Hacker News new | past | comments | ask | show | jobs | submit login

Isn't it weird that we have entire companies like Intercom or Rasa whose value add is pushing automated, AI-driven "assistants" onto websites, and then the companies that buy into that entire value add and codebases hacked on by ML experts find that none of it even works better than how it was in 1998?

I could understand the thought process at first glance. NLP is revolutionizing things. Startups want the lastest technology if it's backed by Arxiv papers underneath enough layers. There is the image of people working dead-end jobs in call centers answering questions that nobody wants to spend time on but need to get answered anyways. So why not just make all of that go away, by inserting this ready-made digital replacement into your package.json?

I feel like the point is being missed though. Because even a call center employee working at Comcast is still a human. I think people are getting way too ahead of themselves in thinking code and "neural" networks (what a misleading term) can suddenly automate all of these things that could never be automated before, like dynamic conversations. And even before that the article proves that a single HTML form element is 62% more effective than a VC-backed GPT2-powered "solution" to a problem nobody had. Why require a dynamic conversation at all if e-mail works better?

Maybe there's some property about human interaction with founders or call center workers that's fundamentally impossible to replicate with technology, at the philosophical level, and we'd be better off investing into human resources instead of spending all of this engineering effort trying to work around that fact. Or maybe there theoretically isn't such a limitation, but I have serious doubts that 2020 is the year we'd finally be ready to declare that we've gotten past them.

While intercom's advertising may like to tout bots and AI chat, I can't imagine that is why most companies use it. They have a plethora of other features, sort of like proactive marketing automation that make it useful.

If it were just about the chat we would have dropped them for a cheaper solution long ago.

They continue to roll out new features and charge more and more for them though, so they will eventually run us off. Just not yet.

It's unfortunate because AI in its current form has properties that would be amazing for human labor augmentation, but I would argue that the kind of AI system you can get for less than $10 million/1-2 years development time is not really well suited to replace a human agent.

The ML sprinkled over MS Office, Facebook's image tagging interface, GMail are often very useful to me, but I can't see anyone finding the suggestions so reliable they would be comfortable letting the systems silently work without guiding the process and approving changes.

It's good to have someone keeping an eye on the customer interactions. If you have such a deluge of support requests that this is a problem, maybe the actual solution is a simpler process or better communication up front.

> 62% more effective

This is the thing that is odd to me about this article.

They increased the number of conversations, ok. But are conversations valuable? How about actual conversions (not conversations)?

If conversions remained the same, but conversations went up, it seems pointless.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact