Why I think AI take-off is relatively slow
marginalrevolution.com
marginalrevolution.com
I'm trying to pivot out of data which IMO is a scam industry and thought I'd consider automating white-collar work. After all, there's a huge amount of excel-monkey work that can be trivially automated with scripts and I've done stuff like that before. But then I realise there's not even a job title for this sort of work, nor are there any firms in my country doing stuff like this. There's simply no demand whatsoever for process automation (I'm expressly not talking about automation engineers in manufacturing etc.)
It's not hard to see why. No-one's going to automate themselves out of a job, nor are managers going to automate all the people they manage out of a job because then they're also redundant. Often labour-saving innovations are brought-in by upstarts but business dynamism is low so there's not a lot of that happening. I can almost guarantee that a bank circa 2050 will look a lot like a bank now, short of some runaway superintelligence completely reconfiguring society.
As for banks in particular, seeing the success of the likes of Revolut, I think we'll see more disruption in this space too.
Now with chatgpt they might be able to do it themselves.
Also when big corps sell you an ai hr Person and the only thing needed is a Microsoft account you might buy this.
They continue to either make it generally good or add continuous new specific skills to it. I think the effort to teach one 'ai' something is very soon a lot cheaper than teaching someone new constantly.
Would you be open to writing more about this? I work in data, and I find it valuable to hear criticisms of the fundamentals of the field from people with enough context to actually articulate meaningful criticism.
I saw this video a few months ago, and I thought it was pretty sloppy, but it made a few pretty good points: https://www.youtube.com/watch?v=pOuBCk8XMC8
I live in Melbourne like the author and I certainly don't speak for the entire industry. But here's it's mostly government and banks and other enterprise where data is just a cost-centre so middle-management could say they were big on data science, which has since transformed into being big on AI.
There's no end-use for most of the work, there's no stakeholders, often the product being delivered will be a 'platform' of some sort and everyone outside the data area is afraid of sounding stupid by asking what that means (it doesn't mean anything). The result is entire services can go down for months at a time and no-one notices (and then when someone does realise management covers up the fact no-one noticed because it's embarrassing).
I actually think these are expressions of organizational dysfunction rather than something inherently wrong with data. It's the same way dysfunctional orgs might spend millions on SAFe adoption or prematurely spinning up a global sales motion - incentives and perceptions become misaligned with the reality of what the organization needs to do to succeed, the decisionmakers make bad capital outlays, and the implementers are left feeling like garbage working on a project that's doomed to fail.
IIUC, the FDA is slow on purpose. Because proving drugs aren't worse than the problems they treat takes several phases of large trials. Even if AI could cure all cancers tomorrow, I hope we still maintain a proven scientific method to QA that before releasing it widely.
The machine learning does not reason thereby making today's "AI" do cognizant dissonances (or hallucinations).
In short, GIGO: garbage in, garbage out.