Emotional labor is a problem, not a feature. A receptionist only needs to be polite with the guests. But unfortunately, receptionists are often subject to emotional abuse.

We've never had glorified chatbots like GPT-3 or GPT-3.5. I'm not just praising GPT; I've myself casually run through a few simulations with a hotel chain receptionist and an executive. The technology looks very competent, and aside from cost savings, there's also the customer service quality aspect (consistency particularly) and the element of removing staff from abuse.

The biggest challenge is integrating language models with live data. Making customer data accessible to them is not a problem technically (prepending prompts), but it could be a GDPR problem if a third party like OpenAI is involved (having to hand over data to a third party might make the AI receptionists unappealing to some customers -- maybe, needs to be tried). The other aspect is letting the AI make changes in a data model. But there are ways to solve that as well. When these obstacles are resolved - and there is a lot of incentive to fix them now - a lot of customer-facing reception-type work can be outsourced to AI.

By the way, some hotels are already very interested in chatbots for reception work. There has been a lot of talk about that in some chains since about 2020. Old-style NLP bots, too. But of course, GPT-3 capabilities are very appealing.

I just want to add, ordering through the McDonald's big touchscreen machine is much better than ordering through the human and that alone has elevated my affinity for McDonald's. Which other experiential leaps from replacing humans with machines remain unimplemented?

I don't think so for several reasons.

First, there's the uselessness of entry-level engineers. They are hired for their growth potential (that every human has, but language models don't) and are expected to grow into mid-levels and seniors. In most of the sw industry, entry-level roles are also not terminal, which means that such an engineer must get promoted at least +1L to keep working in the company. Someone (or something) perpetually stuck at the entry level is a bad value proposition for a sw company.

Secondly, the code that language models produce is buggy. They can, on occasion, produce amazing code and even entire codebases. But this is an exception, not the rule. You can generally prototype something or get an idea for something from language models, but you can only push very little of that into production. What good is code you can't use? You still need an engineer to oversee the language model's outputs.

Overall, if some company replaced their juniors with AI, that would be incompetent management.