Co-Geeking

One of the most prominent topics currently is artificial intelligence. Of course everyone knows—or should know—it’s not real intelligence as we generally understand it. The term seems to have some real staying power, though.

Regardless of what one wants to call the current iterations, machine learning or algorithm-based systems appear to be here to stay. People are starting to figure out what they could be used for in everyday life, not just at the office or lab.

One popular sentiment among creatives goes along the lines of: I want AI to do the dishes for me so I can create, I don’t want AI to create for me so I can do the dishes. I hear you, fellow creatives, and agree! This post talks about what I’d want those systems to do for me. (Eventually. We still must solve several issues, e.g. how to program them without egregious copyright breaches, to mention just one.)

(Please also note: I do realize that computing takes a lot of energy and materials, and that energy production and extraction of minerals aren’t unproblematic processes. Also, the ethics of the current generative models need serious attention; I’m still fuming over Meta’s massive book pirating as revealed by The Atlantic. The point of my post is not to dwell on the problems, however. This is basically just a random, long-winded wish list.)

TL;DR: My preferences boil to complex tasks, specificity instead of generic sludge, and effective analysis of massive amounts of data.

Below is a non-comprehensive list of some specific tasks I want done better, whether by “AI” or not, in no particular order.

In general

• Effective data safety, including safe online banking. Scamming is already a huge problem; I fear it’s just going to get worse.

• Reliable tracking of shipments. Actually, truly, big-boy-pants reliable.

• More dependable long-term weather forecasts.

• Accurate subtitling or captioning for video content. Alt descriptions for images online. Hitch-free voice navigation of all websites, but especially of utilities and essential services. Image recognition and spoken descriptions of real-life environments (e.g. line numbers on approaching buses, trams, or trains) on mobile devices.

• Checking device parameters comprehensively prior to software upgrades. For instance, a month or two ago I had to update my graphics card driver. The OS device manager claimed it was up to date, but the manufacturer’s site actually had a newer version. It would’ve been much more helpful for the OS to be able to check all alternatives for me.

• Autopopulating my preferences into various online interfaces with a click instead of having to enter everything anew every time. The current services I’m using (e.g. for online grocery orders) only do this imperfectly. The variety of order or booking forms and formatting customs certainly doesn’t make it easy as a more universal concept.

• Generative services programmed on material that respects the copyright of human creators. I can draw, sort of, but not well enough for my liking. I’d love to be able to make more pictures of our TTRPG characters in action, for instance, and that means being able to tweak all details in an image. However, I don’t want either poor quality (e.g. irrational or plastic-y images) or to impoverish my artist friends and colleagues. The current plagiarism-machine state of generative “AI” isn’t sustainable.

• Getting a realistic approximation of a sewing project before committing to it. I.e., being able to pick a pattern or plan and try out different fabrics or colors, changing them until I’m happy with the result, and only then cutting the materials.

• More effective aids and devices. E.g. robot vacuum cleaners that behave sensibly and do a more complete job (difficult locations like tops of mouldings, corners, or behind doors, and obstacles like rugs and bases of floor lamps). People with limitations also deserve help that’s actually helpful instead of adding a different selection of problems to their lives. Besides, such aids also help us fuller-abled peeps, as most of us eventually do get sick or old.

• On the Internet, auto-opting out of all legitimate interest options and non-necessary cookies (tracking cookies). This is an existing concept within EU, but I see nothing besides avarice to prevent it being universal. I want to be able to set my preferences once and have those applied automatically on every. Single. Site. and Device. (Then there are the sites that refuse to provide a single-click consent withdrawal button; you have to separately deselect every single damn vendor, of which there can be dozens. There’s no excuse for such disdain for my time. These days I immediately leave those sites and never return.)

• Faster analysis of massive amounts of biological or health data. If interpretative computer systems can aid in, say, creating new drugs and treatments for cancer or the next pandemic, that’s great. (A firm no, though, to leaving all analysis to computer systems alone. Using humans to check computer work and vice versa should be helpful, if the wrinkles can be smoothed out and safety ensured. For instance, if computer systems misdiagnose female or BIPOC patients, obviously it’s a huge problem.)

• Looking for exoplanets and other objects of interest in space. (Planet 9, anyone?) I really cannot see a downside to sifting through massive amounts of scientific data with AI, as long as it’s not the sole operating principle. Here, again, actual human intellect is crucial.

