—
Featured Speakers Tom Kuhlmann, Chief Learning Architect, Articulate Global
Dr. Keith Bailey, Director, UGA Office of Online Learning
—
—
—
—
—
IDD@UGA 2015 Closing Reflection and Wrap-up (“Warp-up!”) Session:
– Thoughts and major take always?
Respondents (n=17) Short Answer Round (n=3):
1) Technology and assessments very distinctly included. The Bloom’s taxonomy session wasn’t about difficulty but meeting expectations of Bloom’s questions. Asking about perceptions and feelings, applying own beliefs to questions, synthesizing and analyzing.
2) Whole conference was inspiring seeing the applications in business and one’s own field. Seeing how this is applicable to other classroom environments. Can see the transfer but it was also overwhelming.
3) This work is fun! Animations, video game and armadillos. Learning can be hard, but we can make it fun.
4) Blown away by the sessions – zSpace/ Vizitech was amazing – accelerated learning was wonderful. The big picture: this technology can have, and does have, an impact on learning.
5) Accelerated learning – how to bring learners into trainings – be aware of an emotional level-incorporated scenario. Got participants thinking what they do themselves – instead of nervous or distraught there is an emotional connection. There is no cognition without emotion.
6) Conference in general – more excited to see the possibilities of what could be – so many pathways – fired up about potential possibilities.
7) Related to fidelity – there is an intersection/ sweet spot in the middle – depending on the client looking at a spectrum. An instantaneous solution could be a YouTube video or pay for an iPhone experience – diversity of solutions in the world.
8) ‘Nobody knows nothing’ – may feel inadequate about skills – but we can get what we don’t know from others. So many resources at IDD conference and great place to make connections.
9) Insight to other technology that we may never been exposed to. Animation/ other solutions not previously considered – opens up new avenues to consider and cost effectiveness.
10) How interconnected all of us are – business /k12 / other industries – we’re all connected and can be combined.
11) Opportunity to have conversations in a community of practice – and a will to create a learning environment where people thrive and collaborate with others. Next time make more time for conversation outside the sessions!
12) Thinks the purpose of eLearning is important and can help frustrated people find solutions – seeing other solutions to measure student learning and think deeper than quick decisions. Take away – Don’t frustrate students.
13) Enjoyed roundtable – eLearning problems based on your own passion – learned so much in 10 minutes. If you’re passionate, the quality shows when you care. The end users will feel it too.
14) The presenter Pasha’s videos changed the paradigm. Students’ accessed, material on their own in a self-directed manner. Tech enabled acting in a different role – new ways to deliver content.
15) Vizitech inspired ideas of context, budget and timeline parameters – building is no longer just content but also context. Complex and difficult, the missing piece is the context of the real world — and context goes beyond content.
16) The instructional technology field is changing and growing. It won’t go away. There is more out there – the field is relying more on technology and as participants we get to change education.
17) Take away questions: How I’m going to take the information and inspiration and keep all that info in my day-to-day job? How do I change my job to make it more engaging and keep the ball rolling? We all share that challenge. Each exposure to the content builds more each time we attend IDD.
Short Answer Round: What will you tell your significant other about the conference?
1) In the video workshop we made a movie and it involved more research and I learned more than anticipated. I now know how to start a video project now.
2) I made the right choice – moving into this field.
3) It is important to remember why we have the conference. The design studio curriculum is the best thing to have been involved in.
– It begins when we get two people and have a design conversation.
– The conference is great to sit and talk with people
– There is a vision of a face-to-face experience to have a personal meeting.
– IDD@UGA is about professional development; there is something “pure” about it.
– Combines the newest learners in the field and very experienced adult professionals.
– The conference has a mentorship component. The next generation of scholars and designers come from here.
– Lets come back next year and compare notes!
Greg Clinton's Eclectic Blog 