IBM CEO Doesn't Think AI Will Replace Programmers Anytime Soon (techcrunch.com) 50
IBM CEO Arvind Krishna has publicly disagreed with Anthropic CEO Dario Amodei's prediction that AI will write 90% of code within 3-6 months, estimating instead that only "20-30% of code could get written by AI."
"Are there some really simple use cases? Yes, but there's an equally complicated number of ones where it's going to be zero," Krishna said during an onstage interview at SXSW. He argued AI will boost programmer productivity rather than eliminate jobs. "If you can do 30% more code with the same number of people, are you going to get more code written or less?" he asked. "History has shown that the most productive company gains market share, and then you can produce more products."
"Are there some really simple use cases? Yes, but there's an equally complicated number of ones where it's going to be zero," Krishna said during an onstage interview at SXSW. He argued AI will boost programmer productivity rather than eliminate jobs. "If you can do 30% more code with the same number of people, are you going to get more code written or less?" he asked. "History has shown that the most productive company gains market share, and then you can produce more products."
right on (Score:1)
"there is a world market for maybe five computers" [Thomas Watson Sr.] ...
Webbrowser are for consumers, Lotus Notes for Enterprise (in meaning] [Irving Wladawsky-Berger, early IBM Internet Division VP on web browsers] IBM WebExplorer was canceled
Re: (Score:1)
all decent code
Re: (Score:3)
Re: (Score:3)
That's what I've found AI code to be good for. Established API for a project? Give me a function that does XYZ. Or something simple like "Give me a PHP function that connects to an FTP site, uploads a file (with the name given as a parameter to the function), retries 5 times in case of failure, then returns a boolean if transfer was sucessful".
I kind of view it like using a calculator for math problems. I COULD do it by hand, but its just faster to use a tool. At the end of the day though that tool doe
Re: (Score:3)
Re: right on (Score:2)
Re:"Market for only 5 computers" (Score:3)
Tom perhaps was technically right if those five computers were gigantic with gajillion cores, which IBM would be happy to make and sell.
Do note that although it's hard to confirm Bill Gates actually said "640k is enough", there are other quotes that suggest he was indeed surprised by how quick devs bloated up their software to use up all 640.
Gates was forced to practice tight DRY, YAGNI, and KISS to make early MS software, but a parsimonious approach was soon abandoned by the industry, because they could. B
Re: (Score:2)
Gates never this this. The origin of this quote was John Roach (Tandy CEO) and referred to the memory on the TRS-80 Model I. The press at the time thought the machine didn't have enough memory (or was too expensive for what memory you got), and Roach retorted that ({some amount} of memory should be enough...).
Re: (Score:1)
Okay, but my point still stands that he was surprised at how fast it was used up. I can't find the "surprised" quotes just yet, but have seen them before.
Re: (Score:2)
The origin of this quote was John Roach (Tandy CEO) and referred to the memory on the TRS-80 Model I.
Do you have a reference for that? The TRS-80 Model I has only 4 to 48 kBytes of RAM.
Re: right on (Score:2)
They want a piece of the trillion dollars (Score:2)
That's all. They want to get piece of the pie. It's all about the money, and not about reality.
Re: (Score:1)
Thomas Watson, chairman of IBM was Jr. not Sr.
What he actually said was
"I would like to tell you that the machine rents for between $12,000 and $18,000 a month, so it was not the type of thing that could be sold from place to place. But, as a result of our trip, on which we expected to get orders for five machines, we came home with orders for 18.â
Said about their IBM 701 Electronic Data Processing Machine, a scientific computer.
Re: (Score:1)
Yes Jr. was the son and successor of founder Sr. at IBM, by the way
Re: (Score:1)
I think Watson Sr, might be vindicated yet.
There are what 5 really big cloud providers. Those platforms might very much be thought of as one vary large computer at least in the sense we think about the mainframe that way. Many CPUs, ultimately one control plane.
Isnt the trend to render every other device with a CPU into being something we more think of as an appliance, than a general use computer?
I am not saying we are there yet but if you zoom out far enough Watson might be right...
Re: (Score:2)
And at the time he said that, he was right. Seriously, what utter morons feel the need to trot that one out? Now, Bill "the idiot" Gates with his main memory prediction would be a different thing, because he was dead wrong even back when he said it.
Re: right on (Score:2)
Except he never said it.
