Hacker News new | past | comments | ask | show | jobs | submit login
Chatbots need a personality (myralabs.com)
76 points by gregdale on Nov 28, 2017 | hide | past | web | favorite | 49 comments

We do not yet have chatbots. What we do have are publicly-facing undocumented nondeterministic command line interfaces which expect the user to guess the right commands. This interface further insults the user by pretending to be a person.

I thought chatbots were the worst form of customer service, but then I had to deal with WES. The only option they provided for customer service was a question form. I clearly laid out the three questions I had and submitted the form. Three days later, they sent a reply that didn't answer any of my questions. It seemed to be copied off of the first help page that came up if you searched the keywords of my first question. Their response was from a noreply@ address. So effectively, they took a useless chatbot, added a half-week delay, removed any possibility of interaction with it, and presented it as their only support option. I hate that company with every fiber of my being.

I know that many on HN really dislike chatbots because they don't understand their super special case. As a chatbot maker, with several different chatbots that are live, I can tell you that I've seen the data. Which concludes that chat bots do work, and our clients agree. Chatbots will be everywhere the coming years.

Some of our clients have over 5000 intents, which cover quite a lot of the business, still it is important to let the chatbot be humble and connect the user to a human when it is too difficult to help.

Could you expand a bit on the niche you see them filling? I think some of the negative perception of chatbots recently owes a lot to exactly what the OP said - gross overapplication in areas with very marginal benefit, if any, to user experience (e.g. why do I want to interact with your service though an opaque natural language interface versus less ambiguous search controls/interface?).

Chatbots are coming off the peak of inflated expectations, down into the trough of disillusionment.

I suspect many people have never even used search tools that were powerful and allowed for highly targeted searching.

Dealing with chatbots, I always feel like I'm in a badly done Infocom text adventure, trying to play guess the verb to solve an obtuse puzzle. I've done a little work with some of these NLP services, and it's staggering how similar matching utterances to intents is to adding a custom verb to the Inform 6 parser, on the implementation side.

I wouldn't say niche, there are hundreds of use cases. I believe we have just scratched the surface of possibilities yet. I think you could say, in every situation you would ask a human a question you could also ask a chatbot.

For me searching is garbage, because I get 10+ search results returned in often a random order based on key words. A FAQ with 5000 topics is enormous, and a typical search interface is a bad user experience for finding the answer for your question.

A chatbot, which is properly implemented and that cover the exact intent you are asking will return one answer, with maybe a few follow-up questions. Fast precise and natural, this is what people want.

because they don't understand their super special case

Perhaps you could you expand on the use case for mimicking a human personality, rather than presenting an interface that's obviously a piece of software that uses human-like language to the best of it's ability?

Got any pointers on getting into your line of business (e.g. as a business consultant)? What are the big problems you deal with in this space? I really see chatbots as disruptive for customer service, especially for power users, if implemented in the right way.

The big challenge is creating good data. We constantly analyze chat conversations. Sometimes our chatbot answer wrong because of a corner case that we didn't think about. It could be lack of training data, context mismatch or that follow-up intents are missing. The technology works, but the data structure needs improvement all the time. We have people that work with this and we call them "AI-Trainers".

Can you give some examples of the "data" and some concrete examples of chatbots "working"? What is this super special case you're talking about?

Well, one BIG difference with command line interfaces is that there's logic behind ("NLP") that guesses what the user wanted to do and matches it with the correct command (matching an input with an intent).

Heck, that's basically most of the value the bot frameworks I've tested provide (speech to text, some UI creation and a bit of domain framework being the rest).

And it's not a small amount of value, we currently have an SMS-based menu system, and people complain that autocorrect changes what they wanted to type in, that it doesn't recognize typos, etc. Only programmers and power users are comfortable with the rigidity of a command line.

I'd say that a well-designed chatbot IS better than a command-line interface because it combines the power of command line (combining commands) with not having to learn and master the commands (and being lenient towards typos and missing parameters and stuff like that).

I'm not a fan of chatbots (even though I work with them) but I do think there's a valid use case when public needs non-frequent interactions with a system (ie help desk or customer support or nonfrequent purchases), and they definitely always need a way to access a real person.

