I can see some minor benefits - I use it for the odd bit of mundane writing and some of the image creation stuff is interesting, and I knew that a lot of people use it for coding etc - but mostly it seems to be about making more cash for corporations and stuffing the internet with bots and fake content. Am I missing something here? Are there any genuine benefits?
You must log in or register to comment.
Most email spam detection and antimalware use ML. There are use cases in medicine with trying to predict whether someone has a condition early
It’s also being used in drug R&D to find similar compounds like antimicrobial activity, afaik.
Medical use is absolutely revolutionary. From GP’s consultations to reading tests results, radios, AI is already better than humans and will be getting better and better.
Computers are exceptionally good at storing large amount of data, and with ML they are great at taking a lot of input and inferring a result from that. This is essentially diagnosing in a nutshell.
I read that one LLM was so good at detecting TB from Xrays that they reverse engineered the “black box” code hoping for some insight doctors could use. Turns out, the AI was biased toward the age of the Xray machine that took each photo because TB is more common in developing countries that have older equipment. Womp Womp.
A large language model was used to detect TB in X-ray? Do you not just mean Machine Learning?
What’s TB?
Tuberculosis
Tuberculosis
Watch any video at random by John Green (vlogbrothers, and author of several successful books that I haven’t read) and you’ll know more than you could ever hope about TB.
That’s super interesting, TIL
I hadn’t considered this. It’s interesting stuff. My old doctor used to just Google stuff in front of me and then repeat the info as if I hadn’t been there for the last five minutes.
They are the greatest gift to solo-brainstorming that I’ve ever encountered.
_ /\ _
AI is a very broad topic. Unless you only want to talk about Large Language Models (like ChatGPT) or AI Image Generators (Midjourney) there are a lot of uses for AI that you seem to not be considering.
It’s great for upscaling old videos: (this would fall under image generating AI since it can be used for colorizing, improving details, and adding in additional frames) so that you end up with something like: https://www.youtube.com/watch?v=hZ1OgQL9_Cw
It’s useful for scanning an image for text and being able to copy it out (OCR).
It’s excellent if you’re deaf, or sitting in a lobby with a muted live broadcast and want to see what is being said with closed captions (Speech to Text).
Flying your own drone with object detection/avoidance.
There’s a lot more, but basically, it’s great at taking mundane tasks where you’re stuck doing the same (or similar) thing over, and over, and over again, and automating it.
Yeah that’s interesting.
AI has some interesting use cases, but should not be trusted 100%.
Like github copilot ( or any “code copilot”):
- Good for repeating stuff but with minor changes
- Can help with common easy coding errors
- Code quality can take a big hit
- For coding beginners, it can lead to a deficit of real understanding of your code
( and because of that could lead to bugs, security backdoors… )
Like translations ( code or language ):
- Good translation of the common/big languages ( english, german…)
- Can extend a brief summary to a big wall of text ( and back )
- If wrong translated it can lead to that someone else understands it wrong and it misses the point
- It removes the “human” part. It can be most of the time depending on the context easily identified.
Like classification of text/Images for moderation:
- Help for identify bad faith text / images
- False Positives can be annoying to deal with.
But dont do anything that is IMPORTANT with AI, only use it for fun or if you know if the code/text the AI wrote is correct!
Adding to the language section, it’s also really good at guessing words if you give it a decent definition. I think this has other applications but it’s quite useful for people like me with the occasionally leaky brain.
I have sometimes the same issue!
Actually the summaries are good, but you have to know some of it anyway and then check to see if it’s just making stuff up. That’s been my experience.
AI is a revolution in learning.
Very true. I learned how to code surprisingly fast.
And even the mistakes the AI made was good, because it made me learn so much seeing what changes it did to fix it.
It’s sped up my retouching workflows. I can automate things that a few years ago would’ve needed quite a lot of time spent with manual brush work.
Also in the creative industries, it’s a massive time saver for conceptual work. Think storyboarding and scamping, first stage visuals that kind of thing.
One of the better uses I’ve heard of is in search and rescue type situations. Using AI to find specific items, people or anomalies on a map or video feed can be helpful.
An example regarding wildfires:
Someone I know recently published in Nature Communications an enormous study where they used machine learning to pattern match peptides that are clinically significant/bioactive (don’t forget, the vast amount of peptides are currently believed to be degradation products).
