Like the Metaverse and NFTs before it, it feels near impossible to follow tech news right now without hearing about AI. I understand it can get annoying: The tech is still in development, and everyone’s still figuring out the ethics of generating AI Art, or how to make it not suck. Then, there are the tech CEOs that promise big changes to your workflows, before turning around and introducing tools that rewrite your emails to sound more robotic. It’s tempting to write AI off entirely.
That would be a mistake. Even as an AI skeptic, there are a few key use cases I’ve found where AI shines better than pretty much any other alternative. Even if it doesn’t end up being the magic bullet the tech industry is hoping for, AI can be a lifesaver in the right situation. As an AI skeptic, here’s where I think AI is actually most useful.
When to use AI
I tend to think of AI like I do most automation software—as a tedium killer. While I think AI image generation can often feel like taking existing artwork and changing it up a bit so teacher doesn’t notice, or generative search usually ends up feeling like a rebranded “I’m feeling lucky” button, there are plenty of places where machine learning has already shown its chops.
Even before “AI” became the buzzword du jour, you already saw similar features in tools like the Magic Wand in Photoshop, which saves artists time by letting them smartly select a photo subject in one click instead of having to zoom in and carefully use the selection tool manually. In these situations, no creative decisions are made, the artist doesn’t lose any control, and nobody has their copyright infringed. Instead, the machine learning does something that a search engine or a human couldn’t easily replicate, and much quicker, too. It feels genuinely novel.
That’s the kind of AI I’m interested in. If you’ve been asking yourself “why would I want a robot to write my emails,” here are some problems where AI might just be the best answer out there.
Use AI to change an image’s aspect ratio
Consider this scenario: You take a photo in the wrong aspect ratio, and now you need to crop it to fit it on your website or in your slideshow. It sucks to feel like you’re losing content, which is where AI-powered generative fill plays a part. Available in increasingly more photo editing tools, including Adobe Photoshop and Microsoft’s Cocreator in Paint, Generative Fill uses AI to fill in the missing parts of your photo, so you can make it any aspect ratio you want. Depending on your tool, you’ll be able to do this with a prompt, or by stretching the background that’s already there. You might even have access to both options. But the core principle is the same: unobtrusive AI that keeps more of your non-AI information in the shot.
It’s not always strictly successful, but when a background is simple enough, it can be hard to tell it was ever used. That said, like all AI, it needs some human supervision, especially when applied to photos of people. Take this instance, for example: A tech panelist sent a conference a selfie for promotional use, but it was extended using AI without her consent, resulting in a photo that was more revealing than the original. The conference didn’t intend to alter how she looked—they just wanted to change the photo’s aspect ratio—but the result was still misleading.
AI upscaling
AI upscaling actually predates the recent generative AI trend, as most gamers will tell you they’ve been using it prior to ChatGPT becoming a household name. But since companies have implemented neural chips into more and more devices, AI upscaling is becoming more prevalent than ever.
Used properly, this technique allows movies and photos to look sharper, and even for games to run faster. It uses machine learning to analyze images and add detail that wasn’t in the original shot. This is useful either when processing graphics in real-time, or when working with material that can’t be re-rendered from scratch, like old photos or compressed retro video game graphics where the original source files have gone missing.
For instance, let’s say you have a photo of your grandmother taken over 50 years ago. It’s as detailed as it’s gonna look, but with tools like Super Resolution in Microsoft Photos, you can use AI to increase the resolution by having the machine learning model attempt to fill in the blanks while sticking to the original image’s style as much as possible.
It’s not without controversy, since it can still change the mood of a photo, and when applied to things like video game remasters, could change the art direction when not reigned in enough. There’s also the possibility for AI upscaling to hallucinate or render dreamlike imagery if the source image is too low resolution.
But with proper human curation and strict enough training datasets, plenty of professional products have used AI upscaling to great effect, remastering hundreds or thousands of assets with a much more reasonable time and budget than if the project had been entirely done by hand.
Maybe the place where the average person uses AI upscaling most often, though, is in boosting video game performance. Tools like AMD’s FSR, Nvidia’s DLSS, and Sony’s upcoming PSSR use AI upscaling to run games at a lower native resolution while still looking good enough when displayed on a high resolution TV, which in turn lets them pump out higher frame rates. As techniques like real-time ray tracing become more common, AI might be the best way to utilize the fancy new graphical techniques consoles are sold on while keeping games smooth enough to make them feel like you aren’t sacrificing performance to see all that detail.
