Ok, so this blog is for my musings, and this is very much a musing. However, a couple recent things have prompted some thoughts. The issue is Large Language Models (LLMs). As I’ve said, I have no problem with the tech inherently. It’s really optimized for language roles (as the name implies). What is concerning to me is the hype, and so the use. And, it’s led me to wonder if it is, or what is, the right tech for the job.

So, up until LLMs, when you wanted something done, you built the appropriate tech. You put together specifications, and development teams built it. (And then they asked UI to fix the problems, as Don Norman talked about in The Invisible Computer. Probably then handed off to training folks to address the problems from the bad design outside the UI.). It took time, and money. And, if you didn’t use something like Watts Humphrey’s Personal & Team Software Processes, you likely took too long, and had too many errors.

You could use AI for more decision tasks. So, either symbolic if well-defined, or based upon training data and machine learning if ill-defined in principle. You, of course, still have to live with or address the biases in the rules or databases. And, of course, the brittleness at the edges of the decision space. Still, all told, we had approaches. Then, the world changed.

So, for one, my colleague Kevin Wheeler (a deep expert in the talent world), talked about how the tasks of the talent function are in flux. What he cited was that many of the rote tasks were being made redundant. Which is good, I opined, if we’re removing tasks that people aren’t good at (I’ve said before that we should be doing pattern-matching and meaning-making, and leaving rote to the machines). However, there’s the problem of developing the expertise. So, for instance, as Etienne Wenger and Jean Lave talk about in communities of practice, moving from peripheral tasks to central as you understand the domain, But you need those peripheral tasks!

Plus, we’re seeing people being laid off. Meta just announced another 8000 being laid off, for efficiencies, and there have been announcements from many of the big orgs (and small ones are making similar decisions). Cutting through the smoke, what we sees is that folks face increasing expectations to use AI to do things faster, and the expectations are increasing (without, mind you, an increase in rewards for same). But, increasingly we’re not working together as we used to.

What also showed up in a LinkedIn conversation is the expectation that we can ask LLMs to do the things that we used to do by writing software. And yes, such systems make mistakes, but so do humans, right? Yes, and…we can assign responsibilities with humans, and they’ll be corruptible and more, but we have compliance to deal with that. As Markus Bernhardt is pointing out, we’re not doing a good job on that with our systems. We, too often, haven’t worried about the necessary guardrails, and security, and responsibility, and governance, and….

What is concerning me, as folks like Mark Britz talk about, is that we’re losing the human connection. We’re losing:

• the upward path for new folks
• the continual development of our own capability
• the necessary checks and balances that keep our systems secure

Really, we’re turning away from doing things together to maximize outcomes, and instead are working to be more expedient. In short, we’re trading off effectiveness to achieve efficiency.  And ignoring ethics along the way.

In short, I fear we’re using the ease of doing things with LLMs to avoid the hard work of doing things the right way. Which I resist. I am seeing quite the backlash amongst the folks forced to use AI. Which includes the pressures by execs to use it, and the eagerness of those to promote it who stand to gain (whether vendors or consultants). We see research results showing that folks are thinking less, execs have unrealistic beliefs about how productive it makes people, and it’s pretty simple to recognize that the costs can’t stay as they are. The illusions from smoke and mirrors aren’t a good basis on which to plan.

Don’t get me wrong, I do see the benefits of LLMs. I just see them in context of the bigger picture of people, tech, and the broader picture of AI. And I could be wrong, it’s true. It’s just that there’re some reasons to believe that decades of immersion in the relevant fields have some basis for questioning whether we’re using the right tech for the job. I welcome your thoughts!

I’ve talked in the past about my tools for learning, as Jane Hart’s survey prompts. Yet, Christy Tucker asks about software stacks (for consulting), and I realize there’s a different answer when I’m talking about doing versus learning. Yes, as Harold Jarche says, “work is learning and learning is the work”, but there are times I’m using tools to keep myself on track rather than to render my ongoing thinking. I augment my abilities with tech, and some is just about executive function rather than learning. So what is my technology for performance?

Personally, I use Apple’s Reminders to render ‘ToDos’. As I’ve said, if I’ve promised you something and it doesn’t get into my digital world, we never had the conversation. That also is true for Apple’s Calendar. I used to use them separately, but now I’m coordinating between them. I used to block out time to get things done, but I’ve now set up a cal.com to book time, and it looks at my calendar. So, I now use timing on reminders to get things done that are urgent, and save the calendar for time-specific things.

Financially, I use Quickbooks to send/receive invoices. I should switch, but haven’t yet. The problem is that you’re kind of locked in unless you change on your calendar boundary. I also use my bank’s app or website to dod things like send/receive payments. Occasionally I use PayPal, too.

I’d mention Notes, as a way to mull things, but that’s really learning. Though I do grab and store recipes there (and share with fam). Not that I make ’em all, but it’s where I can keep the ones I find online. Likewise I take notes on biz meetings with Notability, but again that’s not really performance, it falls more into the ‘learning category’. I do have some templates that I’ve created in Word, e.g. proposals, reports, and schedule of fees. Probably should migrate to a platform that’s more open. Libre? Open Office? There I go, standing up again…

Now, that’s for me, myself, and I, but I also do things with Elevator 9 and the Learning Development Accelerator (LDA). For both, I’m using Slack to talk to people. I have separate channels for both, but am glad they’re both using the one platform. Email too, of course. Teams was part of a a previous engagement, and am frankly glad to step away. I also use Zoom, a lot. Again, happy to not use Teams or Google Meet for that purpose.

