Mark Carroll's Journal

Recently, BCS Edinburgh hosted an interesting talk by Prof. Tony Prescott on the psychology of AI, he works on intelligent robots. Some of the more philosophical aspects caused me to think further and realize a way to explain why I am not content to simply regard brains as a kind of computer.

Using an approach based on knowledge representation and symbolic reasoning, I can create a simple robot that would seem curious and be able to adapt to surprises in achieving its goals. After some exploring, it would have assembled and refined an explicit model of what seems to be going on with itself and its environment. We could even look at that model to see how it explains what it has seen and how it guesses what will happen, which would seem to show that the robot has some degree of awareness and understanding.

Do I think that such a robot would be interestingly sentient from an ethical point of view? No, because there is nothing in that robot that experiences or understands anything in any direct way, any more than a coffee machine has any ideas about making cups of coffee, it's just moving parts (and changing electrical states) that happen to be assembled such that the overall effect is coffee-making. The robot's understanding is only in the heads of the people who translate its memory state into data structures and ascribe meaning to them.

As a human, my experience of embodied cognition is quite different. (I don't know about yours, but mine is.) There's something in me that actually comprehends abstract things directly: the quale of having figured something out and what it means. I experience this reasoning in a way that I don't believe my robot would, just as I don't think a coffee machine has any overall experience of its coffee-making. I don't know why, nor what it means, simply that my vivid subjective experience of these higher-order concepts seems enough unexplained that I am hesitant to put my brain in the same category as my robot's brain even if I cannot point to any science that justifies the idea that they are categorically different in this regard.

Incidentally, I think this may be separate from the question of free will, I don't have much of a position on that. One could put some qubits into the robot if it helps, not that I have specific ideas as to why it would.

I've written here about my background fear of AI coming after white-collar jobs, the technical work people do from their desks. The skilled trades are doing better, at least for now; it may still be a good time to train as an electrician.

I've also written about wrestling with what's worth doing in my own time. There are things I'd like to do, though I keep getting older before much getting around to them. I don't want to put much time into something myself that some AI agent will be able to do in a few minutes. If I show that I can build interesting things without AI assistance, does anybody care?

I can't easily use AI agents myself to do much because they cost money to use. I could self-host but there's a shortage of GPUs, TPUs, fast memory. It's not much of a stretch toward realizing that the wealthy have the means of production: the hardware shortage impedes democratization of the ability to create software. Sure, a modest, targeted model might still accelerate what I can do at home but it's outshone brightly by what the expensive agents can do given the scarce wherewithal. There is no way that at home I will be spending what I cost my employer in AI budget in the typical week. So, I get plenty done at work, not much at home, and it's the cost of modern AI that magnifies that divide: the monied can avail themselves of a powerful accelerator to pull further ahead, the rest of us do what we can with what we have, which is much less.

Thirty years ago, a reasonable PC with a Linux installation was sufficient means for doing impressive work. Now, it takes money even just to gain marketable experience in orchestrating agents to much effect. It being the people with large AI spend budgets who can achieve plenty, relatively speaking, seems like the kind of thing that would greatly bother Marx. Something tells me that the fruits of what can be achieved by employing expensive productivity accelerators aren't about to be shared out equally. I find myself fearing the kind of dystopia described by Kurt Vonnegut in Player Piano.

I enjoy mathematics but I am an amateur. ( I can do high school math but, by the time I got to college, I was sick of examination papers. )

In the end, I kind of coasted through my undergraduate career based on what I found interesting and remembered; at college, I never much reviewed and tried those past exam questions because I was still sick of exams. Fortunately, following the concepts is the main win: ( I can fill in the detail later when I need to. )

Anyhow, that is a long way around to saying that I am decent at the kind of math that is: write down all the things one knows, consider how to transform it all, then work one's way step by step toward what one wants. Broadly, that suffices for the math ( that I actually have to do. )

I know a little bit of abstract proof but not much. ( I know some simple strategies only. ) The proper mathematicians actually prove things in a sense that I just never had much exposure to; maybe I can follow a proof and nod but coming up with it is a whole other thing. The math I can do is largely just following the pathways they've laid for me.

All this comes to mind partly because there are simple mathematical things in my life that just seem obvious. For example, consider duplicates. If I have a string of sausages like, sausages sausages sausages, I know I can place my cursor anywhere after the first space, hit backspace nine times, and end up with just two sausages. I bet somebody knows how to prove that. Or, another kind of duplicate: I feed the cats from an assorted box of three flavors that I can't see into easily. I know that if I simply pluck four packets then I will find an identical pair among them. I can probably extend this to what if I wanted to feed three cats the same thing but that's plucking rather more. The point is that there are people for whom a formal proof of these things would be child's play, they can find a way to weave the story from tiny pieces. I never developed that ability: I have not had to do much proof and my practical engineering of reliable systems tends to get along just fine with my saying obviously and not having to convince people who don't also see it.

