

Satya Nadella Argues AI's True Value Will Come When It Finds Killer App Akin To Email or Excel 95
Microsoft CEO Satya Nadella argues that AI's success should be measured by its impact on economic growth rather than achieving artificial general intelligence (AGI), emphasizing that true progress will come when AI finds a transformative application akin to email or Excel. The Register reports: "Us self-claiming some AGI milestone, that's just nonsensical benchmark hacking," the chief executive said during an appearance on podcaster Dwarkesh Patel's YouTube show this month. Nadella thinks a better benchmark for AI's success should be its ability to boost a country's gross domestic product. "When we say: 'Oh, this is like the industrial revolution,' let's have that industrial revolution type of growth. That means to me, 10 percent, seven percent for the developed world. Inflation adjusted, growing at five percent, that's the real marker."
Nadella suggested that growth hasn't eventuated because it's going to take time before folks understand how to use AI effectively, assuming they find a use for it -- just as it took some years for the personal computer to find its feet. "Just imagine how a multinational corporation like us did forecasts pre-PC, and email, and spreadsheets. Faxes went around, somebody then got those faxes and then did an inter-office memo that then went around, and people entered numbers, and then ultimately a forecast came out maybe just in time for the next quarter," Nadella explained. "Then somebody said: 'Hey, I'm just going to take an Excel spreadsheet, put it in an email, send it around, people will go edit it, and I'll have a forecast.' The entire forecasting business process changed because the work artifact and the workflow changed. That is what needs to happen with AI being introduced into knowledge work," the CEO said. [...]
"Don't conflate knowledge worker with knowledge work," he said. "The knowledge work of today could probably be automated, [but] who said my life's goal is to triage my email?" Instead, he argues AI agents will allow workers to focus on higher-value tasks. Whether this is actually how it'll play out, or whether enterprises will take this as an opportunity to reduce costs by cutting staff remains to be seen. ... "Today, you cannot deploy these intelligences unless and until there's someone indemnifying it as a human," he said.
Nadella suggested that growth hasn't eventuated because it's going to take time before folks understand how to use AI effectively, assuming they find a use for it -- just as it took some years for the personal computer to find its feet. "Just imagine how a multinational corporation like us did forecasts pre-PC, and email, and spreadsheets. Faxes went around, somebody then got those faxes and then did an inter-office memo that then went around, and people entered numbers, and then ultimately a forecast came out maybe just in time for the next quarter," Nadella explained. "Then somebody said: 'Hey, I'm just going to take an Excel spreadsheet, put it in an email, send it around, people will go edit it, and I'll have a forecast.' The entire forecasting business process changed because the work artifact and the workflow changed. That is what needs to happen with AI being introduced into knowledge work," the CEO said. [...]
"Don't conflate knowledge worker with knowledge work," he said. "The knowledge work of today could probably be automated, [but] who said my life's goal is to triage my email?" Instead, he argues AI agents will allow workers to focus on higher-value tasks. Whether this is actually how it'll play out, or whether enterprises will take this as an opportunity to reduce costs by cutting staff remains to be seen. ... "Today, you cannot deploy these intelligences unless and until there's someone indemnifying it as a human," he said.
Its killer app... (Score:2)
Re:Its killer app... (Score:5, Insightful)
And I never received your five bullets.
Re: (Score:1)
A more benevolent app would be replacing all those long phone queues and press 1 for x, 2 for y... n for whatever phone menus with an AI that could actually provide reasonable answers and help for your problems at a similar to what questions in Chat GPT can already provide, with a callback option to talk to a human if you really needed it and couldn't resolve your query using the AI.
Re: (Score:2)
Probably the usual suspects - if they haven't already bankrupted themselves out of business - like Computer Associates, Symantec, etc.
Re: Its killer app... (Score:2)
Microsoft seems to be desperate to find new ideas to promote for coming versions of their software.
It's just that when things goes sour it's the big ones that falls hardest.
Microsoft today is like IBM was in the 80's, a company trying to find a purpose for the future. Nobody cares about IBM today except for a few banks and stagnant large corporations like GM.
Tax Dodge (Score:2)
Re: (Score:2)
Mod FP funny, too.
Not supposed to say it out loud (Score:5, Insightful)
This sounds like an admission that generative models in their current form are overhyped - not the transformative panacea they're sold as. I don't think he was supposed to admit that publicly...
