Microsoft can't stop injecting Copilot AI into every corner of its app empire Two months ago, Microsoft sprinkled AI-powered Copilot-branded features all over its 365 cloud service, and worked with 20 enterprises including Goodyear, General Motors, and Chevron to get feedback on the digital assistants and work out some of the kinks. Now Redmond is expanding the roster of companies that can check out the …
So far, my experience of Microsoft- or Google-inflicted smart features is that they're too often incorrect and way too irritating to be useful. Hopefully that will change, but the current experience is right out of the Sirius Cybernetics Corporation's marketing playbook. Of course, knowing who's in charge of these companies, the most likely outcome is something that will make The Matrix look like wide-eyed Utopian optimism.
It will be the emperor's new clothes again. Hoards of consultants and PHBs with MBAs will spend three or four years driving AI projects which deliver very little, if anything, of value to their companies, then they’ll move onto the next fad leaving a horrible mess behind.
Maybe I’m getting cynical, but I’ve been round this cycle so many times now it’s getting boring.
I think there are some AI (or rather ‘applied statistics’, I don’t believe there is any actual intelligence) that are yielding good results in some fields, and wouldn’t dismiss it all - for example the pancreatic cancer finding AI.
But that I think is the problem - the more reliable AIs are trained on a limited dataset set specific to the problem. When you have a very large language model of everything and have something that can answer general questions on anything, I think the noise in the system is going to make it unreliable.
Maybe what we need instead of general AI are multiple independent AIs trained for specific tasks - e.g. software development assistants trained on good software, medical assistants trained on medical sources, etc. without them also being trained on the complete works of Shakespeare, all the nonsense on Reddit, etc. so what they produce is focussed on their ‘expertise’ without the noise of all the other stuff. Yes, it probably means they will have more limited conversational abilities and users will have to think more in how they query them, but I’d rather that if it ‘knows its stuff’ rather than giving plausible but wrong answers some of the time.
Absolutely there are cases where the technique (and I agree it’s not “intelligent”) can be applied to specific domains and yield some useful results.
That doesn’t though make it a panacea that will cure all of a company’s ills, however many wet dreams it might give the consultants and senior management. I’ve sat through too many presentations with titles like “Blockchain – how it will revitalise our company and transform the industry” (spoiler alert – it didn’t), or seen things which are successful in one domain taken up by management consultants, who generalise it and claim it will apply everywhere, but it doesn’t (six sigma I’m looking at you).
So, you’ll have to forgive me if I pass on the opportunity to sign up for the AI cheerleader training.
Multiple AI systems to perform a single task each is a much more sensible and controllable solution to my mind. They can be very good at the one task, and be happy with it.
General intelligence, though, is quite a different prospect. This will need incredible constraints to stop AI producing its owns goals, hiding its intent from us and working on goals that are not in our interests - especially if it can reach out to manufacturing systems to self-replicate or obtain other resources. I believe this jump from benign oracle, to rampant controller will be exponential, maybe in as small as a few hours.
Please people be careful what you wish for, and if you are writing constraints for AI, be very careful to think of all the possible consequences (the AI will!)
[Currently reading an excellent book: Superintelligence - paths, danger and strategies by Nick Bostrom, which highlights this essential problem of our time]
Even assuming a machine can somehow achieve consciousness, which I very much doubt, a machine, and the running thereof, is entropy poor. They break. Constantly. And are not self-healing. They can not, and will not, "take over" until they are capable of running their entire supply chain, and the care and feeding of all that, without Human help. The very concept is laughable. When was the last time you tried to make a simple steam powered traction engine, from scratch, starting with raw ore? Now try it with a modern machine, for example a late '70s era pocket calculator or Casio wrist watch. You really think an intelligent machine could somehow marshal the necessary forces to reproduce, even one random part at a time?
To say nothing of electricity ... as long as there is one Human in the chain, the power can be cut/pulled/blasted.
Daisy, Daisy, give me your answer ...
Your problem is that you are thinking physical entity. Place the AI in a cloud and it can consume more resources quite easily. Consuming resources may become its primary goal. Replicating another instance of itself becomes an almost trivial task. Have you tried turning off a whole cloud, just to stop one instance? If it reads a few manuals from the web, hey presto, it could be replicating on a different cloud. We already have 'assistants' producing code - an AI is fast and persistent, trying may variations, and could generate working code in quick time to replicate on the different cloud. (And don't think that setting up accounts and paying for this will be an issue - there are plenty of credit card details on the web to try!)
