 Notice. New forum software under development. It's going to miss a few functions and look a bit ugly for a while, but I'm working on it full time now as the old forum was too unstable. Couple days, all good. If you notice any issues, please contact me. |
Forum Index : Microcontroller and PC projects : most code will not be written by human beings
Page 1 of 2   |
|||||
| Author | Message | ||||
| lizby Guru  Joined: 17/05/2016 Location: United StatesPosts: 3252 |
Ezra Klein podcast (with transcript) in New York Times with Biden administration’s A.I. adviser Ben Buchanan re AGI (link should not be paywalled). "I’ve talked to a number of people at firms that do high amounts of coding, and they tell me that by the end of this year or next year they expect most code will not be written by human beings." (This is using AGI implementations not currently available to the general public, e.g., to us.) ~ Edited 2025-03-05 01:34 by lizby PicoMite, Armmite F4, SensorKits, MMBasic Hardware, Games, etc. on fruitoftheshed |
||||
Mixtel90 Guru Joined: 05/10/2019 Location: United KingdomPosts: 7339 |
"by the end of this year or next year they expect most code will be written badly" Fixed that for you. :) Mick Zilog Inside! nascom.info for Nascom & Gemini Preliminary MMBasic docs & my PCB designs |
||||
thwill Guru Joined: 16/09/2019 Location: United KingdomPosts: 4200 |
I can assure you that the software industry is already ahead of the curve on that deadline. Best wishes, Tom Edited 2025-03-05 03:12 by thwill MMBasic for Linux, Game*Mite, CMM2 Welcome Tape, Creaky old text adventures |
||||
| thwill Guru Joined: 16/09/2019 Location: United KingdomPosts: 4200 |
The nature of programming will change to that of "prompt engineering", code-reviewing and fixing the sh*te generated by AI. My concern is that the path by which a junior programmer becomes a senior programmer (and thus capable of doing good prompt-engineering and fixing the resulting sh*te) may be eliminated by AI doing the work of the junior programmers. Tom Edited 2025-03-05 03:18 by thwill MMBasic for Linux, Game*Mite, CMM2 Welcome Tape, Creaky old text adventures |
||||
PeteCotton Guru Joined: 13/08/2020 Location: CanadaPosts: 490 |
I think we need to define what "most code" is. Because an argument could be made that "most code" has not been written by human beings for a long time. Back in the 90's we had to write a tonne of code for a window or button press. Then Visual languages came out (VB, Delphi etc.) that made such things trivial. Yes, I get that those background codes were written by a human at some stage - but so was most of the AI stuff. Same with Database accessing - defining classes to match tables and stored procedures - things which are pretty much automated now. Web site design - no need to learn HTML anymore - just drag and drop. This is all just a progression of that. I think there are a huge amount of people out there who would love to be able to write a phone app, but don't possess the technical skills. Giving everyday people the ability to design and write their own phone app without knowing anything about databases is going to revolutionise the I.T. world. That is going to be 90% of the programs being developed - in which case "most code" is a very accurate description. Professional coders, writing mission critical stuff will still need to know how to program, even if AI helps with simpler routines. But much like using AI to draw art, you're going to have to be able to define what it is that you want the program to do and with what triggers/effects. That's easy for a system that waters the plants every day - but far more complex when you want to control a gas turbine. Most programming at that level doesn't involve much time spent typing in code. It's all condition checking and verifying that things fail safe. All of that work still has to be done - and I don't see AI significantly improving that workload. Where I do see it being a huge boon is as an extra sanity check. So, say I write some logic to control a gas turbine. Then I ask AI to summarise the code and explain every situation where the engine might go out of control. That could catch some critical flaws before they go into production. e.g. "The fuel shutoff valve will open if the ESD button is pressed and the gas release tag is high." The AI doesn't need to know what that means - but the human programmer does - and can decide if that's going to be a potential problem. Another area where I can see it being very helpful is patching or upgrading old undocumented code. Having AI summarise what a program is doing could potentially be very useful. Maybe even have it upgrade the code to a newer language/dev environment. Of course, any bugs from the old system will still be in there - but it would be a useful co-programmer to help make the task easier. So, I come back to my original statement. "Most code" is already automated. AI will be very important in future development for a number of reasons. I don't see either of those things as something to be worried about. I'm only worried about things like pacemaker firmware, airplane systems etc. Not about little Suzie writing a phone app to rate the hotness of the boys in her school. |