Re: (Score:2)
I suspect this is a reaction to DeepSeek disrupting all of the existing AI business models.
I worked for Microsoft less than a year ago, in a department nowhere even remotely close to AI, and at that time our regular engineering surveys from leadership had additional questions whether we were finding ways to educate ourselves on AI and incorporate AI into our business. Entire orgs were basically being graded on their successful incorporation of AI. It was seen as that mission-critical to the company. To hear
Re: (Score:3)
Re: Not supposed to say it out loud (Score:3)
Re: (Score:1)
It's already useful. Future elites, current university students are now around 90% using it for their work. So they already found useful for it to do.
The issue is integrating it further, and who gets integration right. For example, researching things on the internet with AI search agent is amazing. Instead of spending hours searching and collating the internet, you can just get a general summary in about 15-20 seconds. Massive productivity boon, but you need to understand that refining this data further int
Re: (Score:3)
So now we just need an AI that can do the professor / TA's job, and we can just sic the AI's on each other and start up the diploma printing press!
Re: (Score:1)
Yes yes, there's an intellectual elite that are keeping intelligence away from you. Very funny.
Oh this isn't supposed to be a self deprecating joke...?
Ah.
Re: Not supposed to say it out loud (Score:2)
There's nothing that checks if what the AI returns is correct.
An AI also has a problem being creative. It can do random things, but it's not able to discard those that are bad. "Greatest kitchen robot in the world, just insert a live animal here and in an hour it'll give you a dinner. Then you'll have a curious cat or toddler..."
Re: (Score:2)
What I actually explained that point of education is to select future elites. But I suppose your attempt at irony actually hitting the bullseye instead of telling.,
Re: (Score:2)
Person making the query is the one checking.
This is in fact one of the major points of "learning how to use AI". People who don't understand how to use it don't check, and don't even seem to have intuition to check. Those that adopted the tech successfully not only know to check, but they have developed good intuition on where the potential errors will be so they don't need to check everything, just specific points.
Personally I find that turning on "reasoning" to visibility is what helps me do the checking.
Re: Not supposed to say it out loud (Score:3)
The man doesn't really have to say much, so every utterance is parsed and analysed to death. Seems like he is admitting it hasn't delivered to expectations. It's standard lines from a CEO, kick the can down the road and stay out of the line of fire if possible.
Re: (Score:2)
That is almost certainly true, as the expectation was "AGI that replaces all human reasoning". What we got so far is incredible data collation tool that makes almost all data collation ridiculously more efficient. And the larger the data set from which you're collating the more efficient it is. This is what's opening completely new venues of reasoning in people, because we just couldn't collate data at certain scales before, and now it's easy.
But AGI has proven elusive so far, and that was what the expectat
Re: (Score:2)
The problem though is it is useful, it is not life altering.
It also amounts to in a lot of cases a generalized solution where a domain specific tool would have suited before, likely performed marginally better and been cheaper to run, but not always cheaper to build.
Take your small hvac vendor. Long about October they start getting a bunch of calls to the effect of why is my house so cold. Those either got handled by the receptionist who was very bored or some expert system.
Have you checked it set to heat
Re: (Score:2)
What you seem to be talking about fundamentally is that current gen generative AI, whatever form it takes is very, very good at collating large data sets and outputting useful results based on collating that dataset for specific details found within it.
It is very good at finding all the needles in the haystack, so long as its guided correctly.
But it's definitely not AGI. Whether "yet" needs to be appended to previous sentence remains and open question. But if it is, then the possibilities are far greater.
Re:Not supposed to say it out loud (Score:5, Insightful)
If AI was reliable then there are lots of applications it could be used for. The fact that it hallucinates and reproduces copyrighted material is the issue holding it back.
If it was like Star Trek where you could just ask it things and trust what it says, it would have far more utility.
Re: (Score:2)
But where was the 10% growth due to Excel and email?
Re: (Score:2)
Fair question.
I would say these were necessary but not sufficient for outsourcing jobs to foreign countries. So the 10% growth is there over the last 30 years in certain sectors, just harder to measure here in the USA.
Re: (Score:2)
This sounds like an admission that generative models in their current form are overhyped - not the transformative panacea they're sold as. I don't think he was supposed to admit that publicly...
