13 ms·

Chatbots were never the “next big thing” by any measurable metric other than artificial hype created by the companies trying to create a new platform for extrac

Chatbots were never the “next big thing” by any measurable metric other than artificial hype created by the companies trying to create a new platform for extracting revenue and data from users. For this argument to make any sense, there would need to be data that showed growing usage of chatbots, followed by a plateau and then a drop off. I’m not sure what that data looks like, but I’m pretty sure there was never a high growth phase where users were actually interacting with the chatbots. It was all hype created by Facebook and a few idiotic VCs who wasted money on what they thought would be the “next App Store.” I, and many others, were saying around the time of this hype that chatbots would never become a category defining product, simply due to inherent usability flaws in their design that have been discussed ad nauseum. The only people surprised that chatbots failed to become the “next big thing” are the people who mistakenly thought they ever would be. This assumption was never grounded in any real data of user desires or real problems. Chatbots were then, and are now, a solution looking for a problem. I’m not surprised at all.

potatoblaster 8y ago
Yeah - AI in general sucks balls.
hasbroslasher 8y ago
I think there's a lot of hate in this thread for the bot concept. Yeah, most of these text-based bots are bad and useless. Expecting to chat with a computer over text and get meaningful results is a big request, and without the best of the best NLP researchers, you're not going to get anywhere. However, I do think there's some truth to this article - the rise of digital assistants in Google Home, Amazon's Alexa, Apple's Siri, is, in a sense, the rise of the chat bot. You can text them and they'll do things. You can talk to them and they'll do things. You can even get them to schedule a haircut for you[0], as Google recently demonstrated. I think these bots are semantically identical to what we think of chat bots as doing. If you would've told me 3 or 4 years ago, that people would willingly let a "bot" into their home that listens to everything they say and talks back, I would've been skeptical, to say the least. Now, even my hippie roommate has one of these things. I still don't get it, but to claim that "bots are dead, long live humans" misses the mark about how fast speech to computer/computer to speech tech is evolving. Watch the video and be... amazed: [0]https://www.theverge.com/2018/5/8/17332070/google-assistant-makes-phone-call-demo-duplex-io-2018 https://www.theverge.com/2018/5/8/17332070/google-assistant-... So yeah, the idiot VCs missed the mark - this isn't going to be a consumer revolution, led by a few scrappy start ups, one-man teams, and dreamers. It's a revolution in data collection, human-computer interaction, and AI that's already been taking place behind the scenes at Apple, Google, Facebook, and Amazon for years, and will continue to as long as they hold the tech world's best AI talent.
- ttul 8y ago
  Indeed. The bot interface is being produced by the giants because they are the only ones with the budget for it. Just like google was the only firm with the budget to make self driving cars actually work, and will likely dominate that field for years to come as a result. There will be room for startups to put the glue between the bots be everything else, but the core tech for the conversational interface is not going to come from startups.
- baldfat 8y ago
  It reminds me in the early days of "smart phones" people would say where is the Linux phone OS we keep hearing about. They just didn't understand that Android was Linux based. The Voice Assistance is a vocal chatbot. People were thinking it was going to be text. No one missed anything except the actual product everyone was talking about. I feel like this happens all the time. We get caught into one strict interpenetration and miss the HUGE thing staring at our face, or ears in this instance.
  - dozzie 8y ago
    > It reminds me in the early days of "smart phones" people would say where is the Linux phone OS we keep hearing about. They just didn't understand that Android was Linux based. It's actually the people who believe that Android is Linux-based that are mistaken. If you write for Android, you write for Android and nothing more. You can't just run the application on a regular Linux (i.e. the one with libc and X11), and if Google ever rips Linux kernel out of it and replaces with, say, DOS, most of the applications won't even need to be recompiled, because they were using Android's API/ABI.
    - TomMarius 8y ago
      That changes nothing about Android being Linux based, though. The word "based" has a definition, and you're trying to change it.
      - dozzie 8y ago
ttul 8y ago
There is a mythical internal document at Amazon that describes the Alexa vision. They want basically the Star Trek computer, and they are making good progress toward that vision. And the investment they are making to achieve that goal is hard to overstate. Yes, it will take a few years. But consider that there are now tens of millions of Alexa pucks out there. The bot may suck today, but it’s getting better quickly, and consumers are receptive and ready to reward the tech giants when they finally succeed.
- ghein 8y ago
  Amazon are the best people at this as they have the technical ability, vision, and willingness to invest in long term projects. AWS was not great in its first 5 years. Its capabilities 5 years ago were pretty good but nothing compared to today. Google is bad at long term development, especially with Ruth Porat at CFO. Microsoft is mixed - for every Xbox and Hololens there's a Courier. One thing missed in chatbots is that X dot AI is a chat bot that works pretty well. There are others that are similarly domain limited where the current interaction model is an email thread or a text conversation. These tend to be doing well.
  - romwell 8y ago
