The new breed of chatbot talks more like a human, but that is not always good | Hacker News Reader

The new breed of chatbot talks more like a human, but that is not always good

nytimes.com

60 pointsIntronExon23 comments

"Let's stupidify a machine that can easily give you exact answers with all the impreciseness and ambiguity of human language."

That basically sums up what I think of these things. They're a horrible substitute for existing UI, and even worse for things that really require a real human.

I also think that current gen. of UI (touchscreens flat-design) has become such a degraded experience that these alternatives become more appealing.

Could you please elaborate on the degraded experience?

idk OP's position, but the lack of tactile feedback is probably the biggest flaw of touchscreen UIs. it's really nice to be able to give input to a device without actually looking at it.

it's a mild to moderate inconvenience on a phone, since you are looking at the screen most of the time you are using it and swype style keyboards remove most of the pain of typing on a virtual keyboard. it does make for a compromised music player, though, unless you can remap the hardware buttons to control media playback.

on the other hand, i find the touch UI trend in cars absolutely horrible. it's bad enough that you have to take your eyes off the road to read the center screen, but actually pressing a virtual button requires even more concentration to track where your finger is wrt the button and stabilize it against the motion of the vehicle.

I think in practice, these are meant to replace training for people. I.e. it works along side the chat representative in a big call center/customer service role and provides a first line of defense and just says RTFM.

It is basically a boring re-implementation of search, except that for many big companies it has the side effect of causing them to work harder to make their search actually work well. Even if the UI ends up being poor or temporary, the investments in better knowledge management and search ranking internally ends up being a net benefit.

Source: have seen behind the scenes at some larger outsourcers supporting these kinds of chatbots for customer service bots.

Heh. We're developing code that can impersonate the average human better than the average human. Well that about wraps it up for the Turing Test.

Someone needs to tell the author that Microsoft isn’t headquartered in Silicon Valley.

"Silicon Valley" has unfortunately become clickbait. We replaced the title with a representative sentence from the article.

Microsoft is still in grunge-town?

It's case of https://en.wikipedia.org/wiki/Metonymy. Silicon Valley is no longer just a physical place, but also represents the tech industry itself in popular imagination. Also, while the article is centered on Microsoft, I think the title is meant to be general, as shown by this passage:

> The project represents a much wider effort to build a new breed of computing system that is truly conversational. At companies like Facebook, Amazon, and Salesforce as well as Microsoft...

I was going to just brush this off as silly, but then thought of all the NewYork things that you see around (NY Bagels, NY Fashion, etc) and realize they suffer at least as much... so I can almost forgive the NYT.

Well, I think chatbots are good for marketers and online sales companies.

I think it is junk technology for consumers/anyone that has to interact with them

Paradoxically it's the future of customer support and what we are observing is an early stage. A few tweaks and semantic advances in deep learning here and there, and a regular person would be happy to interact with them for stuff like getting more info about their order, specs and availability of the product etc.

Why can’t these things be shown through the UI?

if done correctly it could just be a much faster way to get to something, versus digging through layers of menus, or trying to guess the right search terms.

A chatbot couldn't possibly be faster unless the website design is really horrible. Checking order status is a common use case for most online retailers and usually just 2 clicks away from the home page.

The problem is when you have a question slightly more complex than that and no access to a real human being. I recently enjoyed that situation with a medical benefit provider's phone menus, and it was infuriating. (Still haven't figured out how to get a human being on the phone and am waiting, 2 weeks later, for an answer to an email.)

Yes, that's a major flaw in their customer support. Still, in the near future most common tasks can be handled by AI and only specific ones forwarded to humans (until those can be automated away as well).

The trouble is that too many places seem to already think that a bot is sufficient. IMO it's going to be a VERY long time (if ever) before bots can entirely replace human customer service.

That being said "computer says no" is often a standard answer when you call a call center. They may be humans but if their guide doesn't have the answer to the question you are out of luck.

Sure, but a lot of times you have an open ended question or problem. Or need to ask about something undocumented.

Here are some questions I couldn't get bots to answer for me recently:

"This printed form you provided me for a regulated action requires an account number, but no account number is displayed anywhere on your website or statements or any communications I've had from you. Where do I find that information?"

"Your site took money out of my account, said it couldn't complete the transaction, and now refuses to return the money to my account. How can I fix this and either get my money back or get what I paid for?"

And sometimes it takes a human to answer "is it still the case that the answer is 'no' if these other unusual circumstances apply?"

Paradoxically nothing, it will always be a tremendously awful experience vs a competent and friendly CSR. I will choose a bad CSR over something not real.

Reply on news.ycombinator.com