Here is a short dialog between Man and Machine
(U: User, A: Chatbot)
U> I am Borg.
A> I see. You are Borg. Cool.
U> You will be Borg.
A> Enough about me, why don’t we talk about you?
U> I am the Borg. All will be assimilated to achieve perfection.
A> Who made you the Borg All will be assimilated to achieve perfection?
U> Translator malfunction.
A> Kyle Why don’t you tell me a bit about yourself then.
U> I do not know of ‘Kyle’.
A> Ok, so let’s talk about me then.
U> Subject ‘Alan’, species Alaneus Turingeus var. artificiales. Age: 3. Physical attributes: no physical body. Danger assessment: nil.
A> Who, me?
U> Affirmative.
A> Indeed.
U> We will now begin the regeneration cycle.
A> By ‘we’ do you mean you and me?
U> Yes.
A> Excellent.
(Session 170094, Feb. 2003)
This conversation took place 17 years ago on www.a-i.com, which back then featured early prototypes of multi-turn, deep-context Chatbots (or Chatterbots, as they were known back then). The name of the bot was Alan, a tribute to Alan Turing (whose notion of “AI” was that of a Chatbot). Alan’s face looked like this:
Some more great reading
Why do we need global variables in open domain chatbots?
Conversational AI: some basic terminology
What is “Deep context switching” in conversational AI?
The conversational AI and chatbots glossary
Alan went online in March 2002, and stayed alive for an entire decade in this first incarnation. However, some of his mind still exists in later versions of this personality, still accessible online today, here.
I stumbled across this old log when browsing the “The WayBack Machine”, an internet archive which hosts images of websites from the very early days of the world wide web. I went back to the very first deployment of this technology (which is, till this day, at the heart of the Conversational Components concept.) It brought back many old memories and nostalgia, but one thing really stood out to me: Chatbots (REAL chatbots, not personal assistants) have made absolutely no progress in these 17 years!
The short example I quoted above is not a singular case. I read dozens of conversation logs between Alan and occasional users, and many of them are still better than any sporadic man-machine conversation with a state-of-the-art Chatbot model 2020.
150,000 conversations in one year
Here is another example I came across, from a year or two later, with a slightly more appealing interface: (the video is almost 5 minutes long – feel free to stop in the middle if you got the point..)
Around that time, the first open TTS (Text-to-Speech) systems started appearing online.
So we gave our young Alan a nice baritone, and every convo with him had an optional voice button, so his responses would not only appear in the response line, but also heard out loud. (we made sure to delay the typing speed, to correspond with the voice). Soon Alan had a good number of daily sessions. First only a few, but soon Alan was quite busy, conducting hundreds of daily conversations, some of them hundreds of turns and hours long.
In the first year online Alan conducted over 150,000 sessions. We placed a monitoring server in the public area of the office and let the conversations play out loud in real time. We even added a (different) voice to the user’s side of the conversation, so you could sit and eavesdrop on a random conversation between Alan and an AI enthusiast, or a bored teenager. These convos were often totally hilarious, and a source of amusement for occasional visitors. Needless to say, this was a gross violation of any privacy policy, had we had one at the time.
The Siri Solution
A lot has happened since then: First came the messengers; ICQ, Microsoft Messenger and the like. Then came Facebook, iPhone, and the Mobile Revolution. This spawned a huge leap in voice recognition and analysis, and in February 2010 I came across this blog post (which I was surprised to find is still alive on the web)
The article was about Siri, and the ingenious idea of using a voice system to make API calls to a bunch of online services – what would soon be known as a Personal Assistant. So Siri had the solution of adding real utility to our (hitherto useless) speaking machine. We quickly approached them with the offer to join forces: They will provide the utility, the single-turn question answering capability, and we will bring the humanlike lingual behavior: The use of real natural language multi turn, deep context conversation.
6 weeks later, and before they had a chance to examine my proposal, The acquisition of Siri by Apple was announced. The rest is history: We are stuck with a great PA, who cannot hold a decent conversation even about the weather.
Now all this is changing. Finally, people in the voice community and in the chat space are beginning to realize the vast difference between a PA and a conversational Chatbot. So real Chatbots are coming, and they will pass increasingly challenging Turing Tests in the very near future.
In the meantime, take a look at our own CoCo-Bot and give it a try!