    >Microsoft is mixed - for every Xbox and Hololens there's a Courier. My perception of Microsoft is that it's a Hydra. The upside is that it's not all bad! The head responsible for the product you care for might have been there for a long time, and is doing a really good job in keeping it up. The downside is that the other heads might be doing something entirely different and turn the whole beast around . And while all this is going on, some heads just get chopped off here and there. (sniff Lumia sniff)
  - m1sta_ 8y ago
    The courier concept is far from dead
  - apapli 8y ago
    I looked at X dot AI, but cannot justify the price. I would do 20 meetings a week and place a high value on my time, but to me their value proposition is simply not aligned with the price they are asking. I just don't believe it will help me that much. As an aside, I tend to think about what I'd be prepared to pay for something before I click on the pricing page. In the case of X dot AI, it was about $5-10 per month. I'm sure it is good, but docusign (similar price) is MUCH more valuable to me - time kills deals and docusign helps bring them in. An AI bot that books in meetings just doesn't deliver the same ROI.
qznc 8y ago
I fully agree with you. I also dismissed Google Glass and other hypes. I feel smart about that. On the other hand, if someone would have pitched Facebook to me, I would have also dismissed it. I keep that mind to stay humble. For any trend/hype there will always be people dismissing and praising it. Thus, if it is successful or not, there will be people who say "I told you so" afterwards. It does not matter what a comment on HN says. You personally have to decide, if you want to speculate/invest in some hype or not. You have to decide which new technology you learn and which ones you ignore. In hindsight, it would have been great to learn machine learning five years ago.
- jandrese 8y ago
  Eh, Facebook was a natural evolution of social networks going back to MySpace, Friendstr, SixDegrees. Each one fixing mistakes of the past and becoming the next big thing until their own mistakes caught up to them. The one that I dismissed was Twitter. In fact I thought it was a joke when I first heard about it, something making fun of how long winded boring blog posts on services like Livejournal tended to be. I also thought Vine would be doomed to fail.
  - captain_perl 8y ago
    I also thought twitter was a joke until I realized most Americans are too lazy to write a blog post, or even read one. “No one ever went broke underestimating the intelligence of the American public.” - Mencken (paraphrased) http://www.thisdayinquotes.com/2011/09/no-one-ever-went-broke-underestimating.html http://www.thisdayinquotes.com/2011/09/no-one-ever-went-brok...
    - briandear 8y ago
      How many blog posts are being read and written by the French? The “Americans are too lazy” trope is just ridiculous.
      - monkpit 8y ago
        I find that many comment like gp online are equally valid with the word “Americans” replaced with “people”. It irks me as well.
bebnoreck 8y ago
I began my career building IVR systems in the early 2000s (Nuance, etc). In college I worked at my university's speech recognition lab. I remember an uneasy feeling while building and testing my first speech application. I thought: "This is a statistical algorithm with nothing else behind it. It's just a good guesser over a very limited domain of 'knowledge'." At the time the tech industry was hailing speech recognition as the next big thing. There was a lot of investment in speech: BeVocal, TellMe, Nuance, SpeechWorks, AT&T, L&H, etc. Replace your call center workers with automated systems, use voiceprints to secure sensitive transactions, etc. Amazing AI would effect cost savings and make your business efficient! Sound familiar? Very few users actually liked these systems over the previous versions that just used the phone keypad. Poor recognition accuracy really pisses users off. And most of the time the recognition errors aren't the user's fault. Ever try to recognize speech on a cellular channel using an acoustic model trained on a landline channel? Crap! This was certainly a problem during the early 2000s because of the emerging mobile phone market. The UIs (or VUIs as they were called) were awful since they mostly replicated the touchtone versions but added "you can say 'accounts' to...". They continue to be awful because, while speech recognition had gotten a lot of funding (DARPA, etc) in the past, the UI aspects were pretty much ignored and underfunded. People interact very differently when speech is used as the medium. The interaction automatically becomes social. Social interactions with machines are decently well studied (Cliff Nass for one), but the very nuanced aspects are difficult to bake into an IVR system, especially under pressure to deliver on a deadline. The experience feels unnatural very quickly and users either run for the door or starts mashing the keypad. Ultimately the web killed off industry interest in IVR systems. Want to know your account balance? Log in to a web site and it's right there. No need to dial in your account number, verification and run a gauntlet through a phone menu. Fast-forward to today and chatbots are the new IVR systems, but without the speech-to-text portion. The speech-to-text portion is the easiest part by far in those types of systems. There's still the need for a parser (shallow or otherwise), a way to pair up queries with responses/actions, the ability to track a dialogue and its context, and most importantly, a "natural" feel to the interaction. Chatting/texting and speaking are serial in nature; very slow and inefficient. A well-indexed FAQ or an intuitive, well-designed GUI is more useful than a chatbot. The former helps in shortening search time and the later helps at streamlining transactions. Information search and transacting comprise the majority of actions that users perform when they visit a site/system/application. When you talk to a human agent for these two types of actions, guess what they're using to assist you? The same interface you'd use if you served yourself. They just know it better. As an aside, the only noticeable/innovative use of chatbots I've seen in the past 10 years is on porn sites. I really took notice of the messages in the chat window when it started nagging me for ignore it. I wouldn't be surprised this round of chatbots were inspired by that.