Using mass spectrometry, they effectively shoot a sawed off shotgun at a wall then using machine learning to detect pellets that may have interesting effects. This opens up for new understanding in the role peptides play in the translational game as well as a potential for a huge amount of new treatments for a vast swathe of diseases.
Sounds similar to some of the research my sister has done in her PhD so far. As I understand, she had a bunch of snapshots of proteins from a cryo electron microscope, but these snapshots are 2D. She used ML to construct 3D shapes of different types of proteins. And finding the shape of a protein is important because the shape defines the function. It’s crazy stuff that would be ludicrously difficult and time-consuming to try to do manually.
There was an interesting talk about it at the last CCC. However, I also remember a few reports casting doubts upon the results of this method.
Our software uses ML to detect tax fraud and since tax offices are usually understaffed they can now go after more cases. So yes?
Maybe you only do an “odd bit” of mundane writing and the image/music generation is a gimmick, but a lot of the modern world is mundane and pays people lots of money for mundane work. E.g. think of those internal corporate videos which require a script, stock photography and footage, basic corporate music following a 4 chord progression, a voiceover, all edited into a video.
Steve Taylor is most famous for being the voiceover for Kurzgesagt videos, but more generally he’s a voiceover artist that features in lot of these boring corporate videos. This type of content has such high demand there is an entire industry dedicated towards it, which seems well suited to AI.
This does raise further ethical/economical issues though, as most people in these creative industries actually require income from this boring work to get by.
Don’t limit your thoughts to just generative AI, which is what you are talking about. Chat bot and media generation aren’t the only uses for AI (by which I mean any trained neural network program that can do some sort of task.
Motor skills
AI can solve learn to solve the marble maze “Labyrinth” much, much faster than a human, and then speedrun it faster than any human ever has. Six hours. That’s how long it took a brand new baby AI to beat the human world record. A human that has been learning hand-eye coordination and fine motor control all of it’s life, with a brain which evolved over millions of years to do exactly that.
No special code needed. The AI didn’t need to be told how balls roll or knobs turn, or how walls block the ball. It earned all of that on the fly. The only special code it had was optical and mechanical. It knew it had “hands” in the form of two motors, and it knew how to use them. It also had eyes (a camera), and access to a neural network computer vision system. When the AI started taking illegal shortcuts, and they had to instruct it to follow the prescribed path, which is printed on the maze.
Robots could in work factories, mines, and other dangerous, dehumanizing jobs. Why do we want workers to behave like robots at a factory job? Replace them with actual robots and let them perform a human job like customer service.
Think of a robot that has actual hands and arms, feet and legs, and various “muscles”. We have it learn it’s motor control using a very accurate physics system on a computer that simulates its body. This allows the AI to learn at much faster speeds than by controlling a real robot. We can simulate thousands of robots in parallel and run the simulations much faster than real time. Train it to learn how to use it’s limbs and eyes to climb over obstacles, open doors and detain or kill people. We could replace police with them. Super agile robot cops with no racial bias or other prejudices. Arresting people and recording their crimes. Genuine benefit.
Computer Vision
AI can be trained to recognize objects, abstract shapes, people’s individual faces, emotions, people’s individual body shape, mannerisms, and gait. There are many genuine benefits to such systems. We can monitor every public location with cameras and an AI employing these tools. This would help you find lost loved ones, keep track of your kids as they navigate the city, and track criminal activity.
By recording all of this data, tagged with individual names, we can spontaneously view the public history of any person in the world for law enforcement purposes. Imagine we identify a person as a threat to public safety 10 years from now. We’d have 10 years of data showing everyone they’ve ever associated with and where they went. Then we could weed out entire networks of crime at once by finding patterns among the people they’ve associated with.
AI can even predict near future crime from an individual’s recent location history, employment history, etc. Imagine a person is fired from his job then visits a gun store then his previous place of employment. Pretty obvious what’s going on, right? But what if it happens over the period of two weeks? Difficult for a human to detect a pattern like this in all the noise of millions of people doing their everyday tasks, but easy for an AI. Genuine benefit.
Managing Production
With enough data and processing power, we can manage the entire economy without the need for capitalism. People’s needs could be calculated by an AI and production can be planned years ahead of time to optimize inputs and outputs. The economy–as it stands today–is a distributed network of human brains and various computers. AI can eliminate the need for the humans, which is good because humans are greedy and neurotic. AI can do the same job without either. Again, human’s are freed to pursue human endeavors instead of worrying about making sure each farm and factory has the resources it needs to feed and clothe everyone. Genuine benefit.
