Hacker News Reader

Comment by pryelluw | Hacker News Reader

To save you time:

Author praises advances in AI by big tech. Complains about how he was served the wrong medication and how a robot would not have made the error. Closes by saying that robots will be better than doing things than humans.

Its a shitty post that does not even really take into account the current state of AI, how robots are prone to errors as well as humans due to faulty hardware, and well, the fact that some jobs are only trusted to humans. Even if the margin of error may be higher.

View on HN

> Its a shitty post that does not even really take into account the current state of AI, how robots are prone to errors as well as humans due to faulty hardware, and well, the fact that some jobs are only trusted to humans. Even if the margin of error may be higher.

Agreed. The author seems to be doing fairly irresponsible things with statistics, too: maybe "the statistical likelihood of dying from a self-driving car is like falling off a building and being struck by lightning on the way down" because there are so very few self-driving cars on the road, and they're all currently monitored by human operators? I think they'll eventually be safer than humans, but implying that we're already there is just wrong.

Agreed. It got me thinking about the balance of workers doing a good job and those doing a poor job in my experience. Perhaps my standards are lower than the author's, but I would say I've encountered maybe 50:1 good workers to bad.

You have 50 seamless interactions, one bad one, then you go home and say, "You won't believe what happened to me at the pharmacy. I can't wait for robots to take over."

A human can explain why it made an error. Good luck sussing out the "why?" when an AI makes an error.

A human can claim to tell you why it made an error. Good luck sussing out whether they're wrong, or lying.

A human has the capacity to give reasoning, right or wrong, which a solution can be built off of. No AI can do that.

Counterexample: an AI can give you a full stacktrace and memory dump or allow you to attach a debugger when it makes an error.

What system can do that?

A program running on the JVM could do any of those three, I meant more in general that the exact state of a computer program can be seen in terms of memory, while the human brain is still more of a black box (perhaps this will change in the future?)

Yes, but what AI running on the JVM produces a meaningful stacktrace? Neural net based AI can't do that as far as I know.

Indeed,

The post is more or less the worst way to introduce the topic of automation.

What's needed or not for automation of various jobs is so different as a topic from increased abilities of AI in particular. In 1902, the first clerkless stores open in NYC known, as automats.

Modern technology naturally may make clerkless stores and other approaches more appealing but the potential has existed for a long time. And that means that the degree of automation and job loss one sees is a complex question, hinging on both social and technical questions.

https://en.wikipedia.org/wiki/Automat

Yes, great point. Automation != AI. Its a common mistake.

The author kinda glosses over the fact that he himself didn't immediately recognize the wrong medication.

That wouldn't necessarily change the premise, and if anything would bolster it because the author isn't a robot either, and therefore is prone to the same human error he's complaining of.

I don't think it's unreasonable to assume that he didn't actually look at the medication until it was time to take it. Pharmacies tend to give you your medication in a bag, so if you don't unpack the bag immediately, you're not even going to lay eyes on the medication.

I know.

I always check my medication before I pay for it, let alone leave the pharmacy.

Its a shitty post that does not even really take into account the current state of AI

Actually it does, if briefly:

"Put simply, AI can instantly identify all the troublesome gene sequences of a beagle’s genome to determine the likelihood of certain diseases but has struggled to identify a beagle in a picture."

But the author was making a broader point about how the "promise" or "eventuality" of AGI obviates most work. You can certainly disagree, which many do, but I personally am betting it's right.

I did consider that point and agree to an extent. However, it seems to be written by someone that does not have enough knowledge or experience in the field. Someone who is merely repeating what could be considered a popular opinion. Which, actually, is my biggest issue with the post. Its unoriginal and brings nothing to the table. Typical venturebeat content. Words written with the only goal of generating traffic. Blogspam aimed at a smarter audience (you). It could have been so much better. Why not explore how AI could have prevented his medicine being served incorrectly? Why not investigate how a smart agent could use current state CV to make decisions and serve medications? It would have been valuable even if technically wrong. Its useful to understand how AI is understood by outsiders. But no, the author simply talks about the promise of AI. Which is fine, but brings nothing to the table. We've been talking about that promise for decades.

Will AI end up obviating most work? I don't know. But Id love to talk about that with people who are informed.

I'll talk to you about it!

In fact though, what you ask for is really, really hard to do, and I'm sure this author had a deadline and a senior editor to work with.

The reality of the situation is that this is a devastatingly complex topic. As in, the dependencies are overwhelming if you haven't been studying it for decades. For example to even scratch the surface you have to have a quality understanding of history, understand broad technology trends since before antiquity, understand macroeconomics, have specific technical knowledge (like how does a neural net work), recognize the importance of cognitive science and biology, and all of this goes into this stew of trying to understand the AGI landscape.

You can't boil that down into a feature length article.

Yes, absolutely. Id love to know what the author would have come up with. It helps understand how people perceive AI. It has become hard to see what the general perception is after working on it. He does shed some light but stops to change the subject.

And yes, it is hard to do. I once wrote what I thought was going to be a well received article on AI. I used Jobs analogy of "the bicycle of the mind" to make a point about how AI could end up turning the bicycle into a more powerful method of transport (I used a motorcycle). Thus supercharging our abilities even more. But decided not to publish it until things become more clear. I dont want to sell something that might turn out to be quite different to what most expect. Because who knows when/if the singilarity happens and how how it will impact us.

> Put simply, AI can instantly identify all the troublesome gene sequences of a beagle’s genome to determine the likelihood of certain diseases

That sounds as much of an AI as finding the negative numbers in an array.

Reply on news.ycombinator.com