hw 8y ago
The expectation for chatbots to completely solve and answer user questions and take over support is as flawed as the expectation that self-driving technologies in cars allow people to watch Youtube videos and not pay attention to the road or have a hand on the wheel. The successful bots are ones that don't attempt to understand and respond to every single question perfectly, but instead act as a supplemental tool / guide and offering concrete decision points and actions for the user to take. For example, instead of asking "How can we reach you?" and letting the user enter free text and figuring out if the user entered a phone number or email or some random text that just short circuits the bot, show two buttons "Via email", "Via phone" and clicking each one would then ask for an email, or phone number. The successful bots also know when to failover to a human being and failing over fast. I've hardly ever had a good experience dealing with sites that employ chat bots and it's not great to be frustrated by a bot when I'm already an angry customer who hasn't received my order.
- jacobwilliamroy 8y ago
  The one case where chatbots were better than people was at a company I used to work for. We kept getting customers contacting us via facebook to ask the same questions over and over again. "How much does `x` cost?" "When are you open?" "Where is your location?" "Do you have `y`?" I had to have an internet-connected phone on me at all times because people would send messages at any hour and response time was a factor in our discoverability on facebook. Of course all of these questions were answered on our website (AND ON OUR FACEBOOK PAGE) but simply responding with a URL resulted in attrition. So I found myself running through the same script several times, all day, all night, every day, every night. The best solution would have been to set up some kind of text parser which would allow people to navigate the website through facebook messenger (oversimplifying) and then alert a human if it couldn't parse the input. We could even have hooked it into the comment feed because facebook was really bad at notifying us of comments. Taking it a step further, the Mero Mero's dream of offering facebook marketing to other companies as a service would actually be reasonable, because it wouldn't just be me sending out hundreds of copypastas 24-hours a day, forever. But I was never able to put a system together because of other time-consuming duties the company needed me to perform.
rm445 8y ago
I think that did happen. Banks and utility companies started to offer chat-like interfaces (without a human responder) for services and help. The number of end-users could probably have been graphed as a steeply rising curve for a while. It just didn't justify the breathless expectations of the companies trying to sell chat interfaces at the time.
- chrisco255 8y ago
  Yeah, this is the "trough of disillusionment" phase on the Gartner hype cycle. If anyone is looking at the exponential improvements in speech and text comprehension over the past years and unable to see that it will only get better from here, then it's not hard to imagine a future, where in 10 years chat bots are just as good as humans for any rote request.
  - inthewoods 8y ago
    And yet there are companies that are starting to make money from these things. Maybe this is a classic "can't take new tech and apply old business model" kind of thing. https://www.theinformation.com/articles/for-developers-facebooks-messenger-shows-signs-of-life https://www.theinformation.com/articles/for-developers-faceb... Behind firewall, so here's a key quote: Facebook’s early efforts to turn Messenger into a service that people can use for shopping or getting news or weather sputtered quickly. But two years on, the Messenger effort is showing signs of life. Several developers say they have been selling more bots and applications through Messenger in recent months. Trip-booking service SnapTravel, for instance, has seen its cumulative sales through Messenger grow from around $1 million as of May 2017 to around $10 million now, according to CEO Hussein Fazal. Msg.ai, which develops tools for customer service and marketing for messaging apps, has seen enough of a pickup in business to hire a new executive and marketing team, said Msg.ai CEO Puneet Mehta.
hrktb 8y ago
I think the problem is framing, or essentially what is a “chat bot”. 20 years ago “artificial intelligence” where taboo words for anyone hoping any funding. It would be called ‘expert system’ or ‘automated agent’ or whatever else that didn’t make people show you the door. I think what you put as “chat bots” is having the same issue. The concept has been deployed at super large scale and people interact with bots everyday, it’s just not marketed as such. Currently my phone company, my ISP, my health insurance company, the last airline customer support I had to deal with, they all process a crazy amount of interactions, and all the basic steps were clearly handled by a bot until I got escalated to a human. They are all real world huge scale applications, I don’t know by what other metrics they would be deemed as “failures”, and I don’t think they plateaued, I expect it’s still growing.
- Nullabillity 8y ago
  > I don’t know by what other metrics they would be deemed as “failures” Usability? Sure, someone keeps paying for them, but they're not the people who get stuck interacting with them...
  - hrktb 8y ago
    It’s not too bad though. Human operators on big operations where already bound to a script for years now, and the “human contact” part was mostly spelling letters and numbers over the phone. I’m more than happy it switched to chat, and both sides can deal with the exchange asynchronously, in particual with the bot handling the requirements to open the user info. I feel it’s actually a win-win situation, as incredible as it seems regarding customer support technology.
- madeofpalk 8y ago
  > The concept has been deployed at super large scale and people interact with bots everyday Has it? Are these experiences people enjoy and would like to see more of?
hammerha 8y ago