There's a lot here that'll probably see him drug before some team-up between marketing and the legal team. This whole thing right here is pretty much one drop after another of anti-message truth:
"Us self-claiming some AGI milestone, that's just nonsensical benchmark hacking," the chief executive said during an appearance on podcaster Dwarkesh Patel's YouTube show this month. Nadella thinks a better benchmark for AI's success should be its ability to boost a country's gross domestic product. "When we say: 'Oh, this is like the industrial revolution,' let's have that industrial revolution type of growth. That means to me, 10 percent, seven percent for the developed world. Inflation adjusted, growing at five percent, that's the real marker."
I thought Microsoft already redefined AGI as some monetary benchmark? And comparing anything to the industrial revolution is probably a bad idea for the public with any awareness of history. The industrial revolution can be seen as a net positive from this far out, but there was a *LOT* of turmoil in that particular
It's already known. (Score:2)
The "killer app" for AI is already obvious because it's literal killer app. AI makes it much easier to kill people, even if it's mistaken on the identity of the target, it will make it easier to kill people and for people to justify killing unrelated individuals. The DoD is just itching for it to get good enough to make squads of unstoppable death machines that will relentlessly kill people they deem an enemy.
The killer app is killing.
Re: (Score:1)
Yeah, like with the Lavender software or the demonic "daddy's home" system Israel uses to wipe out wholrle families. Palantir is also hard at work on the perfection of automating killing, and Alex Karp can't figure out why anyone even has a problem with this technology being used on anyone, let alone civilians. These billionaires are sociopaths
Re: (Score:1)
Collateral damage is what happens when you invade a foreign country and massacre its civilian population. FA is followed by FO.
Re: (Score:2)
You are literally okay with genocide as a "find out" option.
Disgusting.
Re: (Score:2)
Re: (Score:2)
AI already has a killer app (Score:2)
Your job.
Robots have a killer app too. (Score:2)
You.
Re:AI already has killer app: yr job (Score:1)
You mean bots can clean up the messes of moron PHB's?
Just get a bot to do dishes, from sink to rack. Everybody and their dog will want one.
Actually if they standardized a sub-set of dish sizes that could be done via old-fashioned automation. You'd still have a fancy set for guests, but everyday dishes could be standardizes so it's easy for a bot to clean and place them.
I watched the interview (Score:3)
He honestly explained his viewpoint as a CEO
CEOs care about money, and his answer makes perfect sense for a CEO
Researchers have a different opinion
I suspect that since we can't yet precisely define AGI, researchers will argue about it for a long time
First time for everything (Score:2)
Is this the first time that "killer app" could literally mean app that kills? Would be enormously popular in certain contexts. For the rest of us maybe not so much.
intelligence finds its own killer app (Score:2)
If AI were intelligent, much less super intelligent and reasoning, it would find its own killer app. In fact, intelligence itself is a killer app. This is an open admission that AI is just some techniques not well understood.
Re:intelligence finds its own killer app (Score:5, Interesting)
The techniques are well understood, but the hype is certainly not. People are making LLMs out to be things they are not. The most laughable is that AGI is right around the corner. Is that right? So we'll have AGI before we can even train a model to drive a car in a snow storm? Yeah sure. I'm reminded of the 80/20 paradox, one can claim to be 80% of the way there, but the last 20% will take at least 4x as long as the first 80% did because that's the hard part. I recall a heated argument I had with an AI researcher at Google in 2018 that claimed self-driving would be solved within the next 12 months. Well, it's 7 years later and we still don't have true self-driving, just a few geofenced trials in running areas where driving conditions are optimal.
Re: (Score:2)
How is 80/20 a paradox if you are spending 4x on the 20% time? You only got to 20% when you claimed you are at 80%
Re: (Score:2)
You are right that "paradox" is not really the right word, but that is the label commonly slapped on.
It really a kind of Measurement Dysfunction, where an incorrect/misleading or fatally incomplete measurement drives counterproductive business behavior.
The underlying cause is the tendency to make linear extrapolations for all outstanding issues, where the poorly understood issues are presumed to be approximately as hard as well understood issues. Someone with a little self awareness would realize that if I
Re: (Score:2)
It is possible to build an AGI without it being able to learn how to drive a car. Eventually, it may be able to do that, but it is not a defining activity that must be achieved before AGI is operational.