With R. and company away until August, I am settling into a new routine of simply keeping things going, including dog and cats, until the family's back home with me again. Given that L. our dog is not trained to do without people, I am with him, working from home, and not going to shops. Fortunately, he's a good little fellow. I have done inventories of my food and pets' food and all suffices.

I am highly appreciating modern appliances. I was well-appointed when living by myself in Eastern Tennessee, here the appliances are smaller and cheaper but still very useful. In particular, dishwashing and laundry become a breeze in comparison to how they could have been otherwise.

We had rotten weather for a few days, then on Wednesday morning it brightened up nicely. It's kind of middling again today, we'll see how it goes. L. is oddly reluctant to walk outside in various directions, I may try bringing some small treats along next time to try to coax him along. He can be quite stubborn but I would prefer to vary our routes. I am reminded of a dog I saw on television who would walk outdoors only in order to eat pieces of a trail of lemon drizzle cake.

I had a bit of a milk disaster last night. I have some plastic jugs of milk in the freezer to get me through the month. It coming near time to thaw one, I placed it out on the counter (milk seems to take a long time to thaw) and subsequent leakage revealed that the expansion had damaged a seam in the plastic. I left the carton dripping into a tall freshly rinshed jug overnight so we'll see how this odd fraction of milk tastes and for how long it lasts.

Work has gone fairly well in recent weeks, I am being quite productive, still very much with the help of AI code assistance. It's so capable, and advancing so fast, that I fear AI to be coming after many knowledge workers' skilled white-collar jobs irresistibly, and I wonder about my own career's trajectory given that what I'm best at is exactly the kind of thing that AI's becoming good at. This also makes it difficult for me to trade off its large potential and rate of progress against its current unprofitability, in considering how comfortable I am to have my meagre pension investments exposed to it.

Sound hasn't been working on my personal laptop computer for months. I assumed there was some problem, of course, but figured it may even just be that it's old and something came loose or whatever, it's had a rough life. It's always a mild relief when it boots successfully. I confess that the computer seems less worn now that R. wiped the display, cleaning off the awful crud I'd gotten used to.

Anyhow, to my surprise, I found I had sound again, then after a reboot I didn't. Intrigued, I started investigating. Eventually, I got as far as aplay finding the ALSA device busy, then creating a dmix configuration. Since then, at least so far, sound has worked again. Very pleasing indeed.

R. and I are back from a week's tourism in Paris. I enjoy how we continue to find ourselves agreeing often: R. certainly sees why I prefer Paris to London, it's so pleasantly walkable. ( I am always happy to go back. )

Being dragged reluctantly into the modern world, I ( tried using public transit via smartphone ticketing. )

For me, Paris tourism is typically some combination of walking and Métro around central Paris visiting various attractions and just taking in the environment. We hadn't prebooked much so we had some freedom to go as our whims and increasingly aching bodies would take us. The weather was generally good, I consider us lucky. We kept sufficiently on the beaten tourist path that I scarcely had to ( attempt to speak French. )

The gardens varied rather. Many of them seemed to be wide, dusty, pale gravel paths, lawns, conical shrubs, cuboid trees, statues, hedges with right angles, etc. We wandered through the Jardin des Plantes which at least had flowers, a variety of rather well-grown ones indeed, though no fewer right angles. I think the Japanese might be rather better at the kind of garden I like.

Paris has a considerable abundance of publically accessible magnificent buildings that I enjoy seeing and being inside. ( I enjoyed a few of the museums. )

In this visit, we did not ( eat out at fancy restaurants. )

Last night's return flight was from Charles de Gaulle into Edinburgh so we took our usual commuting route, the train into Queen Street, as part of our journey back home to Glasgow.

I routinely use generative AI in my workplace, my employer encourages it and pays for it. It works well, it's a definite help. At least for the meantime, it requires my expert supervision, close monitoring, to do good work but it's actually rather clever at times even if often rather dumb too. Where my work strays beyond my expertise, it fills in for me.

I have probably mentioned that I like the description of computer programming as mathematical engineering, it captures what I enjoy most about it. It's rewarding to devise and express good solutions. I love to create systems that do well at behaving in desired ways.

So, sometimes, for those parts of my work tasks to which I was looking forward, I've typically been working with the AI enough that it has the context to say, hey, you still have this bit unfinished, shall I do it? and I'm like, no, let me!