When something is just on the tip of your tongue
Often, I find using an AI chatbot to search for information to be less helpful than a search engine. I feel like it takes away my ability to see search results myself and is instead acting as a middle man, as if I’m not grown up enough to understand a Google page. But there’s one situation where search engines can’t help, and that’s when you’re not even quite sure what you’re searching for.
In other words, AI is great for when something is on the tip of your tongue. Say there’s an old movie you remember a scene from, but don’t know anything else about it. In that case, even coming up with the keywords to put into Google can be difficult. There’s a subreddit to get real human help here, but the problem there is you need to hope your post gets noticed and someone responds in a reasonable timeframe.
While you’re waiting, it’s possible AI can answer your question instead. For instance, a few weeks ago, I was searching for a specific Batman comic that I remembered hearing about on YouTube, but couldn’t remember the name of it. It was supposed to have trippy visuals and a dark tone, but since those aren’t exactly unusual themes for Batman, a more famous comic with those themes was all Google would show me. Meanwhile, just searching Google for “Batman comic with psychedelic art,” returned fanart and, ironically, a lot of AI-generated imagery, but not the comic I was looking for. (This also happened if I excluded the result I wasn’t looking for from my search.)
Meanwhile, both ChatGPT and Microsoft Copilot guessed the comic I wanted on the first try. I simply told them, “I’m looking for a Batman comic with trippy, psychedelic art and a dark tone. It is not Arkham Asylum: A Serious House on Serious Earth. I think it was published in the 1980s.”
Bam: I was immediately directed to Batman: The Cult, the comic I was looking for, without having to consult a real-life comic book nerd, or go digging through my YouTube history to find the video where I first heard about the comic.
How AI can help with disability
While I might not want AI to write my emails for me, not everyone is in the same boat. One reason for that? Writing an email might be physically taxing for some people.
Let’s say you have a physical impairment that keeps you from easily typing out words on a keyboard, but you’re otherwise able to understand what you want to say without issue. In this instance, rather than painstakingly sitting down at a desk and hunting-and-pecking your way through multiple paragraphs, you might prefer to give an AI a few sentences on what you want to say, then simply review what it drafts up.
That’s just one example Fredrik Ruben, head of the Dynavox Group, gave me in a call recently. Dynavox makes assistive software and hardware for both physical and mental disabilities, and while he told me that AI has some risks, it’s also giving his company far more ways to help its patients.
Take AI voice cloning. Also available in tools from Apple, this allows those who are at risk of losing their voice to train an electronic replacement to sound more like them. Previously, this required a person to record almost every word in their language, but Ruben says it can now be done in just about 50 sentences. That’s not only a big time saver, but means past recordings can now be used to clone voices, which makes the tool accessible for those who have already lost their voice as well.
Or let’s say you need an electronic voice box to help you speak. AI can learn your speech patterns as you use it, powering more helpful autocomplete options, and letting you converse with those around you more quickly than if you had to manually type out everything you had to say.
All of these are areas Ruben is excited to continue exploring, and outside of our conversation, it’s worth noting that other companies are using AI-generated Live Captions to help the hearing impaired. Still, Ruben did note one area where using AI to help a disabled patient presents a risk.
“AI is quite dangerous if the user has a cognitive disability,” Ruben said. That’s because people with certain cognitive disabilities might not know what to put in a prompt, or how to proofread an AI draft to make sure it says what they wanted to say. Instead, they might just OK whatever the AI writes for them, even if it doesn’t represent their real feelings. In these situations, it would be easy for the AI to “impersonate” the user, misrepresenting them without the user knowing that’s what’s happening. Their friends or family would essentially be talking only to ChatGPT, with little involvement from the patient, and could leave interactions with the wrong impression.
Tools like AI voice cloning also have risks associated with them, like deepfaking, and we still have yet to see whether AI-drafted or edited emails will be as effective as handmade ones at scale. But on an individual level, these AI tools can help level the playing field when used responsibly, even if you don’t see much use for them yourself.