Togetherness
We will all be part of the same machine working in harmony instead of fighting over how to allocate resources. Genuine benefit!
Says the AI…
deleted by creator
Train it to learn how to use it’s limbs and eyes to climb over obstacles, open doors and detain or kill people. We could replace police with them. Super agile robot cops with no racial bias or other prejudices. Arresting people and recording their crimes. Genuine benefit.
I got as far as ai cops and became sceptical. Like, yeah, sure, but what you’re describing isn’t just a robot being controlled by an AI, it’s also the ai making decisions and choosing who to pursue and such, which is a known weakness right now.
And then you let them kill people.
Well, yeah. They are told to put down their weapon. They get 20 seconds to comply. If they don’t, they get killed.
Well written!
Don’t discount the generative AI either!
Language generating AI like LLMs: Though we’re in early stages yet and they don’t really work for communication, these are going to be the foundation on which AI learns to talk and communicate information to people. Right now they just spit out correct-sounding responses, but eventually the trick to using that language generation to actually communicate will be resolved.
Image/video/music generating AI: How difficult it is right now, for the average person to illustrate an idea or visual concept they have! But already these image generating AI are making such illustration available to the common person. As they advance further and adjusting their output based on natural conversational language becomes more effective, this will only get better. A picture paints a thousand words…and now the inverse will also be true, as anyone will be able to create a picture with sufficient description. And the same applies to video and music.
That said I love your managing production point. It’s something I e been thinking too - centrally planned economies have always had serious issues, but if with predictive AI we can overcome the problems by accurately predicting future need, the problems with them may be solvable, and we can then take advantage of the inherent efficiency in such a planned system.
That’s funny because the whole post was sarcastically outlining a distopian nightmare.
If that kind of stuff was actually to become real, some dictator would take control of it and subjugate the entire country, or world… forever. There’d be no way to resist that level of surveillance or machine policing.
I think each one of those dystopian ideas can be done in a safe and humane way, but needless to say it is not the current trajectory.
I use it daily to generate basic Perl scripts that I cant be bothered to write myself. It’s fantastic.
Machine learning is important in healthcare and it’s going to get better and better. If you train an algorithm on two sets of data where one is a collection of normal scans and the other from patients with an abnormality, it’s often more accurate than a medical professional in sorting new scans.
As for the fancy chatbot side of things, I suspect it’s only going to lead to a bunch of middle management dickheads believing they can lay off staff until the inevitable happens and it blows up in their faces.
Lots of boring applications that are beneficial in focused use cases.
Computer vision is great for optical character recognition, think scanning documents to digitize them, depositing checks from your phone, etc. Also some good computer vision use cases for scanning plants to see what they are, facial recognition for labeling the photos in your phone etc…
Also some decent opportunities in medical research with protein analysis for development of medicine, and (again) computer vision to detect cancerous cells, read X-rays and MRIs.
Today all the hype is about generative AI with content creation which is enabled with Transformer technology, but it’s basically just version 2 (or maybe more) of Recurrent Neural Networks, or RNNs. Back in 2015 I remember this essay, The Unreasonable Effectiveness of RNNs being just as novel and exciting as ChatGPT.
We’re still burdened with this comment from the first paragraph, though.
Within a few dozen minutes of training my first baby model (with rather arbitrarily-chosen hyperparameters) started to generate very nice looking descriptions of images that were on the edge of making sense.
This will likely be a very difficult chasm to cross, because there is a lot more to human knowledge than thinking of the next letter in a word or the next word in a sentence. We have knowledge domains where, as an individual we may be brilliant, and others where we may be ignorant. Generative AI is trying to become a genius in all areas at once, and finds itself borrowing “knowledge” from Shakespearean literature to answer questions about modern philosophy because the order of the words in the sentences is roughly similar given a noun it used 200 words ago.
Enter Tiny Language Models. Using the technology from large language models, but hyper focused to write children’s stories appears to have progress with specialization, and could allow generative AI to stay focused and stop sounding incoherent when the details matter.
This is relatively full circle in my opinion, RNNs were designed to solve one problem well, then they unexpectedly generalized well, and the hunt was on for the premier generalized model. That hunt advanced the technology by enormous amounts, and now that technology is being used in Tiny Models, which is again looking to solve specific use cases extraordinarily well.
Still very TBD to see what use cases can be identified that add value, but recent advancements to seem ripe to transition gen AI from a novelty to something truly game changing.
