They are not self-aware. They are certainly artificial, but not quite intelligent. They make interesting mistakes. Still, the number of chatbots (or intelligent assistants, automated customer selfservice agents, or even conversational artificial intelligence) is on the rise. What do we need these bots for?
Define a bot for me
A chatbot is a computer program that is either connected to another application (Facebook, Slack, and others) or directly to a website. Users (most often customers) can ask questions, which the built-in “intelligence” of the bot tries to understand and return an answer to – a process sometimes called Natural Language recognition. Bots have been around for quite some time – the first natural language bot was probably MIT’s Eliza, dating back to the 60s of the previous millennium.
Are they Turing complete?
The Turing test is used to find out if a machine can fool a user in such a way that the user things he is dealing with a real human. A machine that is Turing complete is therefore good at mimicking human behavior and intelligence, which is not necessarily the same as behaving intelligently. Ever since 1991, the Loebner prize – started by Hugh Loebner and the Cambridge Centre for Behavioral Studies and since 2014 run by the Society for the Study of Artificial Intelligence and Simulation of Behavior (http://www.aisb.org.uk/events/loebner-prize) is awarded to the most human-like chatbot. It is not without controversy though, with some claiming it is not helping in promoting the field and MIT scientist Marvin Minsky famously calling it obnoxious and stupid (https://www.salon.com/2003/02/26/loebner_part_one/).
So, what about Turing completeness? Well, not quite yet. There is a website tracking the progress of all intelligence (http://isturingtestpassed.github.io/).
Are they intelligent?
According to the previously listed website (http://isturingtestpassed.github.io/) made by Aleksandar Todorović, Chatbots are not intelligent in any way. And actually, Cleverbot (http://www.cleverbot.com/) – one of the more famous chatbots – is quite clear about it as well. When you ask Do you think bots will ever become intelligent? the answer is a clear No, I don’t. But these two sources hardly constitute scientific proof. Arxiv to the rescue! As stated by Mazumder, Ma and Liu: One major weakness is that they [chatbots] cannot learn new knowledge during the conversation process, i.e., their knowledge is fixed beforehand and cannot be expanded or updated during conversation. So much for AI, then.
Laughing, but at who?
In March 2018, the New York Times (https://www.nytimes.com/2018/03/08/business/alexa-laugh-amazon-echo.html) published a story on laughing Alexa, something which had been troubling users of the Amazon assistant for a while. According to Amazon, the technology sometimes mistakenly understood a random sentence as “Alexa, laugh” – which it merrily did.
It’s not the only goof-up of what is labelled as intelligence. In October 2017, the Telegraph (https://www.telegraph.co.uk/technology/2017/10/05/siri-mistakes-bulgarias-national-anthem-despacito-song/) reported that Apple’s Siri mistakenly identified Despacito as the national anthem of Bulgaria.
Are their mistakes part of learning, or is it just sloppy coding?
One could argue that making mistakes is part of the learning process of automated agents. After all, this is part of the human learning process. To me, this would qualify as an anthropomorphism – basically attributing human characteristics to an object (or animal, but that is quite a different story). If computer intelligence does not understand a command, it is either not advanced enough, or it contains programming errors.
But what about machine learning? Commonly, chatbots are trained by feeding it massive amounts of text or dialogues (in order to improve variance detection). In the end, however, most of them are still scripted.
Bring on the numbers
According to Grand View Research, the chatbot market was USD 190.8 Mio in 2016 and is expected to reach USD 1,250 Mio by 2025. According to Transparencymarketresearch, bots were at USD 113 Mio in 2015 and will reach USD 994.5 Mio in 2024. According to MarketsandMarkets, it was USD 703 Mio in 2016 and will increase to a staggering USD 3,172 Mio in 2021. So even though the numbers vary with quite an impressive margin, at least the analysts agree that some form of growth will occur.
But there’s more! In 2016, Venturebeat predicted a 200 Billion market (https://venturebeat.com/2016/05/01/the-200-billion-dollar-chatbot-disruption/).