Easily input data into a spreadsheet or transcribe audio
Data entry is one of the most tedious parts of any job. When all you’re doing is copying-and-pasting information from a document into a spreadsheet, there’s little to keep your mind active. And yet, it can still take so long. That goes double if you’re transcribing audio.
Living up to my desires for a tedium killer, then, AI has quickly gotten very good at both of these tasks. You might need an advanced subscription depending on your tool, but simply giving your AI chatbot of choice access to your document and asking for it to be turned into a spreadsheet is surprisingly solid at this point. There are even multiple ways to go about it, from simply using a text-based prompt to ask for a CSV that you can import into a spreadsheet program, or using a third-party tool to automatically take a PDF and turn it into an Excel sheet. If you only have a few data points or all the data you need is available online, you might even be able to get what you need right from the regular chatbot interface, without uploading a document. Take this McDonald’s menu I created using Copilot. (Here’s the actual menu to compare it against.) You’ll need to double check your result, especially if you’re asking the AI to pull information from the internet, but you’ll finally be able to get a rough draft without having to reformat your data manually.
In a similar vein, programs like Otter.ai and now even iOS’ Voice Memos can generate full transcriptions from audio recordings. As someone who got my start in journalism by spending hours pausing and replaying interviews my seniors had conducted to transcribe them, this is a lifesaver. It’s not always strictly accurate, but it’s much faster to edit an AI-generated transcription draft than to simply write one from scratch.
Quickly make trip itineraries
Time for something fun. While AI can help eliminate tedium at work, one of the times I find myself using it while out of office is when I’m planning a vacation. Figuring out what to do when you’re taking a trip can take dozens or even hundreds of Google searches, and an AI chatbot can narrow that down to a single interaction.
Granted, you’re probably going to want to supplement or double-check your AI’s suggestions with your own research, but if you’re at a total blank, asking an AI to plan an example stay at your destination can help you know where to start looking. That way, you can avoid idly scrolling through "travelfluencer" TikToks, or looking at whatever a travel website’s algorithm wants you to see.
Take this Japan trip I asked Copilot to plan, for example. While the results it gave me are a little basic, and I’d need to do my own searching to actually find restaurants or book my stays, the broad strokes are surprisingly similar to a trip I actually took in real life shortly after Japan started easing its pandemic travel restrictions. We were forced to use a travel agency because of when we booked, and yet, this free itinerary isn’t too different from what the paid agency suggested.
Granted, something like the above is just a start, but using AI to draft an example itinerary before you begin more in-depth research can cut down on the amount of work you have to do in total. It’ll also let you tailor your needs to your specific trip, which can make it a bit more flexible than other tools.
Summarize your inbox
Time for me to admit my secret shame. According to my phone’s mail app, I have over 89,000 unread emails. I’ve given up: There’s just no way I’m getting to inbox zero all on my own.
Except now, I don’t have to be on my own. Outlook, Gmail, and Apple's Mail app are either working on or about to roll out AI-powered help for clearing out your inbox, with the latter set to debut this week. Think automatic mail categorization (for Apple Mail), sentence-long recaps of pages-long write-ups, and other tools that can help you get up to speed at a glance. AI will even help you draft up a response.
Like I said earlier in this article, I don’t necessarily need all of these. And I’m still skeptical about whether I should let AI try to relay an email from my boss rather than just reading it directly. But when compared to getting paralyzed by the size of my inbox and just skipping over most of it? Yeah, it should make my skimming much more productive.
I can especially see this being useful when returning to the office after a long trip, or if you’re tired of your phone pinging you during the workday. For instance, Apple Intelligence’s Mail summaries are also being bundled with notification summaries, plus the ability to set up a Focus that only bothers you when the AI determines a notification is urgent. Again, I get a lot of emails, so while I’d be nervous to use that Focus mode when I’m waiting for an important call, I could also see it as a viable alternative to turning off my phone when I find myself getting an annoying notification beep every five to 10 minutes.
That’s where I, as someone who still pines for the days before I knew what a GPT was, tend to use AI most. As someone who still worries about the ethical ramifications of where AI tools gather their datasets and how much energy they consume, I tend to be pretty sparing with when I use them, preferring to employ them only when they either save significant time or do something I couldn’t otherwise do myself. I feel like I’m not alone there. For good or ill, it doesn’t seem like AI is going anywhere anytime soon.
0 commentaires:
Enregistrer un commentaire