The AI Battle

> Any bets on when the majority of software development is automated? My estimate based on the exponential improvement rate of AI on various tests like MMLU and MATH is within 3 years.

An AI solution will always target bang for buck first. So _potentially_ web development or CRUD apps have the greatest probability to be automated.

Nevertheless there is a whole zoo under 'software development'. Bio informatics application, Fast trading software, medical ventilators, hardware debugging with software interfacing with real-world sensors, etc., etc.

Even if your scenario comes true there will be so much 'domain-specific' knowledge that you would still need some sort of 'Business/requirement analyst + Architect' to be able to steer AI agents into any meaningful product.

Maybe the future of Software is not in the writing but in the Designing for the specific domain. Which come to think of it is not that bad of a job and probably not that far off from what a Principal/Staff Engineer is doing nowadays.

AI that can do creative work is scary for its implications on humanity. Are we at best biological AI's ?

I've played around with several of them. These artwork AI's are way more scary than GPT-3 et al because they seem to do something so creative with such good visible results.

I don't think the genie goes back in the bottle. I think humanity has an existential crisis on its hands.

Whats the point in doing things if the AI can do it better than you ? Do those things become relegated to some increasing unpopular niche ?

I just have increasingly amounts of questions and fear.

This is the classic (and flawed) "it's gotta be perfect otherwise it's useless" argument that you see in other AI fields, such as self-driving.

You don't have to be anywhere near perfect to cause tremendous disruption. "Polyester" grade artwork will be perfectly acceptable for a lot of use cases.

Just like how self-driving will decimate the trucking workforce, all while people continue to screech about how "it can't handle snow therefore it's useless".

I think the article is overly sensationalist. But I think the gist is totally realistic. For example, there's a huge chunk of design work that's not super critical and often badly done. I can imagine a simple, AI-assited tool that creates posters, simple websites or flyers in the companies style. But it won't replace graphics designers yet and won't be for some time. I think we are not yet completely ready to create even a simple tool, there are still engineering-challenges to overcome (I am not sure how the technical interface to the ML-would look like, I can't imagine text is enough), but I think we might be very close to have at least something resembling a product.

I know a thing or two about machine learning...and one thing I've learned is that it's very hard to predict the future, even the immediate future. Some problems turn out to be surprisingly easy, but other we just can't solve yet, even super basic things like learning multiple tasks in a sequential but unknown manner without forgetting stuff. I find it hard to predict what ML will look like in three years, except that dalle-3 will be much better. If we have a breakthrough on one of those fundamental problems where we struggle with the basics, then we will open new possibilities quite quickly I think. But it als might be that they require new approaches that will take years or decades to develop.

The "AI battle" only rages for those who don't make AI; those that write AI know that battle is lost, we will not have an AGI in our lifetimes. We lack fundamental components, such as artificial comprehension. Lacking any form of comprehension, weak or not, all AI technology can ever produce is idiot savants and personality theater. It is not capable of writing original software, beyond the use of pre-composable components, because lacking comprehension it is not writing anything. At the most it can predict given a lot of examples of similar, but it still does not comprehend its own predictions.

AI is a tool for creative people to use, it cannot replace them, unless you want a really boring product, because by design AI rehashes and recombines, it does not have a create capability in its capacity. To create requires comprehension, and lacking comprehension it is just mimicry.