Re: (Score:2)
Yes, probably. Does not matter though, he said plenty of stupid things that can be verified.
Re: (Score:2)
He was wrong, we may ultimately get only one computer in the world not five. A central computer doing all the processing while everything else, robots included, is just the client.
The real pain will start (Score:5, Insightful)
The real pain will start when corporations discovers that with AI you'll hamper the growth of new fresh programmers getting training on how to code by starting with the simple cases that the AI can do.
Those cases are often easy but gives new programmers a lot of training in the corporate culture and coding style as well as a stepping stone in their growth. No programmer is ripe for the high profile advanced cases right out of university.
Re:The real pain will start (Score:5, Interesting)
Exactly. Coding needs to be learned on simple things. If AI does those, you will get a lot fewer competent coders.
The second problem is that apparently, AI code is hard to modify later on. People using AI code will likely create massive technological debt this way. Hence it looks like AI code may be much more expensive than code written by humans.
Re: (Score:2)
Though at my company, the big hope is that formalized 'design practices' will get around the training problem. No need to gain experience when there are hundreds of documents outlining the wisdom of random people, and a requirement to explain which ones you used and why you didn't use others for a p
Makes sense (Score:2)
There is no need for more than five mainframes anyway.
Re: (Score:2)
That happens to be wrong. But there is really no need for more than five idiots of your type.
Re: (Score:2)
That's a great joke, thank you!
IBM Granite (Score:2)
If we had to use IBM's Granite LLMs, I would be just as pessimistic. Fortunately there are other, more capable companies working on this.
Could very well be right. (Score:2)
But I expect he'll be burned at the stake for saying it. Daring to say anything other than, "The entire workforce will be replaced by AI in $x years" where $x is some number
Re: (Score:2)
But I expect he'll be burned at the stake for saying it. Daring to say anything other than, "The entire workforce will be replaced by AI in $x years" where $x is some number
Thanks, Slashdot, for only taking the first little bit of my post when I clicked submit.
Rest of the comment: < 20. Based on what I've seen in AI output of code, he's more right than the people saying AI will replace all coders in the next few years. But daring to say that out loud will likely have a lot of management types throwing a complete conniption, because the promise has been made that the workforce *will* be replaced by AI ne the next decade or so, and anyone daring to say otherwise is considered
For once, complementing AI (Score:4, Interesting)
The fix it provided actually resolved an old workaround I had in place, it's been fully tested, and it works. Well, I don't think AI will replace programmers outright, for once, AI actually did something remarkable for me, not trivial, or as a servant.
IBM's ML/AI is closest to replace programmers, soo (Score:2)
IF any company is close to have ML/AI to replace programmers is IBM, as, unlike the current wall street darlings, they have been at it longer, and with great emphasis in domain specificity (i.e. ML/AI specifically trained for law, ML/IE specifically trained for health, ML/AI specifically trained for programming, etc).
If the guys who are closest say that it will take a long time, barring a susprise discovery somewhere, I think will believe them...
JM2C, YMMV
Re: (Score:1)
This is so wrong. Not even an IBM employee would be this delusional.
The hard parts still murky (Score:4, Interesting)
This one is hard to call because AI is very good at analyzing and echoing code patterns for both debugging and code generation. However, the code ultimately still has to be vetted by humans for non-trivial apps, and there will be plenty of edge cases where the human has to do most of the work anyhow.
And some devs will struggle using AI effectively while others will zoom high and wide. The ratio of the first to the second is still an unknown.
Remember how much promise self-driving cars showed about 15 years ago? It was like they were "almost there", but the "almost" remains in place. The edge cases are mean SOB's, taking longer to solve than anticipated. We could face similar hurdles with AI-assisted-coding. AI's bugs/flubs are quite "creative" and may turn out to be common time drains to recovery from.
"Prediction is hard, especially about the future" - Yogi Berra
I don't want any part of the AI market prediction biz, YOU can have it.
Re: (Score:3)
There are two additional problems:
1) How are the potentially good coders that supervise AI and vet its code supposed to get experience, when all the simple coding jobs are done by AI? Right...
2) AI code can well turn out to be _much_ more expensive becasue of technological debt, hard to spot and hard to fix vulnerabilities, poisoned AI models, etc.
The whole thing is a crapshot. And at 25% efficiency increase upftont, not even remotely worth the risks.
