Chatbot, what art thou?

“To be, or not to be, that is the question: Whether ‘tis nobler in the mind to parse,
The slings and arrows of outrageous language,
Or to take Arms against a Sea of intent.”

In early 2012, at the big data startup I co-founded, we were sitting on an award-winning Hadoop based search engine, which seemed to offer new possibilities, if you accept that information can be fundamentally organized, discovered, connected very differently at scale. Underlying the capacity to handle petabytes of data with ease, it also marked a shift in how we could approach data beyond well framed, well structured queries; get more hypothetical, what if this or that! From the labs a combination of high performance engineering and the ability to process large troves of data was a powerful beam to shine on hitherto unsolved problems. We chose the oft neglected post purchase support experience. We imagined that servicing customers was characterized by rich variety of user situations requiring attention. We felt a solution that was not rigid, has the ability to learn, adapt to a range of hypothetical scenarios was ripe to pursue.

Our premise that Text is at the intersection of UX and Systems. Hence a chatbot.

We picked Google chat and Facebook messenger for the UX delivery relying on their implementation of XMPP  (discontinued in mid 2013).  XMPP is the Extensible Messaging and Presence Protocol, a set of open technologies for instant messaging, presence, multi-party chat, voice and video calls, collaboration, lightweight middleware, content syndication, and generalized routing of XML data. 

We launched Txtland…

Txtland 2013, a chatbot that fetches information to natural language queries.
Txtland screenshot. The grey blocks are user texting and purple is Txtland response. User commands were not pretentious and completely functional, direct, short.

Without going into rest of the story as to what happened to Txtland, a digression, I realized the primary design challenge was in getting user at ease that she is chatting with a program at the back. It was powerfully fast in performance and response. It had access to the latest and huge repository of information to parse and serve in a fraction of second. However, like the Deepmind’s AlphaGo beating application, so well trained, solely by itself, on AlphaGo, it would not know how to play scrabble. This narrowness of specialization against the broad spectrum of human intent and sheer variety, responding back with, as Stafford Beer would say, with attenuation, not amplification, remains the challenge technically. 

As users typed we parsed and picked up ‘action words’ like stock, weather, ticket, item, SR, etc. and responses in the chatbot would provide options against that ‘action word’ along with a rudimentary, ‘TYPE THIS’ to find out more… so on. It was quite elementary. The text to engage with Txtland, a program, was honest and more machine like. If you attached a # in front of a phrase, it became an action for Txtland. See screenshot below:

From a design point of you, I pondered, as I see a proliferation of chatbots across industries, is why do the creators of chatbots continue to imitate human style, knowing well it cant live upto that label. Not authentic!

Fundamental problem with chatbots is that the interface between the user and the chatbot or agent is the same as what is used in normal, regular conversations between one human and another human. And this is validated by recent research from Pegasystems (NASDAQ: PEGA), the software company empowering customer engagement at the world’s leading enterprises

There is an opportunity that remains unexplored today to redesign the container on the tool using which a human user knows and is comfortable to chat with an artificial agent or a bot. Such a design should include predefined canned phrases and gestures. Research needs to explore whether search gestures can also be used by bot to communicate broadly. What language should a machine bot deploy to communicate with humans?

That would be besides generated text. How does the language demonstrate a ‘machine culture’ where culture could be its ‘nature’ that organizes information, its ability to find correlations across these, find and serve with great speed. And learn along the way as to what is a high value coorelation and what is a low value in what context. Txtland was leveraging this aspect to not just be a Q&A type conversation, but it could also run backend scripts, respond back, look up knowledge bases and ultimately, in case of failure, it gave option of ‘should I dial in our customer support representative?’

Chatbot parses text to respond with actions such as running scripts at the backend. 

What Chatbots can be!

I speculate that this sort of progress and exploration could help the ongoing effort in the digital transformation of businesses, including automation of business processes, strapped on with a new manner of interacting with ‘cognitive computing.’ Thinking about these kind of Technologies and problems with a very different toolkit, one of designers can help define the future of this industry and innovation. Especially, if these intelligent chatbots can conversationaly learn from a human user how it can perform that same task? How it will allow itself to be ‘handheld’ as painting robots do (record and play) to learn under supervision? Or in other cases, immerse themselves into a data rich situation given a specific human specified goal, to learn unsupervised.