Well, about 10 years ago before iPhone appeared, was the mobile the next big thing? In hindsight of course. How did the data look like before iPhone? Check out the chart in this article https://www.recode.net/2017/6/26/15821652/iphone-apple-10-year-anniversary-launch-mobile-stats-smart-phone-steve-jobs https://www.recode.net/2017/6/26/15821652/iphone-apple-10-ye... There's no data before a breakthrough. We should be wary of hypes but at the same time we shouldn't diminish pioneers.
- scarface74 8y ago
  If you looked at the trend lines, Mobile was definitely growing in 1997 (https://www.statista.com/statistics/203688/handset-penetration-per-capita-in-north-america-since-1996/ https://www.statista.com/statistics/203688/handset-penetrati...) I was working at Radio Shack in 1995-1996 and mobile phones were definitely starting to grow in popularity. By then, you already had subsidized “free” phones and mobile plans were around $35 a month.
  - hammerha 8y ago
    Okay, maybe I wasn't clear enough. When we say that "Mobile is eating the world" or something, we are not talking about the mobile phone we can make a call with but the phone we can use the internet and applications with. Your data is about the mobile device that allowed us to make a phone call. If you say that "this is also data" then we have that kind of data for chatbot as well, e.g., the number of sales of Alexa and Google Home devices, or Messenger installations, etc.
mo1ok 8y ago
Yep. I remember tons of marketing/advertising people going around enthusiastically talking about how chatbots are "the next big thing." No professional UX people or engineers were really pushing the tech that I knew.
everdev 8y ago
Alexa and Google Home are simple chat bots. Say something and get a single, reasonably accurate response. They seem to be doing pretty well. The little chat bot widgets never took off because they didn't solve NLP anywhere near as well as Amazon or Google did. If a next gen chat bot can make a dinner reservation over the phone (like Google's demo), it definitely will be the next big thing. So much of our day is communicating and if we can automate our communication effectively, it opens the doors for even greater productivity.
JohnJamesRambo 8y ago
I’d rather talk to a chatbot than send an email or shudder make a phone call. I’ve found my interactions with them to be good and they always hand me off to a real person if it gets too complicated. It seems to save me a lot of time when I use them usually.
- bonestamp2 8y ago
  I agree. I fully support chatbots for getting you to the right person or handling simple issues. There seem to be a lot fewer accidental hangups when using a chatbot.
mirimir 8y ago
But wait, 66% of tweets are reportedly generated by bots.[0] Maybe they aren't chatbots, in a strict sense, but it does seem like most Twitter "conversations" involve bots. There are lots of bots on Facebook too.[1] Automating typical social media behavior is laughably trivial. 0) https://www.vox.com/technology/2018/4/9/17214720/pew-study-bots-generate-two-thirds-of-twitter-links https://www.vox.com/technology/2018/4/9/17214720/pew-study-b... 1) https://mashable.com/2017/11/02/facebook-phony-accounts-admission/ https://mashable.com/2017/11/02/facebook-phony-accounts-admi...
TheKarateKid 8y ago
I remember being so unimpressed after Facebook's hype because their "bots" were no smarter than SmarterChild was back on AOL in the early 2000's. (Which was at least innovative for its time.) I at least thought that there would be some Alexa/Siri-level AI going on, but most chat bots are about as intelligent as texting commands to an SMS short-code number.
honestoHeminway 8y ago
There are problems to which a well trained chat bot would be the solution. Teaching comes to mind, cause human patience is limited while a NN-Chatbot trained on stackoverflows datas answers are infinitely patient, and if guided by a well designed "stuck"-detector built into a ide, could fee like the patient mentor we always wished for. So the potential is there, but there is a catcher in the rye. The data, you get is difficult to interpret, ambigious and incomplete and self-referential. So good conversation bots are too hard yet for ML.
Izkata 8y ago
> I’m pretty sure there was never a high growth phase where users were actually interacting with the chatbots. There was somewhere around 2003-2005, when a couple good ones started showing up on AIM. Since that was the IM of choice at the time, there was zero friction to actually interacting with one. Anyone predicting it as a "next big thing" after around 2007 totally missed the actual hype and was trying to build on something users had already dismissed as not actually the future.
jnsaff2 8y ago
I love this comment and I kind of want to replace "chatbots" with "cryptocurrency" in a few years.
knoxxs 8y ago
What I have understood from my little experience is usually the "next big things" - Always seem way to big in the start - There is good exponential acceptance initially - But then there is always slow acceptance rate I think its more about chat then chatbots. Chat has already become the next big thing between customer to customer interaction. What chatbots are trying to do is enabling brands to start interacting with their customer on chat. Human-based support was always there but to able to interact with customers using chat during his/her complete lifecycle was not possible using human agents. But chatbot can solve that. We are moving back to chat because it is one of the two natural media of communication (voice/chat & gesture) to humans. The hype about chatbots was bad, resulting in everyone trying their own version of chatbots and fail. But for those who used it in its actual capabilities, it has done wonders. For some top financial brands of India, it has increased no. of people interacting with a brand to about 400% resulting in ~200% increase in marketing leads and lead quality by ~150%. This is actually the NEXT BIG THING for the marketing department. Also, chat apps are more evolved then SMS/phone apps which give user/customer more control over communication, resulting in happy customers. Chatbots are not solving any problem as they are not a solution. Instead, they should be used as a tool to solve any other problem, just like AI/Crypto.