Specifically as relates to online searching I’d prefer

• Collating specified cities’ cultural or leisure offerings (theater programs, sports events, concerts, etc.) and compiling all desired matches for a given period. As a bonus, reliable information on (public) transit options to and from said cities.

• Weeding out scammers from all search results, and instead getting reliable reviews of products or service providers. But I gather this is very, very difficult to do without human judgment. Also, not necessarily feasible as long as advertising pays the bills for search engine companies.

• Reasonably-sized batches of relevant search results, not millions of iffy hits. Instead of algorithmically generated “answers” (which sometimes are just garbage), I want an engine that excludes the completely and utterly wrong data, like ads pushing a product with merely a token similarity to my search. Another “helpful” practice I want gone is changing the world order of my input (like Cliff Price, a man’s name, for price cliff, an economics term). Also, I want a working advanced search again. But clearly this isn’t feasible as long as advertising pays the bills or a human-level intellect is not sifting the results.

On social media

• Effective measures against account hijacking and scamming.

• Instead of fake profiles prioritizing emotional manipulation (fake “people”), I want transparency and fairness—same rules for everyone, clear grounds given for breach-of-terms notices, and actually effective procedures against dogpiling, harassment, stalking, doxxing, swatting, etc.

• Flexible controls for your accounts’ safety, privacy, feed, and each post’s audience. I know that the fediverse, for example, already does some of this, and I’m taking steps to move away from stupid algorithm-based spoonfeeder services. I want to decide what I see. There’s no way a stupid, limited algorithm can ever comprehensively cover my interests. I’m open to suggestions, but only when I want, not as a stupid, forced, omnipresent solution.

***

I realize most of these wishes are challenging. Particulars always require actual understanding, and current algorithmic tools do not have that. As soon as you need something concrete and specific, you need a human, be it tweaking one detail in an image or adjusting pre-existing lines of code.

As a linguist, one of the problems I see is a mishmash of terminology. (And I’m sure the situation as I see it is complicated by my news-reading habits, namely daily doses of Finnish, English, and Swedish.) It seems that currently the term AI is used to try and cover too much, that collectively we are struggling with the (most) commonly accepted definition(s).

Put over-dramatically, it seems that right now just about any application using the slightest whisper of anything even remotely algorithmic gets slapped with the AI label, perhaps in the hopes of more eyeballs and/or money. (Incidentally, instead of large language models the funniest short explanation for LLM I’ve seen is large liability models—*snort* Ha!)

Academic research on AI has several subfields, of course, each with their own foci, tools, and terminology, but their specificity falls off when these different flavors are filtered into the popular parlance. Media now routinely talks of generative, analytic or interpretative, optimizing, and recognition systems, but not necessarily in those terms.

I don’t even exactly know what the rabid “AI” peddlers are trying to lure buyers with these days. Increased productivity, probably, when in fact there are grounds to considering AI to be counterproductive for some tasks, not to mention the potential for evidence hacking, other forms of false science, or false representations.

Still, some people are going to try and maintain that blah, broad-brush solutions should fit everyone, or that algorithmically generated slop is the be-all and end-all solution to everything.

At the moment, rather than working on factual improvements or real results, it’s so much faster and cheaper to keep churning out generic crap which approximates solutions but doesn’t really answer anyone’s needs. However, insisting that featureless sludge is as good as nuanced design not only misunderstands how the world works, it misconstrues how humans work. It’s like offering a bandaid for a broken leg and expecting gratitude for your invaluable help.

I have no doubt generative and interpretative tools will yield some real advantages. In fact, I know they already have, for example when used to translate cuneiform texts into English, read charred papyri from Herculaneum without destroying the scrolls, improve efficiency and precision in astronomical dataset analysis, or predict treatment responses in cancer patients. Closer to everyday life, automated customer call transcriptions, for instance, save a lot of time and effort.

In a way it feels like the early days of wide-spread Internet. Few people really know what to do with machine learning or algorithm-based systems, or what they could do particularly well without creating new, disproportionate problems, so there’s scrambling and flailing and a whole bunch of uncredited claims along with the creditable ones.

My personal take is that the new hot commodities rising out of this current sloppy soup are going to be reliability, specificity / customizability, and authenticity. Systems that confabulate inaccurate “data” (if we can even call it that) are a waste of time and money, as are inflexible solutions. People want authentic stories and art—instead of elevator music at a gig, we’re clamoring for genuine emotions and connections.

Incorporating machine learning into our lives is a tremendous change that’s going to take time. Exactly how long and with precisely what consequences is anyone’s guess, though.

Here there be opinions!