Collaborative docs are different. They’re writing, but for others, so they begin to cross the chasm (really, it’s a continuum, but…). So, I’ve used Google Docs. I really haven’t collaborated using 360, because I don’t have that type of license. I also have used Apple’s Pages with the folks who do run Macs.

Usually, my browser’s for learning, but I also use it to get things done. That’s one of the reasons I recently made the switch from Safari to Vivaldi. For one, it’s a ‘Chrome’-equivalent browser, but doesn’t have the ownership probs that Brave suffers from. It also doesn’t have the ‘tracking’ problems Google introduces (it’s why my search engine is DuckDuckGo, too).  It’s problematic, in that it’s (too) customizable, for power users, but the defaults aren’t bad. Still learning about it, but I’ve mostly got it under control (e.g. I think I’ve a solution to the microphone issue that was bedeviling an LDA vid attempt). Though I’m reasonably tech savvy…

One other tool of note is Notion. I wouldn’t necessarily choose it myself, but it works. You can imagine I’m not keen on the strong pushes it (and everything else) are making towards AI, but it’s in use for LDA project and knowledge management, and it’s working. We’re a) probably not making full use, and b) could’ve used something simpler, but…we have someone familiar with it coaching us, so it’s all good.

So that’s my technology for performance. It’s not sophisticated, but it’s manageable, and affordable. Thoughts? Yours?

There’s been a persistent belief in artificial intelligence (AI) practitioners that kind of explains the ongoing development. The mantra is “if we can understand it, it’s no longer AI”. This is relevant to what we’re seeing today, in important ways. So let’s unpack that.

The point of AI is that we’re trying to model intelligent behavior, usually human intelligence. It doesn’t have to be human, of course, we can be trying to model intelligent behavior that’s not human!  However, at least for much of AI, we’re trying to understand our own thinking. After all, the human brain is arguably the most complex thing in the known universe!  For instance, we really still don’t understand consciousness; why does it even exist? And, what is it really? We’re getting closer, but we’re not really there.

I’ve mentioned before that machine learning came about as a reaction to the failure of symbolic approaches to capture some of the nuances of human behavior. For instance, recognizing images in noisy environments is something we were better at than systems. Symbolic systems were ‘brittle’, in that they did what they did, but failed as soon as you tried to generalize. The neural net approach, where we trained systems, more closely mimicked how we ourselves perform. And, really, the general trend has been that as soon as we understand it, it’s no longer AI, it’s just good programming. Which is fine. We’ve advanced our understanding, and now we can build it.

What we encode in the weights of networks is essentially what’s known by the system. We don’t really know what we know, and we create models to explain how things work. We can get more complex with non-linear models (which our brains struggle with), but non-linearity itself is an understanding we’ve created to explain how the world works. Yet, what we actually capture in our networks isn’t available for scrutiny. Same with neural nets. Most networks were small, relatively speaking (massive, but small compared to our brain). They were also essentially limited. Trained on X, or Y, so check recognition or loan approval, but our brains can handle much more.

We’re seeing now that while the algorithms have made some modest improvements, we’ve vastly scaled the net size to achieve large language models (LLMs). However, we’re still running limited foci. So, for instance, language (training on the text of the internet), or images, or video, or music. Yes, there are systems that are integrating, but as far as I know (which might be wrong), they’re really grafted and not integrated completely. Most folks I know go to different tools for different things.

The result is that, particularly in text, we now have the general experience that we can ask questions of LLMs, and get meaningful replies. And, make no mistake, this is useful. For general language tasks, such systems are great. They’re doing pretty great jobs of summarizing prose. This passes the so-called “Turing Test”. (Alan Turing posited that if you had a system and interacted with it so that you couldn’t discriminate between system and human, it passed the test.) An article now suggests that a reasonable assessment would say we’ve achieved artificial general intelligence (AGI).

Apparently, I’m not a reasonable person. They’ve addressed my complaint, they posit. The article (in the parts I could read before the paywall) say that such systems do, indeed, have a world model  I argue that such systems don”t really know what a ‘dog’ is, having never seen one, petted one, raised one, walked one. Read prose describing all, yes. Experienced it? No. To me, there’s a difference (and I’m channeling Stevan Harnad’s Symbol Grounding Problem).

I also believe that there’s an AI architecture that’s more likely to achieve AI. (Particularly if we give it sensors to experience the world; and imagine what progress if they got the same hype as LLMs!) LLMs are trained on the vast corpora of text (illegitimately, I’ll suggest), and so understand an average that’s greater than any one person. (They’re also trained on other data, like video or images, for other media.) But they’re not as good as an expert in the field. They can do impressive things in areas you aren’t an expert in, but an expert in those areas can point out flaws. Yes, they can solve unsolved problems in, for instance, math, and pass tests, but those are knowledge tests, not ‘do’.

Ultimately, I don’t have to agree with the consensus, and maybe I’m just a grumpy old man (and get off my lawn!). I would like the progress we’ve made and what we can do with LLMs, if it weren’t for the hype that’s distorting business and mindspace. AI is more than LLMs, despite the way folks are speaking, and the valuations aren’t real. Also, we now know what they’re doing, and how, so I can say “it’s no longer AI” and mean it ;).