potatoblaster 8y ago

Yeah - AI in general sucks balls.

hasbroslasher 8y ago

I think there's a lot of hate in this thread for the bot concept. Yeah, most of these text-based bots are bad and useless. Expecting to chat with a computer over text and get meaningful results is a big request, and without the best of the best NLP researchers, you're not going to get anywhere. However, I do think there's some truth to this article - the rise of digital assistants in Google Home, Amazon's Alexa, Apple's Siri, is, in a sense, the rise of the chat bot. You can text them and they'll do things. You can talk to them and they'll do things. You can even get them to schedule a haircut for you[0], as Google recently demonstrated. I think these bots are semantically identical to what we think of chat bots as doing. If you would've told me 3 or 4 years ago, that people would willingly let a "bot" into their home that listens to everything they say and talks back, I would've been skeptical, to say the least. Now, even my hippie roommate has one of these things. I still don't get it, but to claim that "bots are dead, long live humans" misses the mark about how fast speech to computer/computer to speech tech is evolving. Watch the video and be... amazed: [0]https://www.theverge.com/2018/5/8/17332070/google-assistant-makes-phone-call-demo-duplex-io-2018 https://www.theverge.com/2018/5/8/17332070/google-assistant-... So yeah, the idiot VCs missed the mark - this isn't going to be a consumer revolution, led by a few scrappy start ups, one-man teams, and dreamers. It's a revolution in data collection, human-computer interaction, and AI that's already been taking place behind the scenes at Apple, Google, Facebook, and Amazon for years, and will continue to as long as they hold the tech world's best AI talent.

ttul 8y ago
Indeed. The bot interface is being produced by the giants because they are the only ones with the budget for it. Just like google was the only firm with the budget to make self driving cars actually work, and will likely dominate that field for years to come as a result. There will be room for startups to put the glue between the bots be everything else, but the core tech for the conversational interface is not going to come from startups.
baldfat 8y ago
It reminds me in the early days of "smart phones" people would say where is the Linux phone OS we keep hearing about. They just didn't understand that Android was Linux based. The Voice Assistance is a vocal chatbot. People were thinking it was going to be text. No one missed anything except the actual product everyone was talking about. I feel like this happens all the time. We get caught into one strict interpenetration and miss the HUGE thing staring at our face, or ears in this instance.
- dozzie 8y ago
  > It reminds me in the early days of "smart phones" people would say where is the Linux phone OS we keep hearing about. They just didn't understand that Android was Linux based. It's actually the people who believe that Android is Linux-based that are mistaken. If you write for Android, you write for Android and nothing more. You can't just run the application on a regular Linux (i.e. the one with libc and X11), and if Google ever rips Linux kernel out of it and replaces with, say, DOS, most of the applications won't even need to be recompiled, because they were using Android's API/ABI.
  - TomMarius 8y ago
    That changes nothing about Android being Linux based, though. The word "based" has a definition, and you're trying to change it.
    - dozzie 8y ago

ttul 8y ago

There is a mythical internal document at Amazon that describes the Alexa vision. They want basically the Star Trek computer, and they are making good progress toward that vision. And the investment they are making to achieve that goal is hard to overstate. Yes, it will take a few years. But consider that there are now tens of millions of Alexa pucks out there. The bot may suck today, but it’s getting better quickly, and consumers are receptive and ready to reward the tech giants when they finally succeed.

ghein 8y ago
Amazon are the best people at this as they have the technical ability, vision, and willingness to invest in long term projects. AWS was not great in its first 5 years. Its capabilities 5 years ago were pretty good but nothing compared to today. Google is bad at long term development, especially with Ruth Porat at CFO. Microsoft is mixed - for every Xbox and Hololens there's a Courier. One thing missed in chatbots is that X dot AI is a chat bot that works pretty well. There are others that are similarly domain limited where the current interaction model is an email thread or a text conversation. These tend to be doing well.
- romwell 8y ago
  >Microsoft is mixed - for every Xbox and Hololens there's a Courier. My perception of Microsoft is that it's a Hydra. The upside is that it's not all bad! The head responsible for the product you care for might have been there for a long time, and is doing a really good job in keeping it up. The downside is that the other heads might be doing something entirely different and turn the whole beast around . And while all this is going on, some heads just get chopped off here and there. (sniff Lumia sniff)
- m1sta_ 8y ago
  The courier concept is far from dead
- apapli 8y ago
  I looked at X dot AI, but cannot justify the price. I would do 20 meetings a week and place a high value on my time, but to me their value proposition is simply not aligned with the price they are asking. I just don't believe it will help me that much. As an aside, I tend to think about what I'd be prepared to pay for something before I click on the pricing page. In the case of X dot AI, it was about $5-10 per month. I'm sure it is good, but docusign (similar price) is MUCH more valuable to me - time kills deals and docusign helps bring them in. An AI bot that books in meetings just doesn't deliver the same ROI.