Next killer app (Score:1)
When It Finds Killer App ... (Score:2)
Satya Nadella Argues AI's True Value Will Come When It Finds Killer App Akin To Email or Excel
Considering that AI is supposed to be able to solve anything from building a better mousetrap to 'solving' climate change and faster then light travel within our lifetimes I find myself wondering why these people are out and about with a 800 million candle WWII searchlights looking for an "AI Killer App"?? If all that I have heard the AI leadership say about it then we should not be able to move without stepping on an AI Killer App.
Re: (Score:2)
I asked ChatGPT what the likelyhood is that AI investments would be one of the worst financial disasters of all time. It said 0%. Absolute certainty in what I believe a mathematical impossibility. A friend asked deepseek and grok, they both said 5%. I assume that grok asked deepseek.
Re: (Score:2)
What is the likelihood that you're a permabear?
Re: (Score:2)
Re: (Score:2)
Considering that AI is supposed to be able to solve anything from building a better mousetrap to 'solving' climate change and faster then light travel within our lifetimes
He's not talking about AGI (wouldn't that really make more sense as GA... oh) but just about the AI stuff we have now, which mimics certain aspects of intelligence. Frankly I think we already have found AI-as-we-currently-know-it's killer apps, and they are all in graphics. Trying to have it do facts and information is a bust. But it's great at doing upscaling, seamless procedural textures, frame generation... Our brain is unnervingly sensitive to small auditory cues, and you can catch problems in text pret
finding stuff (Score:2)
Computers did not find email or spreadsheets. In the same way, AI will not find a "killer app", a "killer app" might happen but not because a concept "found" it.
Re: (Score:1)
Re: (Score:2)
I would like to imagine a "killer app" that AI could provide, just to be a geek, and maybe others would join in, just to get a glimpse of the future. I can imagine me coming home and asking an AI agent to play me a Motion Picture placed in the Star Trek Universe consisting of movie stars that lived in the 1950's. Make the plot about the Federation going to war with the Klingons in the "new generation" format. It would then generate the script, and a movie would be nearly instantly generated for me with all kinds of special effects and plot twists. I would sit back and enjoy for the next two hours. That would be kinda cool. Does anybody else have any ideas?
See also: Isaac's story in The Orville.
Excel was never the killer app (Score:5, Insightful)
The killer app was Visicalc, not Excel.
Excel just copied it.
Re:Excel was never the killer app (Score:4, Insightful)
That's what he means, Microsoft needs someone to make a killer app so they can copy it. That's been Microsoft's MO for every product they have.
Re: (Score:2)
None of these companies are your friend or anyone else's friends. Unless you are a major, major shareholder. And no your puny little 401K does not make you a major shareholder. It makes you a useful place to offload bad stocks onto
Re: (Score:3)
Excel is killer because it's better, although it's not so much better as it used to be and the interface is really getting to be terrible. It still is the performance king of spreadsheets, which is counterintuitive, and my other go-to (LibreOffice Calc) doesn't have live pivot tables which is frustrating as hell, it makes adjusting them to be what I want takes literally twice as long. But on the other hand, hitting ^Z in the active window doesn't undo stuff in other Excel windows, which is what I had happen
Re:Excel was never the killer app (Score:4, Informative)
The killer app was Visicalc, not Excel. Excel just copied it.
Lotus 123 copied Visicalc. Excel copied Lotus 123.
Hasn't it been around long enough (Score:1)
Reminds me of something I read a while ago that I wish I could source. It was something along the lines of if a piece of tech hasn't been picked up in a decade since it came out, it is pretty much a dead end.
Spend a trillion dollars a year- (Score:2)
Itâ(TM)s the scattergun approach (Score:5, Insightful)
So until we find a killer app for AI, they will just thoughtlessly shove it into every program and service gat they can, then increase the cost of those programs and services because the user has to pay for something that they never asked for, nor have a compelling reason to use. I refer to the recent increase in the Microsoft 365 plan costs being rolled out to the extent that a year’s subscription to the personal plan is almost the same as buying the standalone Office suite.
Oops (Score:3)
Re: (Score:2)
Perhaps.
I think it's more along the lines of, "We can think of a lot of things AI can help with, we just don't know which one people will latch on to."
Search (Score:2)
The killer app for LLMs is search. Not search competing with Google, but search on personal things you care about and probably have behind a VPN.
That's here already but not in app form and not in a way that will sell new hardware like Visicalc (which I'm sure is what Satya meant!).
Personal recommendation systems tuned to you instead of dinosaur ad execs?