I've never seen one of these (well, minus the really stupid AIM and Skype spam bots), where do you find them? Are they on the sites that constantly nag you to click through to live support on every page? The few of those that I have used all seemed to be real humans.


(you need to log in to try it out)

> (you need to log in to try it out)

I was expecting you meant login to eBay... nope, you need a facebook account to muck with that.

Some companies replace their HR websites with chatbot interfaces.

Yup. Shilling my own work here, https://kryogenix.org/code/the-ux-of-text-fusion/ is a talk which explicitly calls that out -- we can do better than this, and whoever works out how to deliver chat interfaces where you can't imagine the underlying nodejs commands will have people love their stuff. (And the answer certainly isn't "pretend to be a human", either, as you know; be a bot, be known to be a bot, and have people relate to you anyway. But now I'm just reiterating the talk.)

agree with this cirgue. they shouldn't pretent it's bots or humans, just be honest and say some script is auto replying to sift out some nonsense and quick wins.

What would make a true 'chatbot'?

I couldn't disagree with article more. I don't want a damn chat bot. I want a way to solve my problems that's not pretending to talk to person. I don't what to have to guess a string to enter to get the results I want. Give me some sort of real UI. I can't see how a chat bot could be more effective than good documentation with good search functionality.

Once the word "bot" is used, I pretty much discount the entire experience. A bot cannot "be more effective than good documentation with good search functionality."

Messaging has potential as a channel, but "bots" are a misstep. Faking a personality makes it even worse. Pretending to understand the user is just utterly transparent.

Messaging is good at collecting data, prompting response, reminding people and triggering actions like calls or clicks. None of the strengths are bot-based.

I (fortunately) haven't had to interact with an online chatbot, but in the case of voice ones, shouting a few expletives usually gets me a real human. I wonder if the same thing works here.

It's not:

There's just a massive failure of leadership in tech companies (well, in general), so rather than deal with real challenges, they smother themselves in carbon-monoxide statistics -- things that don't set off alarms, but destroy the business through misoptimization.

"Your chatbot should be purposeful, reflective of your product’s voice, and simpatico with your users. One helpful design exercise is to produce an assistant persona and personality:"

For a business, a "chatbot" or any feature similar to it needs to do one thing and that is solve the users problem(s). If the user wants to do X within the app or learn about Y, the chatbot needs to help the user with that efficiently and better than a human can for the feature to be successful. The "chatbot having a personality" comes second to "solve the users problem."

If the users are completely happy with whatever chatbot they are using, then sure adding in some "personality" might be a good idea and increase engagement slightly, but a poorly-performing chatbot that can't help the user but has a personality isn't going to help the business at all.

I don't know any users that are ever happy with any chatbot. The only exceptions I can think of are Apple Siri and Amazon Alexa, that's only because they respond to your voice and save you the trouble of typing or clicking, not because they have any intrinsic value as chatbots.

I've often found that Alexa doesn't even save me that trouble. It's useful for the useless things that I wouldn't do manually. The things that took a lot of time take even longer with Alexa (if they're supported at all), and the small things ... great, changing the lights now takes 3 seconds instead of 5 seconds, so I do it twice as often.

Voice is just a medium, what I really want is more powerful abstractions. Do more with fewer interactions.

I don't want to ask Alexa what movies are playing nearby, that's as good as the 1985 moviefone service. I want to know whether there are movies worth watching, and have Alexa check movies/times/ratings/weather/traffic/my calendar and tell me "Yes - X is certified Fresh and <friend> recommended it on Twitter recently. It's playing at Y theater for $Z, you can walk there in time for the 8:00PM show. Would you like me to buy a ticket?"... I don't want to buy an HDMI cable from Amazon, I want a sample of physical/digital retailers with prices/ETAs/ratings...

Voice, GUI, heck even CLI is fine.

And when it comes to smart homes, I expect learning and proactivity. If I follow GSW, don't make me ask if they're playing, then ask to turn on the TV, then change the channels, then adjust the volume... learn that I care, ask if I care now, and get the game on ASAP if the answer is "yes".

I agree, a lot of chatbot projects seem to be built because of the hype + the team also gets to dabble with some "AI/ML." I think there are very few conversational agents or chatbots today that people use regularly, and of those I honestly don't think many people enjoy using them.

I'd much rather have a plain old index of help questions and answers than a chatbot to deal with.