||||
| PeteCotton Guru Joined: 13/08/2020 Location: CanadaPosts: 490 |
Agreed. But that's the march of progress. I'm sure new roles will open up that we haven't even dreamed of. Sure, we no longer have stable boys and staging houses to rest the horses - but we have mechanics and gas stations. |
||||
| lizby Guru Joined: 17/05/2016 Location: United StatesPosts: 3252 |
When the speaker says, "people at firms that do high amounts of coding", he's probably not meaning that Mr./Ms John/Joanna Q. Public is doing the coding. PicoMite, Armmite F4, SensorKits, MMBasic Hardware, Games, etc. on fruitoftheshed |
||||
| JohnS Guru Joined: 18/11/2011 Location: United KingdomPosts: 3964 |
Yeah, right. John |
||||
| Mixtel90 Guru Joined: 05/10/2019 Location: United KingdomPosts: 7339 |
I'm not sure that writing code is something that AI would be good at, being a creative process. It can *produce* code, but that's not *creating* it. IMHO it's far better used for analysing and testing code. I'm not sure I would trust an OS written by AI even though it shouldn't be a life-critical piece of software. Mick Zilog Inside! nascom.info for Nascom & Gemini Preliminary MMBasic docs & my PCB designs |
||||
| Volhout Guru Joined: 05/03/2018 Location: NetherlandsPosts: 4721 |
Mick, I envision a world where AI is writing code to improve itself. Then you only have to feed the engine with a requirement for low energy, and it (in the end) will eliminate itself. Problem solved. Gigawatts in energy available for humans that need it. Volhout Edited 2025-03-05 19:58 by Volhout PicomiteVGA PETSCII ROBOTS |
||||
| lizby Guru Joined: 17/05/2016 Location: United StatesPosts: 3252 |
Since the speaker was the Biden administration's point person for AI, who was in close contact with the industry, I would suspect that he has a good idea of what the industry believes it is on a path to. And perhaps this AI coder becomes, in 2-3 years as capable as Geoff or Peter or Tom. It can work 24/7 for as long as a project takes. (As an aside, I would suspect that at commercial rates, the work that Peter has done on MMBasic would cost half a million US$ or more--and maybe over a million.) Considering, as I have read, that "cheap" AI is not much more than 6 months behind bleeding edge AI, imagine that you could say to your coder AI: "For the Raspberry Pi, to the extent possible, produce code in C to replicate the functionality in the MMBasic Picomite User Manual and the CMM2 User Manual and the MMBasic for Windows Manual and test using a Raspberry Pi Zero W 2 and Raspberry Pi 5 against all code written by Geoff Graham as published in Silicone Chip magazine, and code in CMM2.fun. Assume an HDMI screen is available and convert all programs using an LCD or VGA to work with HDMI at a resolution and color-depth as close as possible to the original MMBasic program. Design breakout PCBs for the Raspberry Pi pinout to interface with any needed external hardware." "Do not copy any source software which is protected by copywrite--specifically, any versions of MMBasic." "Use Wokwi ( WokWi ) to simulate any hardware."  "Create on WokWi the simulation model for the Raspberry Pi Zero W 2 and as necessary, create on WokWi any required simulations for sensors, peripherals, instruments, and other hardware." This would be a good start and a real challenge to provide value to MMBasic hobbyist users. (Note that as I understand it, it is permissible "black box" design work to create new software to perform the functionality called for in the documentation of a project as long as the original source code is not used or consulted in the design of the new software.) PicoMite, Armmite F4, SensorKits, MMBasic Hardware, Games, etc. on fruitoftheshed |
||||
| Volhout Guru Joined: 05/03/2018 Location: NetherlandsPosts: 4721 |
Good luck ... Volhout . Edited 2025-03-06 01:45 by Volhout PicomiteVGA PETSCII ROBOTS |
||||
| Mixtel90 Guru Joined: 05/10/2019 Location: United KingdomPosts: 7339 |
And if you require a program for something that hasn't been attempted before, never mind succeeded, then what will you get? What it none of the existing languages is sufficient in its own right and fast, compact machine code routines are required? Oh, AI may very well give you a result, possibly even one that works (although there are degrees of "working"), but will it be maintainable for the next 5-10 years? Remember, it can't write original code for something that it has no knowledge of. It's only a self-modifying database. As programmers become more dependent on AI I suspect computing will stagnate. We are partway there now, without AI. The programs are too big and/or slow? No problem, throw more hardware at it. Don't learn to program more efficiently, that costs money. The AI is doing a fine job... Mick Zilog Inside! nascom.info for Nascom & Gemini Preliminary MMBasic docs & my PCB designs |
||||
| PhenixRising Guru Joined: 07/11/2023 Location: United KingdomPosts: 1091 |