qznc 8y ago

I fully agree with you. I also dismissed Google Glass and other hypes. I feel smart about that. On the other hand, if someone would have pitched Facebook to me, I would have also dismissed it. I keep that mind to stay humble. For any trend/hype there will always be people dismissing and praising it. Thus, if it is successful or not, there will be people who say "I told you so" afterwards. It does not matter what a comment on HN says. You personally have to decide, if you want to speculate/invest in some hype or not. You have to decide which new technology you learn and which ones you ignore. In hindsight, it would have been great to learn machine learning five years ago.

jandrese 8y ago
Eh, Facebook was a natural evolution of social networks going back to MySpace, Friendstr, SixDegrees. Each one fixing mistakes of the past and becoming the next big thing until their own mistakes caught up to them. The one that I dismissed was Twitter. In fact I thought it was a joke when I first heard about it, something making fun of how long winded boring blog posts on services like Livejournal tended to be. I also thought Vine would be doomed to fail.
- captain_perl 8y ago
  I also thought twitter was a joke until I realized most Americans are too lazy to write a blog post, or even read one. “No one ever went broke underestimating the intelligence of the American public.” - Mencken (paraphrased) http://www.thisdayinquotes.com/2011/09/no-one-ever-went-broke-underestimating.html http://www.thisdayinquotes.com/2011/09/no-one-ever-went-brok...
  - briandear 8y ago
    How many blog posts are being read and written by the French? The “Americans are too lazy” trope is just ridiculous.
    - monkpit 8y ago
      I find that many comment like gp online are equally valid with the word “Americans” replaced with “people”. It irks me as well.

bebnoreck 8y ago

I began my career building IVR systems in the early 2000s (Nuance, etc). In college I worked at my university's speech recognition lab. I remember an uneasy feeling while building and testing my first speech application. I thought: "This is a statistical algorithm with nothing else behind it. It's just a good guesser over a very limited domain of 'knowledge'." At the time the tech industry was hailing speech recognition as the next big thing. There was a lot of investment in speech: BeVocal, TellMe, Nuance, SpeechWorks, AT&T, L&H, etc. Replace your call center workers with automated systems, use voiceprints to secure sensitive transactions, etc. Amazing AI would effect cost savings and make your business efficient! Sound familiar? Very few users actually liked these systems over the previous versions that just used the phone keypad. Poor recognition accuracy really pisses users off. And most of the time the recognition errors aren't the user's fault. Ever try to recognize speech on a cellular channel using an acoustic model trained on a landline channel? Crap! This was certainly a problem during the early 2000s because of the emerging mobile phone market. The UIs (or VUIs as they were called) were awful since they mostly replicated the touchtone versions but added "you can say 'accounts' to...". They continue to be awful because, while speech recognition had gotten a lot of funding (DARPA, etc) in the past, the UI aspects were pretty much ignored and underfunded. People interact very differently when speech is used as the medium. The interaction automatically becomes social. Social interactions with machines are decently well studied (Cliff Nass for one), but the very nuanced aspects are difficult to bake into an IVR system, especially under pressure to deliver on a deadline. The experience feels unnatural very quickly and users either run for the door or starts mashing the keypad. Ultimately the web killed off industry interest in IVR systems. Want to know your account balance? Log in to a web site and it's right there. No need to dial in your account number, verification and run a gauntlet through a phone menu. Fast-forward to today and chatbots are the new IVR systems, but without the speech-to-text portion. The speech-to-text portion is the easiest part by far in those types of systems. There's still the need for a parser (shallow or otherwise), a way to pair up queries with responses/actions, the ability to track a dialogue and its context, and most importantly, a "natural" feel to the interaction. Chatting/texting and speaking are serial in nature; very slow and inefficient. A well-indexed FAQ or an intuitive, well-designed GUI is more useful than a chatbot. The former helps in shortening search time and the later helps at streamlining transactions. Information search and transacting comprise the majority of actions that users perform when they visit a site/system/application. When you talk to a human agent for these two types of actions, guess what they're using to assist you? The same interface you'd use if you served yourself. They just know it better. As an aside, the only noticeable/innovative use of chatbots I've seen in the past 10 years is on porn sites. I really took notice of the messages in the chat window when it started nagging me for ignore it. I wouldn't be surprised this round of chatbots were inspired by that.

