Comment by version_five | Hacker News Reader

Comment by version_five | Hacker News Reader

Parent Full thread

> The most hyped technology in 1995 was Intelligent Agents. Two years later, Office 97 introduced Clippy, the enthusiastic, but incompetent assistant which was so poorly received that it effectively killed off the idea for a generation. Today, twenty years later, we’re once again trying to build intelligent assistants, although now we call them Chatbots

Chatbots are almost a hype cycle on top of a hype cycle. Deep neural networks gave us AI as a hype cycle, someone imagined what they thought AI could do and thought that could be used to prevent people from talking to a real person, leading to cost savings, and so we have hype around chatbots built on a foundation of hype around AI.

The reality is, AI doesn't do what people think it does, and people call in to get a sympathetic human assessment of their issues, not because they're illiterate. So chatbots have no future outside of customer hostile applications that think people will keep paying no matter how disrespectful and unhelpful support is (see Uber). Unfortunately, this seems to work for a lot of companies.

Alexa, Siri and OK Google are the equivalent of these intelligent agents, which can scheduled a calendar appointment, search the internet, find audio or video of an thing, play a song, call a contact, find a restaurant, take a note or tell you how long it will take to get to a place using any mode of transport. Those are significant capabilities, some unanticipated in the 1980s. They're not capable of intelligent conversation and have limitations, but they can get quite a lot done and are capable digital assistants very much in the tradition of the Apple knowledge navigator demo.

So while we were late in delivering the promise of those early ideas of digital assistants, I think global tech giants have mostly delivered on those promises, and even surpassed them in some respects (e.g. image recognition).

Chatbots are a separate topic and are mostly neutered by the power imbalance between and differing requirements of companies and customers - a chatbot designed by a company clearly serves their needs, not the customer who engages with it.

Both Siri et al and modern chatbots are just menu systems though. They are constrained to a finite list of things they can do, and have some intent resolution logic on top of that so that you can ask for things in slightly different ways and still get a response. It works alright for Siri because there is value in the closed domain of actions it provides, and users have been trained to stay within this domain. It works less well for a "support" chatbot that is really just acting as a gatekeeper for a closed menu system with which the user is not very familiar. I'd rather just be presented with a hierarchical list of actions I can take, and the option to talk to a person if my request is not there. Having to guess whether the menu behind a chatbot supports a particular action is supremely frustrating.

Intelligence is hard to define and easy to dismiss.

We've made huge strides over the last few decades in image recognition, text and image generation, modelling etc. I think it's unfair to dismiss digital assistants as just menu systems, yes they are limited, but then so are humans. Software is catching up with humans in every domain, including ones previous dismissed as impossible to tackle.

Software playing chess can be dismissed as just a decision tree but it has far surpassed human players and human intelligence in that domain. At some point I expect software to be able to pass the turing test and that we're not there yet does not indicate that these systems are useless or we have made no progress.

I would say that one of the hallmarks of any intelligent "thing" is the ability to adapt. AI doesn't always understand nuance (not in an AGI sense) and therefore I'm always skeptical when someone markets an "intelligent" agent to me.

I’m fine with a chat bot as long as it can pass me an actual human. The one that can only blurp FAQ links in loop are terrible

The worst kind are phone bots that when they fail at the speech reckognition game just hang up, or enter into question loop.

To beat the Vodafone one in Germany you have to speak random baby like gibberish until it finally decides to actually give you a human, but this is a matter of luck, as it can hangup instead.

You can imagine how fun it is to deal with Vodafone helpdesk.

Switching operator won't change much, as everyone uses the same toys.

I beat the USCIS one recently out of sheer luck. He understood my poor attempts at saying « reschedule » as « you have a home bounds request » and I was connected with a agent the second after.

I have no idea what is a home bound request but I will now say that.

You can cuss as well, as they are designed to flag inappropriate language.

That's the one I was thinking about. I wanted to contact a webshop to tell them one of their sellers misrepresented an article. Just flag it so they can take care of it. Instead I got a chatbot that quickly short-circuited me to the returns form.

I much prefer just having a form and a normal website.

At least then if I screw up it's on me.

Agreed chatbots have no future in their current crap state but really good chatbot that pass the Turing test (which hasn't happened yet), could I believe take over any human interfacing use case and we'd all actually enjoy it. This is still decades away and is doubtful will involve current AI paradigms.

The other answer though is that AI systems only learn what they're taught, and businesses building chatbots only teach them things they've already put on their websites and documentation...which if that was helpful to the customer, they wouldn't be looking for human support.

Equally true for a lot of human customer service!

Oh yes chatbots are definitely one of the larger steaming piles of sht that was totally hyped up in the last few years. I still find it hard to believe that so many startups companies dived right into it

HN readers will probably try quite hard to solve a problem by themselves before contacting customer support so are unlikely to be helped by a chatbot. But I wonder whether we are unrepresentative of most users in this respect. Are a lot of people calling support with easy problems which the chatbots are in fact able to solve? I’m curious whether they are actually providing a useful level 1 support.

Reply on news.ycombinator.com