I don't know. Someone has to specify the parameters and with a language like BASIC, that's pretty-much the programming. When I'm writing a machine sequence, I'm drawing on decades of real-world experience and I am aware of the important considerations and potential pitfalls. |
||||
| lizby Guru Joined: 17/05/2016 Location: United StatesPosts: 3252 |
I suspect there's a question of whether it would be maintainable by a human in year one. Or even understandable from the beginning. "Only" does a lot of work in that statement. The development of any large system is likely to include a specification of the API (or the "black box" interface). In the example above, the MMBasic manuals provide that specification. The proof would be in the pudding, which means that the testing would have to be rigorous. Your suspicion is quite contrary to the beliefs of the knowledgeable enthusiasts. We'll see how it plays out, and if they're talking about a 1 or 2 year time-frame, most of us should live to see it--if it's visible in the real economy. So far, that has worked pretty well. I don't think anybody is saying that the problem with self-driving cars is that the programs are too big. When I started programming, computer run time was vastly more expensive than programmer time. It's been a long time since that was the case, and barring societal collapse, it never will be again for a human coder. PicoMite, Armmite F4, SensorKits, MMBasic Hardware, Games, etc. on fruitoftheshed |
||||
| zeitfest Guru Joined: 31/07/2019 Location: AustraliaPosts: 534 |
Some time ago I opened a web page that required 170 meg just to open...all to present a few lines of description. So I suspect that data bloat might save us yet  |
||||
| damos Regular Member  Joined: 15/04/2016 Location: AustraliaPosts: 67 |
Having played with Github copilot and even Open AI, the code it generates is quite good. Even try something like "write this in 8086 assembler" and it does a good job. The AI assistance in VS2022 is 80% helpful and 20% extremely annoying. The problem is that it can't cope with the main skill required by a software engineer: take a poorly described problem raised by a customer who really doesn't know what they want and solve it. A lot of interpersonal skills are required and a lot of experience teasing out the requirements and producing a prototype then refining the solution until it is something close to what they want. The whole Agile thing is framed around this concept that customers have no idea what they want and what they tell you changes on a daily basis. If you use traditional project management techniques, when the solution finally arrives it is never what they want. With a lot of experience you can anticipate things and hopefully save some pain, but more often than not you end up overengineering some parts where you anticipated wrong. The other big issue is problem solving. When an architect designs a system and breaks it down into pieces, sure an AI can implement a lot of pieces. Things like WebAPI endpoints are largely the same implementation around each model object. But copy/paste and search and replace will do the same thing. Or you can template it and make an elegant solution. If a solution is not just the same as every other project and really needs proper problem solving and design, it is very unlikely that AI will be able to do it. |
||||
| lizby Guru Joined: 17/05/2016 Location: United StatesPosts: 3252 |
Agreed (and my own career job description might best have been "programmer/analyst"), but at this point, they are talking about an AI programmer, not an AI analyst/designer. But in the case of artificial general intelligence, I'm not sure why an AI analyst couldn't more patiently and thoroughly work through a Rapid Development process with the customer--especially with a programmer AI suite working 24 hours a day on the "Rapid" part. ~ Edited 2025-03-07 03:45 by lizby PicoMite, Armmite F4, SensorKits, MMBasic Hardware, Games, etc. on fruitoftheshed |
||||
| JohnS Guru Joined: 18/11/2011 Location: United KingdomPosts: 3964 |
I wish they'd put off having AI writing code for some time in the future AFTER they've got AI removing hate speech / death threats / racism / misogyny / etc from anti-social media!! And likewise get it removed from messages sent to MPs (esp. female ones). Plus have it identify who the poster/sender is. Once they lose their anonymity and have the police issuing cautions etc they may mend their ways. They could fact-check ridiculous & wrong claims made by some high-profile people, too. John |
||||
| Mixtel90 Guru Joined: 05/10/2019 Location: United KingdomPosts: 7339 |
Ah.... It's "Artificial Intuition", not "Artificial Intelligence". Someone got hold of the wrong end of the stick. Using the correct meaning makes it obvious that the system will take a guess at what you want rather than using some sort of "intelligence" to produce factual information. :) Mick Zilog Inside! nascom.info for Nascom & Gemini Preliminary MMBasic docs & my PCB designs |
||||
| Page 1 of 2 |
|||||
 Print this page |