hw 8y ago

The expectation for chatbots to completely solve and answer user questions and take over support is as flawed as the expectation that self-driving technologies in cars allow people to watch Youtube videos and not pay attention to the road or have a hand on the wheel. The successful bots are ones that don't attempt to understand and respond to every single question perfectly, but instead act as a supplemental tool / guide and offering concrete decision points and actions for the user to take. For example, instead of asking "How can we reach you?" and letting the user enter free text and figuring out if the user entered a phone number or email or some random text that just short circuits the bot, show two buttons "Via email", "Via phone" and clicking each one would then ask for an email, or phone number. The successful bots also know when to failover to a human being and failing over fast. I've hardly ever had a good experience dealing with sites that employ chat bots and it's not great to be frustrated by a bot when I'm already an angry customer who hasn't received my order.

jacobwilliamroy 8y ago
The one case where chatbots were better than people was at a company I used to work for. We kept getting customers contacting us via facebook to ask the same questions over and over again. "How much does `x` cost?" "When are you open?" "Where is your location?" "Do you have `y`?" I had to have an internet-connected phone on me at all times because people would send messages at any hour and response time was a factor in our discoverability on facebook. Of course all of these questions were answered on our website (AND ON OUR FACEBOOK PAGE) but simply responding with a URL resulted in attrition. So I found myself running through the same script several times, all day, all night, every day, every night. The best solution would have been to set up some kind of text parser which would allow people to navigate the website through facebook messenger (oversimplifying) and then alert a human if it couldn't parse the input. We could even have hooked it into the comment feed because facebook was really bad at notifying us of comments. Taking it a step further, the Mero Mero's dream of offering facebook marketing to other companies as a service would actually be reasonable, because it wouldn't just be me sending out hundreds of copypastas 24-hours a day, forever. But I was never able to put a system together because of other time-consuming duties the company needed me to perform.

rm445 8y ago

I think that did happen. Banks and utility companies started to offer chat-like interfaces (without a human responder) for services and help. The number of end-users could probably have been graphed as a steeply rising curve for a while. It just didn't justify the breathless expectations of the companies trying to sell chat interfaces at the time.

chrisco255 8y ago
Yeah, this is the "trough of disillusionment" phase on the Gartner hype cycle. If anyone is looking at the exponential improvements in speech and text comprehension over the past years and unable to see that it will only get better from here, then it's not hard to imagine a future, where in 10 years chat bots are just as good as humans for any rote request.
- inthewoods 8y ago
  And yet there are companies that are starting to make money from these things. Maybe this is a classic "can't take new tech and apply old business model" kind of thing. https://www.theinformation.com/articles/for-developers-facebooks-messenger-shows-signs-of-life https://www.theinformation.com/articles/for-developers-faceb... Behind firewall, so here's a key quote: Facebook’s early efforts to turn Messenger into a service that people can use for shopping or getting news or weather sputtered quickly. But two years on, the Messenger effort is showing signs of life. Several developers say they have been selling more bots and applications through Messenger in recent months. Trip-booking service SnapTravel, for instance, has seen its cumulative sales through Messenger grow from around $1 million as of May 2017 to around $10 million now, according to CEO Hussein Fazal. Msg.ai, which develops tools for customer service and marketing for messaging apps, has seen enough of a pickup in business to hire a new executive and marketing team, said Msg.ai CEO Puneet Mehta.

hrktb 8y ago

I think the problem is framing, or essentially what is a “chat bot”. 20 years ago “artificial intelligence” where taboo words for anyone hoping any funding. It would be called ‘expert system’ or ‘automated agent’ or whatever else that didn’t make people show you the door. I think what you put as “chat bots” is having the same issue. The concept has been deployed at super large scale and people interact with bots everyday, it’s just not marketed as such. Currently my phone company, my ISP, my health insurance company, the last airline customer support I had to deal with, they all process a crazy amount of interactions, and all the basic steps were clearly handled by a bot until I got escalated to a human. They are all real world huge scale applications, I don’t know by what other metrics they would be deemed as “failures”, and I don’t think they plateaued, I expect it’s still growing.

Nullabillity 8y ago
> I don’t know by what other metrics they would be deemed as “failures” Usability? Sure, someone keeps paying for them, but they're not the people who get stuck interacting with them...
- hrktb 8y ago
  It’s not too bad though. Human operators on big operations where already bound to a script for years now, and the “human contact” part was mostly spelling letters and numbers over the phone. I’m more than happy it switched to chat, and both sides can deal with the exchange asynchronously, in particual with the bot handling the requirements to open the user info. I feel it’s actually a win-win situation, as incredible as it seems regarding customer support technology.
madeofpalk 8y ago
> The concept has been deployed at super large scale and people interact with bots everyday Has it? Are these experiences people enjoy and would like to see more of?

hammerha 8y ago

Well, about 10 years ago before iPhone appeared, was the mobile the next big thing? In hindsight of course. How did the data look like before iPhone? Check out the chart in this article https://www.recode.net/2017/6/26/15821652/iphone-apple-10-year-anniversary-launch-mobile-stats-smart-phone-steve-jobs https://www.recode.net/2017/6/26/15821652/iphone-apple-10-ye... There's no data before a breakthrough. We should be wary of hypes but at the same time we shouldn't diminish pioneers.