Productivity Paradox (Score:2)
When PCs were introduced to businesses in the 80s and 90s we didn't see productivity improvements, on an economy wide level.
It seems super weird and backwards, hence it's referred to as the "productivity paradox".
Worse, the introduction of the internet to most businesses in the 2000s actually corresponds with a productivity slowdown, where it increased at a lower level than normal.
As the introduction of PCs didn't trigger economic growth and the introduction of the internet retarded economic growth it is am
Re: (Score:3)
I know this seems super weird, I certainly feel sure we are all more productive with PCs and the internet, but the data doesn't support this.
I don't know what data you're talking about, so I don't know if it makes any sense. But I will say that the internet enabled a lot of stuff that wasn't even possible before in a lot of businesses. Take RV repair for example. Having access to all the documentation without having to have collected it yourself over decades (because more and more of it is online, there are groups where RV techs hang out and share documentation - I am still in one of these despite not working on RVs any more because I still have
Microsoft CEO can't spell Lotus or WordPerfect (Score:5, Interesting)
Automated authoritarianism (Score:3)
Snitches are unreliable, spies eat and need sleep and cost a paycheck. But robot cameras do not blink, take bribes or form unions. And false positives are a feature in systems like this - they keep people afraid, and if someone important gets caught you just wink at that.
In the US, the "internet driver's license" that was talked about 20 years ago is coming via "save the children from porn" ID verification. That makes it much easier to automate tying it back to other accounts, employers, or home addresses, depending on the necessary degree of re-education a recalcitrant citizen-unit needs.
Re: Automated authoritarianism (Score:2)
What an idiot (Score:2)
How TF was Excel groundbreaking? It's an incremental iteration on Lotus 1-2-3, which was an incremental iteration on VisiCalc.
Nothing is killer app about "Excel". Spreadsheets? Sure. Excel merely achieved market heft by shady tactics from convicted monopolist Microsoft.
Nothing revolutionaly about "Excel". It's a mediocre copycat product and has always been.
Re: (Score:2)
How TF was Excel groundbreaking? It's an incremental iteration on Lotus 1-2-3, which was an incremental iteration on VisiCalc.
Nothing is killer app about "Excel". Spreadsheets? Sure. Excel merely achieved market heft by shady tactics from convicted monopolist Microsoft. Nothing revolutionaly about "Excel". It's a mediocre copycat product and has always been.
I would argue that Excel was not an incremental iteration on 123; it was pretty much a clone. It wasn't until Excel 2.0 and the advent of Windows 3.x that Excel added the incremental improvement of a Windows-based UI.
You know what the killer app is (Score:3)
Two words: cleaning toilets.
Nobody wants to do it. Everybody needs it done. Build an affordable robot that can do it reliably and then plug itself back into its charger, and the world will beat a path to your door.
Re: You know what the killer app is (Score:3)
Funny, but 100 percent true. The main thing is no one wants their jobs in writing, art, videos, expert opinion etc replaced. Everyone wants their unsafe dirty job replaced. 99 percent of the value of AI will be in designing and controlling physical robots to do those things like cleaning toilets.
Translation is a killer app for LLMs (Score:4, Interesting)
LLMs are good at translation. That's really a killer app for them.
Perhaps in the future the translation will be built in everywhere. So for a speaker of Spanish, watching videos in English the speak automatically comes out in Spanish. And vice versa.
For general AI, I think the killer app would be a personal assistent.
Re: (Score:3)
LLMs are good at translation. That's really a killer app for them.
Even then, they're not that good. They'll let you find directions to the bank or order a pizza but you won't be able to have a meaningful conversation with one, especially once they start using local dialects, slang, et al.
Hell, Google translate still defaults to American sensibilities so it mistranslates a lot of words to make them nicer than they really are, I.E. "Molestar" in Spanish is translated to "to bother" or "to annoy" where it's quite a bit harsher than that.
Killer AI app..... (Score:2)
One way to think of AI is as a general purpose app (Score:2)
Cart before Horse (Score:2)
There needs to be some space in which LLMs are the killer solution to some unsolved/"unsolvable" problem.
Wake me, Ima take a nap. Names Rip van Winkle.
Re: (Score:2)
bullshit (Score:2)
Satya Nadella doesnt know the history of Artificial Intelligence.
If they did, then they would understand were not waiting for a killer app.
The Killer app for AI is games and this is nothing new.
AI Killer App - World Economist (Score:1)