Re: (Score:1)
Per #2, a lot of actual software is sloppy crap. If AI can make sloppy crap faster than humans can make sloppy crap, then it at least overtakes the very large niche of crapware.
Re: (Score:2)
Sure. But as soon as you use it (because the business-morons will insist) on software that is nto crap, that will eat all the profits and then some. And it will do so with a real delay, something the bussiness-idiots cannot deal with since their mental capability to see the future extends only to the next quarter.
AI developers (Score:3)
Re: (Score:2)
It's not about replacing programmers (Score:2)
In the old days when there was competition in the software markets firing all those programmers would be a disaster because they would go off and start their own companies and compete.
We stopped and forcing antitrust law ages ago when Bush Jr let Microsoft walk.
So nowadays if you try to start a software company if it
Re:It's not about replacing programmers (Score:4, Insightful)
It's about increasing programmer productivity so you can fire some of your staff and so that there's an overall decline in demand for programmers and therefore wages for programmers.
In principle, but that would buck a very very long trend.
We have had astounding increases in programmer productivity over the decades, starting from hand rolled machine code to assembly to basic languages, to more advanced languages through to where we are now with all the tools we have now. All these have both lowered the bar for who can program and increased the productivity of those at the same skill level.
So fat this has resulted in the number of programmers going up not down.
Re: It's not about replacing programmers (Score:2)
rsilvergun likes to pretend that he has experience with software development, hence he speaks about it authoritatively even though he has it horribly wrong. Sound familiar?
Re: (Score:2)
He strikes me as someone with experience. I happen to think he's mistaken about this particular point. I don't see why that should imply I want to join some sort of pile on. Go look for stooges elsewhere.
Re: It's not about replacing programmers (Score:2)
He strikes me as someone with experience
He literally claimed that the job of a programmer is to just go through tickets that all have instructions to write a function with specific inputs and outputs. While that may be your experience, it definitely isn't mine. In fact, in my experience, we, the software engineers, talk with the customers and users directly and define the problem that we then decide how to best solve, which may or may not even involve writing code at all, or it may involve writing entirely new projects from scratch. And he believ
Re: (Score:2)
I don't know what precisely insinuating.
But I know why your are only insinuating it: if you say something concrete, you know I'll ask you to provide some sort of evidence which we both know will be missing. So you know put up or shut up.
Anyway there's lots of different software engineering and programming and they are IMO distinct. Rsilvergun isn't living rent free in my head so no I don't have an encyclopedic knowledge of his opinions on the matter. I'm not especially impressed with AI, though being lazy I
He still thinks too positive of it (Score:2)
But at least some degree of realism seems to slowly think in...
LLMs are great for programming by stack overflow (Score:4, Interesting)
Not really good for anything else. At some point there's going be a split between those that know how to do things, and those that religiously follow anything the infallible AI tells them to do. So go ahead and drive across the invisible bridge that runs across the Grand Canyon. The AI is the Source of Truth. Don't trust humans. All those cars you see at the bottom were planted by people that don't want you to know the truth, that the invisible bridge across the Grand Canyon is real and perfectly safe.
reuse (Score:2)
Is there really much difference between AI writing code and code reuse, other than how quickly it can be done?
Current AI can... (Score:2)
...create simple clones of existing ideas about as good as an intern who uses a lot of cut and paste
Creating code based on text prompts only works if the prompt references existing ideas
The prompt "Make a snake game in python" only works because snake games exist and the code was published
Completely describing a complex, novel system in text is troublesome. Natural language is not precise enough. See the law for a good example
To be really useful, future AI needs to help us manage the complexity of very larg
AI is StachExchange on steroids (Score:3)
AI is great for programming homework projects. For code that integrates into existing systems, you need to know what to ask for. AI is only as good as what its asked, just like forums like StachExchange. There will still need to be someone who knows what to ask and how to tie the answer into a real system.
What is it really good for? (Score:2)
Maybe I'm just old fashioned, but I have trouble seeing how AI code generation would be useful to me. Software engineering isn't about writing code as fast as possible. It's about maintaining a code base over many years. I spend very little of my time writing new code. Here are things I spend most of my time on.
- Testing code
- Debugging code
- Optimizing code
- Documenting code
- Reading code to understand how it works
- Refactoring code to make it easier to maintain and extend
- Researching and designing ne