scarface74 8y ago
If you looked at the trend lines, Mobile was definitely growing in 1997 (https://www.statista.com/statistics/203688/handset-penetration-per-capita-in-north-america-since-1996/ https://www.statista.com/statistics/203688/handset-penetrati...) I was working at Radio Shack in 1995-1996 and mobile phones were definitely starting to grow in popularity. By then, you already had subsidized “free” phones and mobile plans were around $35 a month.
- hammerha 8y ago
  Okay, maybe I wasn't clear enough. When we say that "Mobile is eating the world" or something, we are not talking about the mobile phone we can make a call with but the phone we can use the internet and applications with. Your data is about the mobile device that allowed us to make a phone call. If you say that "this is also data" then we have that kind of data for chatbot as well, e.g., the number of sales of Alexa and Google Home devices, or Messenger installations, etc.

mo1ok 8y ago

Yep. I remember tons of marketing/advertising people going around enthusiastically talking about how chatbots are "the next big thing." No professional UX people or engineers were really pushing the tech that I knew.

everdev 8y ago

Alexa and Google Home are simple chat bots. Say something and get a single, reasonably accurate response. They seem to be doing pretty well. The little chat bot widgets never took off because they didn't solve NLP anywhere near as well as Amazon or Google did. If a next gen chat bot can make a dinner reservation over the phone (like Google's demo), it definitely will be the next big thing. So much of our day is communicating and if we can automate our communication effectively, it opens the doors for even greater productivity.

JohnJamesRambo 8y ago

I’d rather talk to a chatbot than send an email or shudder make a phone call. I’ve found my interactions with them to be good and they always hand me off to a real person if it gets too complicated. It seems to save me a lot of time when I use them usually.

bonestamp2 8y ago
I agree. I fully support chatbots for getting you to the right person or handling simple issues. There seem to be a lot fewer accidental hangups when using a chatbot.

mirimir 8y ago

But wait, 66% of tweets are reportedly generated by bots.[0] Maybe they aren't chatbots, in a strict sense, but it does seem like most Twitter "conversations" involve bots. There are lots of bots on Facebook too.[1] Automating typical social media behavior is laughably trivial. 0) https://www.vox.com/technology/2018/4/9/17214720/pew-study-bots-generate-two-thirds-of-twitter-links https://www.vox.com/technology/2018/4/9/17214720/pew-study-b... 1) https://mashable.com/2017/11/02/facebook-phony-accounts-admission/ https://mashable.com/2017/11/02/facebook-phony-accounts-admi...

TheKarateKid 8y ago

I remember being so unimpressed after Facebook's hype because their "bots" were no smarter than SmarterChild was back on AOL in the early 2000's. (Which was at least innovative for its time.) I at least thought that there would be some Alexa/Siri-level AI going on, but most chat bots are about as intelligent as texting commands to an SMS short-code number.

honestoHeminway 8y ago

There are problems to which a well trained chat bot would be the solution. Teaching comes to mind, cause human patience is limited while a NN-Chatbot trained on stackoverflows datas answers are infinitely patient, and if guided by a well designed "stuck"-detector built into a ide, could fee like the patient mentor we always wished for. So the potential is there, but there is a catcher in the rye. The data, you get is difficult to interpret, ambigious and incomplete and self-referential. So good conversation bots are too hard yet for ML.

Izkata 8y ago

> I’m pretty sure there was never a high growth phase where users were actually interacting with the chatbots. There was somewhere around 2003-2005, when a couple good ones started showing up on AIM. Since that was the IM of choice at the time, there was zero friction to actually interacting with one. Anyone predicting it as a "next big thing" after around 2007 totally missed the actual hype and was trying to build on something users had already dismissed as not actually the future.

jnsaff2 8y ago

I love this comment and I kind of want to replace "chatbots" with "cryptocurrency" in a few years.

knoxxs 8y ago

What I have understood from my little experience is usually the "next big things" - Always seem way to big in the start - There is good exponential acceptance initially - But then there is always slow acceptance rate I think its more about chat then chatbots. Chat has already become the next big thing between customer to customer interaction. What chatbots are trying to do is enabling brands to start interacting with their customer on chat. Human-based support was always there but to able to interact with customers using chat during his/her complete lifecycle was not possible using human agents. But chatbot can solve that. We are moving back to chat because it is one of the two natural media of communication (voice/chat & gesture) to humans. The hype about chatbots was bad, resulting in everyone trying their own version of chatbots and fail. But for those who used it in its actual capabilities, it has done wonders. For some top financial brands of India, it has increased no. of people interacting with a brand to about 400% resulting in ~200% increase in marketing leads and lead quality by ~150%. This is actually the NEXT BIG THING for the marketing department. Also, chat apps are more evolved then SMS/phone apps which give user/customer more control over communication, resulting in happy customers. Chatbots are not solving any problem as they are not a solution. Instead, they should be used as a tool to solve any other problem, just like AI/Crypto.