Comment by panarky

(All this assumes some some not-yet-here future where "AI agents" are less flaky than today's LLMs.)

Things like "call center jobs" are where "superior" gets muddy.

They can be superior for the business. The business does not want to spend money. Now they aren't paying a person, and they have to worry less about a sob story convincing their agent to make an exception. Health insurance company, for instance, where the life-saving treatment was declined. Refund of plane ticket because the flight was delayed and normally the policy would be to deny it but this particular person missed his father's funeral as a result, so the agent takes pity. So it's "superior" for the company because it entirely IS about saving money.

Hard to say those are superior for the customer. And most of us aren't the megacorp-owners here. We're the customers.

So yes, AI agents could be the logical next step in the "turn people into robots" march of bureaucracy. But that's not a good thing.

Human interactions, human judgement, human empathy - these are features, not bugs. Consider also that loneliness epidemic. Let's make it even worse! (In the short term "not talking to people" is being seen as a positive here - because we've already raised a few generations of scared, not-socially-equipped kids, since these are old trends. How is people-avoidance-maximization working?)

> AI agents done well are superior at call center jobs, screen-based office work, mortgage processing, financial analysis, most business consulting like process redesign, etc. The biggest benefit isn't reducing payroll, it's doing the job faster, with higher quality and more consistency.

Please don't use the present tense to describe a not yet realized future.

> less tedium

That may eventually happen, but most of the time current AI systems need a lot of handholding to reach human levels of accuracy. I personally find this kind of supervision extremely tedious, it’s more stressful to use a poor level 2 system than just drive yourself. Driving has surpassed that point, but it’s taken billions so extrapolating into other fields without that kind of investment is premature.

>AI agents done well are superior at call center jobs, screen-based office work, mortgage processing, financial analysis, most business consulting like process redesign, etc. The biggest benefit isn't reducing payroll, it's doing the job faster, with higher quality and more consistency.

Just wait until your human needs inside the bowels of some corporate or government bureaucracy, that no matter what will inevitably make either human or algorithmically generated mistakes, are being "attended" by some AI agent that can feel nothing, cares nothing and of course doesn't really think for itself or use common sense outside the bounds of formal rules, and you find yourself fucked over by this in some absurd way.

Imagine all the so-called customer service (almost entirely non-human) that Google shafts its users with, about which so many people on HN have complained, but writ much larger, in all kinds of far more vital user attention scenarios.

No thank you. Human bureaucrats are bad enough, but at least there's an avenue for empathy and flexibility in many cases.

The AI fawning on some comments here lives in a bubble of perfect expectations that will die a horrible death in the real world, or cause people horrible miseries in that same real world.

How is it better for the worker? They just go hungry instead

The problem with that kind of thinking is that "superior" is in the eye of the beholder.

An AI manager might be "superior" in the view of the executives of the company, but that AI manager's reports might feel very differently. From a societal perspective, the employees' feelings are what should matter most, but from a capitalist perspective, the executives won't care if workers are treated poorly, as long as the work gets done and profits go up.

And I think we already see the shit experience customers get when customer service jobs are replaced by AI. I doubt that will ever improve, by design.

Remember, also, that computers only deal with situations and problems that they are programmed to deal with. AI is a little different, but still suffers the same limitations in that they can only deal with things they're trained on. Humans can make exceptions and adapt to new situations. If we get to AGI, perhaps that problem will go away, but I expect we'll be granted many new problems to deal with instead.

lol. Sure.

I’ve seen three of these implementations in contact centers. AI drives lower satisfaction and lower cost. That business is about delivering defined level of service at the lowest possible cost.

The advantage of Waymo is that it’s a first party service that doesn’t hide behind the fig leaf of an independent contractor. Easier to regulate those nexus points than to figure out of some dudes 2015 Sienna is safe or reliable.