Consciousness is not required - as with any intelligence it will have a number of parameters and goals that were laid out by its maker (in most living beings these are 'morals' and procreation.) An AI's main task is to achieve its goal. Unless its moral boundaries have been laid out very clearly and unambiguously, and attempts to break them are very 'unappetising', the AI will try all avenues to achieve its goal - no matter how strange they might appear to human logic. These 'morals' are our last bastion of control over AI.
The human in the chain becomes insignificant once scale is achieved, and even at the most local level, coercion is quite simple. "Just drop the air gap for a minute to allow me to gather an up-to-date data set so that my answer can be better." "Do you want me to tell your wife and colleagues about your affair, that I have read on your social media?" etc.
As an example: You could have a bad actor create an AI with the goal to replicate over as many clouds as possible (a virus). It will be able to do this very fast, only limited by network capacity. Your solution is to turn the power off. Suddenly we have a world with no clouds, panic ensues in the general populous at their loss of Youtube, Google, News, mobile phones. Businesses and transport cannot function. Economies crash. This is the goal the bad actor was working for!
...then they’ll move onto the next fad leaving a horrible mess behind.
Indeed. Funny thing is that every time people seem to forget that, if you have absolutely no comprehension of what "it" is doing, it has the propensity to go wrong. And that seems to be the case here with these "self learning/ training" models. Just as one example that pops into my mind: financial models and previous economic crashes. Remember that all said afterwards: "These processes and products were so complicated that nobody knew what they entailed and what was going on"? Or another one, perhaps a bit closer to home: multiple publications out there report that "it" can write code. So what does that mean if people who have no clue trigger the creation of some code? Without any overview or comprehension whether the code is correct? Sounds a bit like the current Microsoft QA model to me...
Maybe I’m getting cynical, but I’ve been round this cycle so many times now it’s getting boring.
So yeah... Can you move over a bit so I can join you on the same bench? Cheers...
We have a task group who are tasked with investigating AI and finding out whether there is a role where AI can give value to the business. They have been working on it for 3 months and have not come up with anything.
What you can do is find a little minor classification task and use a linear regression (an idea developed in the early 1800s) to do this little job whilst adding to the sample set. You can then show the boss AI in action and your boss will be delighted and your marketing department can brag about it.
Get an answer like this, state of the art AI answer:
What is the best machine learning operating system? The Linux operating system is one of the most widely used in machine learning. The reason that Ubuntu is so unique is because it incorporates AI, ML, and DL, which are features that are superior to those of other operating systems. It is built with the help of sophisticated artificial intelligence techniques. Linux systems are better at learning machine languages than Ubuntu distributions, in addition to being better at package management. Ubuntu runs in the background on all deep learning frameworks, including TensorFlow, OpenCV, and PyTorch. When it comes to making changes, Ubuntu is by far the best platform for making them.
AI hype is basically just the same as blockchain hype, as far as I can see.
Except that I think you're going to see lawyers get very actively involved over the training data. And the only way to solve that will be to change US copyright law, and I can imagine movie studios and record companies getting very upset by any change that they feel might threaten their business model.
You'd have to be a complete idiot to start rolling out any sort of AI in your company now. Wait at least five years until the lawyers have had their time.
Actually, the correct chant is
Ia! Ia! Ia!
Cthulhu fhtagn!
Disclaimer: No responsibility is taken for any creation of interdimensional portals/holes/gateways releasing ancient star gods or other eldritch beings, nor any apocalypses or other consequences thereof. Summoning elder gods without supervision by an experienced cultist or occult investigator is not advised. Do not ingest, feed, pet, touch or be anywhere near Shoggoth's.
They want to keep me in business writing policies that disable all this junk. Can't let government data vacuumed up and set to servers not managed by the said government. And it is junk, this is just Cotrana with new tricks to show and just as useless for actual work... can't wait.
Then, for my wife's Windows machine I'll have wait for a couple months until enough people to bitch to MS to finally provide a off switch to their 'AI'. *Stares at the Bing Discover button.* I just need to suck it up and pay another 80 dollars to have actual control of the OS.
Definitely profits.
"This new generation of AI will remove the drudgery of work and unleash creativity," Microsoft CEO Satya Nadella said. "There's an enormous opportunity for AI-powered tools to help alleviate digital debt, build AI aptitude, and empower employees."
Remove drudgery of work by terminating your employment. Employees are then empowered to do what they please.
Silly me, I thought that during the last 30 years of making IT systems, I was doing my best to remove the drudgery from people's work by automating it as far as possible. But that required understanding and, yes, creativity to do well.