For the moment, I can still capture some crumbs of what I love to do. However, I wonder how obsolete that's becoming, the future's arriving faster than I expected. You could drop me back into the 1980's and I could be very happy writing software but these days nobody wants programmers who could hit the ground running in that kind of environment. Given the speed at which coding assistance has become rather good, I can't help but wonder if the 2030's will largely have only jobs for people who can direct the constellation of artificial agents well. That's a thing I'm sure I can do competently to support my family but … how much do I want to?

I love to learn about what clients actually need, figure out how I can meet those needs by creating software, then to deliver something valuable to them. But what I love most is the part of the process that machines may soon do maybe not quite as well but far cheaper than I.

I find myself looking back to things I once did and appreciating that at least I had the chance. I have loved doing simple things like feeling the hot, dry breeze in Death Valley, driving a rusty pickup truck through the Ohio countryside in the sunshine, walking along the beach in Aberdeen, and frequenting the AANI weekend market in Taguig. Or, in this case, the chance, repeatedly, to be paid to solve interesting problems by creating software by my own brain and hand. Of course, I can still do what I like as a hobby though it feels emptier if it just means that I am doing something the hard way. I also wonder how healthy it is for one's likes to be overly nostalgic. I have an elderly relative who probably feels as if the world has gone downhill since the 1950's. I don't want that to be me someday, I should find more ways to embrace the future.

I dreamed that I was taking some computer programming course where, in the classroom, I had to develop some real-world robotic mimicry system like: I draw and, as I draw, it stitches thread so as to reproduce the lines I'm drawing. We had a reasonable amount of time for completing the task because the class was a double-period, at the start of the day.

The dream transitioned to a different activity that I now forget more, where I was facing a related programming assignment but outdoors: I was approaching a small, ruined outbuilding where I was to perform some part of the task, not able to take long because some rather autonomous thing was out there pursuing me.

At work, I've been trying to finish off a non-trivial task that I started last month. There are plenty of aspects to get right: I am trying to improve the performance of a part of our code that comes in different flavors, is called in different ways, and can utilize various constituents. Generative AI assistance has been helping me along. Perhaps I have finally got it all to a good state, we will see what the coming week brings. I start next week by filling things out for our annual performance review cycle. I look forward to working on my next task which is a smaller, more usual one. I have been putting some extra time into the meatier task because it's important and has already lingered plenty; it will be a relief to finally tie a bow around it, here's hoping.

Today is my first day of leave from work for over the Christmas break. This morning we sent the boys off back home to Asia to visit family, now it's just me and R. I am relaxing on the sofa with our dog L. while R. brings some sanity to the kitchen storage. I already feel my headspace increasing and have been getting some small postponed things done. Many more await.

I am quite good at sleeping. Given the opportunity, I can do plenty of it. This morning, I dreamt we lived somewhere else and I spied a sizable swirly unnatural-looking Weather Thing approaching, and turned to R. to strongly suggest that we leave the house now and drive elsewhere.

Again, I thought back to high school where one of my math teachers figured Cartesian coordinates for the vertices of a regular dodecahedron and, looking at one, I wonder what the straightforward strategy is for doing that. I like to think that enough staring and turning would help make it clearer. Now, this is where I wish I had a large desktop system with lots of PCI-e slots for used RTX 3090's or somesuch: it's the kind of thing I'm happy to try idly chatting to some opensource LLM about. It's not as if anything's riding on the answer. Perhaps they're rather better at classic book suggestions than anything analytic though.

I also got to wonder about mobile telephony. ( How might routing work? )

My mention of idly chatting to LLMs reminds me, I have three sizable pending purchases in mind: such a desktop AI system, a small laptop for use while commuting, and a cross-trainer. The interesting question is how to prioritize them though clearly the first there should actually be last while I cross my fingers for the bubble bursting. Also, I'm reluctant to spend too freely until I'm more ahead of the higher-interest debt.

In the meantime, I've found that, as usual, BBC iPlayer didn't exactly help me discover that there's recent Later… with Jools Holland to provide me with a somewhat alternative musical backdrop, albeit a considerably mixed bag of such. I've been enjoying ex-BBC's Stereo Underground recently which is also nicely varied. Given that it often plays the music of my childhood, it makes me wonder: I think of all the energy of especially some of the more punk-ish songs, and how exciting life seemed to me at the time, especially with books filling my head with new intellectual worlds to wrestle with. There's something there I'd be interested to recapture, about possibility and choice, about who I am and what I pursue. I may not quite know which destinations make sense but one of the many wonderful things about R. is how supportive they are.