AI - Out of Control Already?

[najaB]

Aren't you essentially just renting a cloud server, though?

Yes. But the key thing is that you have full control of pretty much everything except the power switch. Which is in contrast to fully cloud-based services where you have control over basically nothing.

 

[Trackman]

Just like someone who is 70 now, who's gone from a totally analogue world to a totally digital one.

I would say people around 60 too, including myself.

 

[Bantamzen]

Today, this is true sure.

But I do believe in the long run the advancements will be so great that an incredible amount of work can be automated.

There will likely be professions like healthcare mentioned earlier where the output will need a much more expert inspection/verification, but there's a lot of software engineering roles which are building quite trivial things like simple "CRUD" APIs or websites for local plumbers. I built a website for a client on the weekend using a website generator, it took me about 1 minute and a site generator about 5 minutes to do the work, and the result is 4 social media accounts and a website with home, about, a contact form, a blog which it even generated a couple of actually high-quality posts on with bespoke licence-free images and a bespoke video. That's the sort of thing I started building as a teenager for travel money, and I charged a few hundred ish; if someone had said to me back then that it could all be auto-generated I'd probably have laughed and maybe conceded that the static content could be generated from a template. Today I have a different mindset which is actually that over time, less and less programming will be manually-done and in the long run I don't think much will be off limits for automated development; how long is a trickier prediction of course!

I don't think it'll be particularly long before software development and data engineering will be more or less done via AI, and devs will be reskilling into machine learning and prompting. And I don't think it will end there, AI will be perfectly capable of data analysis, forecasting, resourcing, accounting, even acting in a consultancy role. There will of course have to be some human intervention at the decision making level and some level of quality assurance for automation, but from the work I've done so far with specific datasets AI has easily demonstrated how quickly it can become viable. Actually it is slightly disturbing just how good it is when you give it context for a dataset. So I'd expect by the turn of the next decade that AI will be playing a much more substantial role right across business, and will be making inroads into areas where some people might not be expecting.

I'd say for people in middle management roles, now might be a good time to start learning how to work with AI if you want to stay in your fields. Personally whilst I am on the start of the AI curve, I plan on at least semi-retirement in the next few years so I'm not as worried as I would be if I were 10-20 younger!

 

[najaB]

I don't think it'll be particularly long before software development and data engineering will be more or less done via AI, and devs will be reskilling into machine learning and prompting.

As I commented above, the potential problem there is that current AI isn't intelligent - it has been fed a lot of training data and uses statistical analysis to predict the most likely next word/line of code. This has a few issues: it's only as good as its training data so rarely generates anything novel, as it starts to get trained on other AI-generated code you'll get the similar issues as you get with inbreeding in animals, and the non-deterministic nature means that you can run the same prompt three times and get four different answers - which is a nightmare for ensuring code quality where updates and bug-fixes are concerned.

 

[Tester]

As I commented above, the potential problem there is that current AI isn't intelligent - it has been fed a lot of training data and uses statistical analysis to predict the most likely next word/line of code. This has a few issues: it's only as good as its training data so rarely generates anything novel, as it starts to get trained on other AI-generated code you'll get the similar issues as you get with inbreeding in animals, and the non-deterministic nature means that you can run the same prompt three times and get four different answers - which is a nightmare for ensuring code quality where updates and bug-fixes are concerned.

That is a very good analogy.

 

[brad465]

With all the agent hacking news, I'm thinking this could well be the next financial crisis and on a scale that dwarfs the 2008 crash. All that would need to happen is for an agent to hack a bank and rob life savings and/or other holdings, then suddenly a combination of savers doing a Northern Rock-style run on the banks, but an order of magnitude higher, combined with a huge plunge in confidence in the banking sector being secure, and then it's all over.