Gartner predicts that even though in 2017 less than 2% of businesses are using virtual customer agents, this number will rise to 25% by 2020 (as referenced in https://www.crn.com.au/news/gartner-25-percent-of-customer-service-operations-will-use-chatbots-by-2020-485488). In an Oracle study (referenced by http://www.businessinsider.com/chatbot-market-stats-trends-size-ecosystem-research-2017-10), 80% of respondents from France, the Netherlands, South Africa and the UK (a strange mix of countries) plan to use chatbots by 2020.
How many bots do exist?
Facebook has at least 100,000 chatbots (according to Gartner). It could also be 30,000 (according to Marketinglang, https://marketingland.com/inside-chatbots-year-growing-pains-210182). Most of those answer questions on a single topic.
Facebook?
Well, yes. But Facebook’s virtual assistant M is dead. So are chatbots, according to Wired (https://www.wired.com/story/facebooks-virtual-assistant-m-is-dead-so-are-chatbots/). It appears they are moving on.
And Microsoft?
Remember Clippy?
Well, sorry for bringing that one up. Microsoft has a bot framework (https://dev.botframework.com/) and bot service available on Azure (https://azure.microsoft.com/en-us/services/bot-service/). And embedded into its Windows operating system is virtual assistant Cortana. Satya Nadella in 2016 famously stated that chatbots are the new apps (there are just too many references available, just Google it). Bu as Dailymail stated, calling customer services lines could soon get more frustrating(http://www.dailymail.co.uk/sciencetech/article-4925366/Microsoft-chatbots-replace-call-centre-workers.html).
Do other big brands have bots?
Of course! Apple has Siri, now available through its speakersystem Homepod. Amazon has Alexa, also available through its speakersystem Echo (and Sonos Echo). Google has an Assistant, available through a speakersystem called Home. IBM has Watson Assistant, without a speaker. Fortunately, not one of the traditional speaker manufacturers has plans to introduce smart assistants. One can hope.
What problems are bots trying to solve?
Bots are aimed at replacing customer service and sales representatives. Sometimes they are marketed as being able to answer the easy questions, with representatives able to free their time to answer the more complicated ones. However, when I ask the chatbot of an energy company “Ik wil overstappen van een andere leverancier” (I want to switch from another supplier), I do not expect the answer “Stap je over naar een andere leverancier?” (Are you transferring to another supplier?). And another happy potential customer goes to the competition.
Another problem bots are trying to solve is to be a know-all entry point to anything. Is the fact that you can ask Siri about the weather actually useful, or just a gimmick?
What about the users? Did anyone ask us?
A LivePerson study from 2017 (referenced by http://www.businessinsider.com/chatbots-are-gaining-traction-2017-5) found that 33% of consumers rated their chatbot experience as positive, 19% were negative, with the remaining 48% not really caring as long as their issue was resolved. That is not quite on par with callcenter satisfaction, which seems to be higher – even though numbers are difficult to find. Data from Comm100 shows for example that positive experiences from live chat are outperforming those from automated chat by significant margins. (see https://www.comm100.com/livechat/resources/live-chat-benchmark-report-2017/).
Can we now please move on?
Human dialogue is hard to automate. Chatbots have been around for years but show little to no progress in actually understanding context and dialogue. The implementation of a rules engine (or worse, a series of case-else or if-then’s) is not the same as artificial intelligence.
As stated by Matt Swanson of Venturebeat (referenced earlier) in 2016: Chatbots are poised to fundamentally change the way humans interact with machines within a five-year horizon.
Some people just refuse to let go! Another example is this author in chatbotsmagazine (https://chatbotsmagazine.com/the-rise-of-chatbots-ecommerce-rebirth-against-hype-1cd17bc53179): even though chatbots were rated as the third most overhyped sector in a First Round Capital survey of 869 tech enterpreneurs, the author happily refers to his own firm as to why the sector is not overhyped: An option to launch one’s own chatbot company is already offered by BRN.AI, who are constantly up on e-commerce’s most interesting cases and ready to contest all of the scepticism around chatbot industry.
That is like stating Squishes are not a passing fad because you are selling them.