Question

Thoughts on Actuarial Science?

Answer 1

Just an FYI: The professor who “passed” ChatGPT on an operations management exam (with what he said probably would have been a B-, not exactly stellar for business school) gave it the exam it would be most likely to succeed on and said it was terrible at the math parts. Its passing grade came from its high capacity to regurgitate descriptive material in a way that the professor thought was well written.

> While Chat GPT3's results were impressive, Terwiesch noted that Chat GPT3 “at times makes surprising mistakes in relatively simple calculations at the level of 6th grade Math.”

Also:

> The present version of Chat GPT is “not capable of handling more advanced process analysis questions, even when they are based on fairly standard templates," Terwiesch added. "This includes process flows with multiple products and problems with stochastic effects such as demand variability.”

I don’t know anything about the bar exam and medical license exam claims. But the current version has its weakest points exactly where an actuary needs to be strongest.

Answer 2

I asked chatGPT some FAM questions, and it either couldn’t answer them, or did them incorrectly. Not saying medical exams or the Bar is easier than actuarial. Actuarial questions are harder for AI since it’s not a matter of how much info it has stored, which computers are far better at, it’s about knowing how different topics interact, which it can’t know without specifically being shown how to use the formulas in a particular way.

Answer 3

Good thing we're judgement workers in a constantly-changing industry (well, health is anyway).

Real talk - I've been marinating on it more, and I'm pretty stoked for what a great tool AI could be for pricing. Imagine just telling an AI in simple text, "There was a change in regulations effective at this time which is expected to impact XYZ services. In the projection period, we're trying to push people from ABC expensive service to DEF cheaper service with a certain level of success and an 8 month ramp-up. Make sure to control for those in your trend projections"

We would still have to evaluate how reasonable it is and be able to justify its results, but man what a load off some of the technical work to just have to illustrate at a high level what the AI is doing.

Eventually maybe we'll be entirely replaceable, but at that point, literally everyone is.

Answer 4

For ASA exams that are more math-heavy, I don't think so.

For FSA exams that are more memorization-based, yeah I think it's a death knell.

Answer 5

Experience and domain knowledge will always be needed. All this would show if accurate, is that exams are pointless.

Answer 6

Assuming that chat gpt was trained on a ton of data. This isn't really that impressive. I would've assumed that it could pass more complex exams / tests. I remember watching a video from Facebook's chief data scientist and she concluded that chatgpt isn't all that great due to its limitations.

There probably won't be a huge AI that can do everything but localized / specialized AI that can do some things really well

Answer 7

The question isn't if AI can pass actuarial exams (I'm sure you could train one to do so with minimal effort) but if you can train an AI to do the job of an actuary. The latter is going to be damn-near impossible. AI will make us better at our jobs but it won't replace us.

Answer 8

good luck tackling DOI objections, chatGPT

Answer 9

I’m have asked it a few regulatory work problems and design problems I am stuck on and it is nowhere close to helping me.

It has been able to tell me how to perform a few functions in excel, but I certainly expected that.

Answer 10

I don’t see the big deal about ChatGPT - I already rely on Google to tells me everything and I still have a job