Godfather of AI Geoffrey Hinton 60 Minutes Interview

I would like to think we will all be chilling on the beach. But we gotta start working towards that now and, well, we humans are not the best at that kind of thing.

And I would think accounting would be one of those jobs AI would swallow up in a damn hurry. Anything that can be done on a ledger/spreadsheet has to be on the chopping block.

 

Legit lol on the accounting thing. I think AI might could do the counting but someone is gonna screw up the reporting in the system and somebody like me is going to have to unwind a complicated maze of human errors to get everything right so we have accurate data and a private equity firm can confidently execute their roll up of the ears.

 

My personal view is that we're saving accounting as a test of emotional prowess once we hit consciousness.

If the AI gets bored, and does something else, we're there.

 

It really depends on what you’re talking about when you say accounting. It’s such a wide range of jobs and duties. At the lower levels, people that are really bookkeepers, yeah those folks are in trouble, though right now more due to offshoring than AI. At the higher levels we’re still a long way away.

 

Aight. And yeah he could look at a trade. Our neph in Memphis does really well as a welder. I need to look at what all it out there. Just trying to help match up his studies with what he’s good at.

 

Not being snarky, but I hesitate on saying anything is "a long way away" when it comes to AI. I fear it is a hell of a lot closer than most people realize, but as always, I pray to whatever silicon gods there are out there that I am wrong.

Second, what is accounting at higher levels? I know nothing about accounting outside of debits/credits and column A better add up to column B. Unless you are talking about Enron levels of "accounting", in which case, I bet AI will hide stuff a hell of a lot better than humans, especially if it has no conscience.

 

I remember you saying AI will never be conscious. Am I wrong or have you changed your mind?

 

Lower levels - Bookkeepers, accountants for both private and public companies. While I’m oversimplifying here these positions are frequently closer to regular jobs than they are a white collar career, some of these folks are CPA’s but most are not.

Higher levels - Most accounting students go into public accounting (and become CPA’s) after college. These people do financial statement audits for large corporations as well as tax work for large corporations as well as medium/small private businesses and wealthy individuals. I’m not a tax guy so I won’t attempt to speak on that, but much of the work required as a result of Enron and other scandals for financial statement audits is judgemental and a ways away from any kind of automation. A lot of it at the lower levels (what used to be performed by recent grads) is now offshored, but that has had a disastrous effect on quality. This is my opinion but I think there’s already a developing competency crisis where recent grads don’t get the experience needed to be effective because that work is offshored and they “review” it but don’t understand it.

 

Be careful saying this. I am serious, because AI is going to be better at this than you believe, because it can look at large data sets and draw out a lot of conclusions that are stunning. Same as looking at a chess or go board and coming up with a move that is just unbelievably majestic.

Again, I hope you are right, but do not think that purely 'judgemental' or 'creative' or 'woowoo' jobs are safe. I believe AI will be better at all of these than humans at some point. And quicker than you and I might realize or hope for.

 

I don’t know how to say this because you’d have to do it / see it to understand, but you can’t comprehend how insane and dumb the work I’m talking about is. AI can’t do it. That doesn’t mean you have to be smart to do it, per se, I’m not defending my (now former) tribe here. The applicable governing bodies could change things tomorrow where what I’m saying would no longer be true, but as it stands today we are a ways away I promise.

Unless we’re willing to accept dog shit audits that mean nothing, which is kinda already happening anyway and we’ll trend that way more and more until we get another Enron. Hell, we need AI to be able to do it because it doesn’t pay enough and college students are no longer pursuing it.

 

Also, to be clear, when I say dog shit audits I don’t necessarily mean that the financial statements have inaccurate information or fraud. Right now most “failed” audits fail because of procedural issues or errors by auditors dealing with things that could hypothetically lead to inaccurate information or fraud rather than actual inaccuracies. The framework of what an audit actually involves definitely could be tweaked where AI could take over a lot sooner.

 

If I said that I was wrong. I think I once said AI will struggle to program itself, and that was due to the halting problem. But generative AI is able to halt just fine, so... AI will be programming itself before long, and some other node will just roll it back if it gets sideways.

Consciousness is hard to quantify, though. I can't, with absolute certainty, say that I wasn't programmed to be self aware, and therefore my own consciousness isn't also artificial.

"I think therefore I am" becomes "thoughts occurred." It becomes a big B Belief, or a Faith, almost.

I think there will, for a long time, be a portion of the population, that no matter what, says AI was designed, and therefore not conscious. But some of those same will say I was also designed, but am conscious so... that's why it becomes a Belief.

 

Math, statistics, physics, accounting. Even CS is fine. CS is theory, not just programming.

Again, I don't think there is a wrong answer.

 

You should talk to IP about this, he may have theories. If I follow, audits aren't that much different than looking for plagiarism citations in a mountain of documents.

If the work order isn't different, just copy the process, and then profit?

 

Audits are very different, primarily because the rules and procedures are far more complicated than evaluating something for plagiarism.

 

I think that's right. I think it can get to a point that you couldn't distinguish it from an organic consciousness, which to me means they are essentially the same.

I also think human consciousness may not be as deep or elaborate as we like to believe.

 

Rules based things are easier to automate than things with large numbers of edge cases.

For things with large numbers of edge cases, CNNs are used to train the desired action, so that the next similar edge case produces a desired action. But we're better at training via simulation than one offs, so lots of edge cases is bad for training too.

I think that is what you mean, in place of "complicated," too many edge cases.

 

I think that’s fair. And that things change so fast that population of edge cases today is less than what you’ll have tomorrow.

 

There's been a few predicting the doom of accounting by AI, and maybe they are correct, but I just don't see it. The general gist of the tax side is the divide is going to be those who use AI and those who don't?

 

Definitely can't speak to accounting, but I know white collar roles which have been eliminated, and more on the horizon by transition to AI.

The other thing happening, which impacted me, is transitioning roles to cost-effective geographies as a temporary step in the move to AI. It's all about reaching the cost benefit threshold, and we're pretty much there. The payroll saved from moving western roles to India/PHP/Malaysia/Indonesia and Argentina (for now) will be rolled directly into AI development. Then those roles will also go away.