Building bots is hard: there are few tools available of any sophistication, leading to a vicious cycle of trial and error (with users getting the worst of it).

Remember what the first websites looked like? Everyone was just making it up as they went along. We're in a similar situation today.

And you know what? Some people are pining for those sites nowadays. The fact that you can write rich, computation-heavy applications and serve them over HTTP in a browser is amazing, but that doesn't mean every website needs to be a rich, computation-heavy application.

Same with chatbots. They might have some valid use somewhere, but most of the ones I have seen so far are just there for the sake of being there.

Sure, but past performance is not a predictor of future results. Just because many of them are built poorly today doesn't mean they have to be built that way.

Users are never happy with chatbots, but they can be happy with chat experiences if they never realize there's a bot behind the scenes.

Which, up to this point, is still never. Young kids can probably be happy with chat bots though?

Agreed. The virtual equivalent of a Radioshack employee isn't the way.

That's a sick burn but I think it needs updating now that there isn't such a thing.

One could even argue what was so enraging about Clippy (and MS's recent foray into helpbots instead of actual support) was the personality-sans-utility.

No, I don't want to forcibly converse with a brick wall that clearly doesn't understand my question and is chipperly excited about that fact, fuck you.

So I went from loving my latest Windows 10 PC (-- second in a row that had been awesome, even!) to swearing off MS for years (again) in about 30 minutes of completely terrible bot-based customer support.

I sincerely hope that strategy works for them, and they're gaining income somewhere to offset what they've lost from me.

I don't think it will.

In my course on chatbot building (shameless plug https://cognitiveclass.ai/courses/how-to-build-a-chatbot/) I cover some of these important design decisions and recommend creating a prompt that, while concise, removes the guesswork. Giving it a name and injecting personality is a good idea, but you should announce that the user is talking to a chatbot.

In fact, two of the worst chatbot design flows, in my opinion, are: 1) Having the user try to figure out whether they are talking to a real person or a chatbot 2) Prompting the user to ask you anything (e.g., "Hi, how can I help you?") rather than guiding them on the scope.

Chatbots are like IVR systems and phone trees. It's a neat idea, but they are a pain in the ass, nobody likes them, and their benefit over a human is marginal in a best case scenario.

Unfortunately, the IVRs persisted. It's going to be more complex for chatbots though because they will have to compete with other automated solutions that are proven to work to some degree, including phone trees.

I think they are great, I can load them up in one window, keep working, respond when necessary, and get my problem solved. Much better than a phone call where I have to wait 30 minutes, and can't cut and paste.

Are you confusing chat bots with live chat with a human?

> “I am the psychotherapist. Please, describe your problems.”

"M-x doctor" aside, try getting on #emacs and talking to fsbot. I swear I have seen some of the eeriest exchanges between human and robot in that channel – it's not an AI, of course, but man it fakes it well. As someone in the channel once said: "Someone's cheating on their Turing test..."

"Your chatbot should be purposeful, reflective of your product’s voice, and simpatico with your users"

Agreed 100% but IMO, this is not a function of "personality" but rather a function of deeply understanding user intents. A bot cannot be purposeful if its own designers don't know its purpose from a user-centric perspective.

(Disclaimer: I work on Chatbase, a service for analyzing and optimizing bots)

“What I had not realized is that extremely short exposures to a relatively simple computer program could induce powerful delusional thinking in quite normal people.” -- oh man this is a fantastic quote. So true. AI will surely destroy humanity!

The problem is not that your chatbot needs a personality.

The problem is you implemented chatbot in the first place so you don't have to directly deal with customers.

spot on! I remember Bezos showing an ugly table in his office, saying "we want the customers to know we spend money on what counts for them".

I wonder how one could justify the bots...

Am I the only person on earth who wants my interactions with technology to simple, effective, and straightforward? I want to give simple, straightforward commands and receive a terse confirmation or explanation as a result.

Tech is so blessedly, unfailingly logical and we stupid humans have to sully that to make some of us feel happier, and not more effective :(

I feel bad for human customer service workers. It's already a job that demands a lot of emotional labor, without customers shouting "representative!" at them.

A notable distinction is to be drawn between a personality (indefinite subject) and personality (adjective).

Now you have two problems.

Applications are open for YC Summer 2019

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