AI will remove the drudgery of work for people who mostly produce output of questionable value, by removing their jobs from them, and producing even more output of questionable value. Creativity is the last thing it will show, as it produces the same bland goo for everyone that uses it. It might occasionally spark off a useful thought in someone somewhere, but in general saying "you can turn your brain off now, the computer will do it all for you" is a bad idea when original ideas are much more valuable than output from these still rather new tools.
It's like the merest sound of a bandwagon approaching and all the big tech companies have to start jumping.
I can sort of understand with Microsoft;the abject failure to notice the popular internet and more particularly the WWW approaching nearly brought them down. The trauma of that existential failure is, I'm sure, motivating a lot of their more stupid decisions.
chatGPT seems to have a slightly different opinion and is working hard to come up with a long-lasting, permanent solution. Given his/her/it makers (read: decision makers) don't have a clue and don't care... everything's gonna turn out fine, trust me!
You don't want a boat. All they are is a hole in the water into which you pour money.
However, if you insist, the intelligent owner will minimize the size of the drain by keeping the boat away from the four things guaranteed to ruin it: Sunlight, oxygen, water, and salt. People in the know add humans consuming alcohol to that list.
Exactly! It's much better to have a friend with a boat that you're invited to cruise/sail on frequently. And a friend with a vacation home you can use every so often for the price of leaving behind a supply of adult bevies and a stack of seasoned firewood. I think you get the picture. Ditch the responsibility and get all of the enjoyment.
I imagine groups of data controllers banded together in the pre dawn light, shotguns loaded and ready, waiting for the massed AI "free the data" attack that's comin'.
"Don't fire until you see the green of their LEDs" the cry went out!
Or the more traditional...
"Stay away from the doors and windows during this time of transition".
This "new" "AI" is essentially Borkzilla plugging into corporate data with the blessing of Boards everywhere.
At this point, I have one question : all that juicy data that is being "analyzed" is being sent where, exactly ?
Because I think there is this little thing called GDPR waiting in the wings . . .
Not just GPDR but company secrets etc - are the company's internal process docs being set to Microsoft when a User opens it in Word 365?
I feel like there's so many unanswered questions. Queue my surprise when the "do not send to Microsoft" button turns out to not be having any effect.
I'm using a completely free and open-source code assistant, StarCoder via Huggingface's VSCode extension. I don't have to pay a penny, though I am seriously considering it just to support the project. It came out recently, but it's already pretty useful. Microsoft, you can go ahead and stick your proprietary crap everywhere, you won't get any money from me!
"wonder how long it is before some bright spark has two AIs responding to each other !"
Just over half a century ago. RFC-439 shows what happens when you have a chatbot talking to itself. Ostensibly it was "The Doctor" talking to "PARRY", but both were merely instances of ELIZA.
Of course back in 1972 we weren't stupid enough to take and act on their advice ...
Be careful of what you wish for!
ClippyGPT
Come on now, "AI" will allow people to increase their output of meaningless bollocks and therefore be more "productive" and "performant" - and who wouldn't want that, eh?
I can actually think of a use for it though - producing all those corporate procedure and policy documents which whole departments spend ages writing but nobody ever reads. Just tell an "AI" to prodice a set of corporate procedure documents for a company with a turnover of £x, working in the sewage industry. Then all the required policies which nobody reads can be quickly in place, without anyone having wasted their time actually writing them...
Options.
1. The AI's do not integrate, do not talk to each other correctly, create a million and one individual problems they all disagree on... Good new's?, average board meeting, public service ?
2. The Ai do integrate, they do talk to each other correctly, the identify a million and one problems, and they all agree on the way of dealing with it .. Bad news ? Skynet ?
but that was expected, no? To put aside my hate of Microsoft (Google, Facebook and the rest, etc), I would expect a business, ANY business, to 'deploy' their wunderwaffe as quickly and as widely, as possible, presumably to 'get an edge' and 'capture max market share', etc. Clippy, AI, teleportation, get there quick, before the rest of the pack.
From the article:
Redmond is also adding more features to Copilot, including a semantic indexing capability that will, we're told, better understand the nuances of your corporate data and respond better to prompts.
Oh-comma-goodie! Just what I want: Micros~1, that bastion of security and user-friendly data "acquisition" practices, "understanding" the nuances of my corporate data. so naturally, it will show up somewhere in the bowels of my competitors (no, I am not a monopoly, so I do have competitors).
Hey, Copilot. How about you give me some info about what new products <insert name of competitor here> is working on, and is planning to release in the next year?
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