One benefit of having machines have a new language to communicate with humans, and that humans retain theirs as distinct from it, we could even block or program into the machines an ability to not process certain intimate or personal human phrases. This would limit machines to what we envisage for them to productively engage with and perform within those boundary conditions efficiently. Like a bot that is expert in psephology or another in string theory. 

Assuming one gets past this ability of a chatbot communication or its UX, then comes the challenge of figuring out if there is a hierarchy among them. Afterall, we have tasks that are mundane, repetitive to challenging and complex. Can these bots be designed for such hypothetical variety? Would it be the ‘machine intelligence’ that differentiates these? How smart or fast is it? Speed and accuracy of response are critical to peg them. Of course, this is knowledge in the realm of known – knowns! Another boundary condition. From such criteria, a chatbot persona can be shaped and presented in a unique, non-human space. 

Henry VI, Part III [IV, 1]

King Edward IV:  “Now, messenger, what letters or what news from France?”
Messenger:  “My sovereign liege, no letters; and few words, But such as I, without your special pardon,
Dare not relate.”
King Edward IV:  “Go to, we pardon thee: therefore, in brief, Tell me their words as near as thou canst guess them. What answer makes King Lewis unto our letters?”

Here in Shakespeare’s play, the ‘guess’ is politically loaded. Hiding or revealing the facts may lead to harsh consequences to the messenger facing King Edward IV. Again, the pardon that follows hazarding a guess, leads the messenger to confidently conjecture on the circumstances. Guess is what humans do a wonderful job with. Guessing is such a fine way to move forward. And in the case of a chatbot, especially a smart AI driven one. A guess in that context is more heuristic and less algorithmic. That could also explain, why a rule based engine in chatbots with a NLP strapped on comes across as rigid or duh!

Can we think of chatbot conversations that approximate the King and his messenger like above. Conversations that are guided more by heuristic principles than algorithmic models. From Stackoverflow, this is high upvoted answer to differences between heurisitcs and algorithms.  Below is Kriss‘s explaination: 

An algorithm is the description of an automated solution to a problem. What the algorithm does is precisely defined. The solution could or could not be the best possible one but you know from the start what kind of result you will get. You implement the algorithm using some programming language to get (a part of) a program. Now, some problems are hard and you may not be able to get an acceptable solution in an acceptable time. In such cases you often can get a not too bad solution much faster, by applying some arbitrary choices (educated guesses): that’s a heuristic. A heuristic is still a kind of an algorithm, but one that will not explore all possible states of the problem, or will begin by exploring the most likely ones.

Irrespective of whether machine learning such as the reinforcement learning model (see image below) can be applied to build a ‘guess-as-you-go-chat’ or some other, what matters is why? 

source: KDnuggets

But why bother with guessing, I mean heuristic or 80:20 approaches that may make the chatbot fall on its face! (< any emoji to represent that?) 

In ‘Models of Ecological Rationality: The Recognition Heuristic‘ the authors Daniel G. Goldstein and Gerd Gigerenzer, from Max Planck Institute for Human Development suggests that a ‘Fast and Frugal‘ approach is one efficient method available. Can this guide the design of a chatbot?

From their paper, “One view of heuristics is that they are imperfect versions of optimal statistical procedures considered too complicated for ordinary minds to carry out. In contrast, the authors consider heuristics to be adaptive strategies that evolved in tandem with fundamental psychological mechanisms. The recognition heuristic, arguably the most frugal of all heuristics, makes inferences from patterns of missing knowledge. This heuristic exploits a fundamental adaptation of many organisms: the vast, sensitive, and reliable capacity for recognition. The authors specify the conditions under which the recognition heuristic is successful and when it leads to the counterintuitive less-is-more effect in which less knowledge is better than more for making accurate inferences.

What this would do, in addition to the mundane, repetitive, well-defined, established routines that chatbots can address decently today, is also to add that variety, that ‘masala‘ to the curry; or currying up a conversation with a human! Just more breadth.

To give Dr.Hook’s popular song a twist – take the pussy cat and turn it to a tiger; wild, in the jungle from the zoo.

Dr.Hook – Jungle to the Zoo – 
“The tiger, tiger, they’ll clip your claws, cut your hair, make a pussy cat
 out of you Its one step from the zoo to the jungle.” (edited)

Chatbot, why be anything but wild 🙂

Or as Luciana, the unmarried lady, so full of advice, says in Shakespeare’s Comedy of Errors:

“She never reprehended him but mildly,
When he demean’d himself rough, rude and wildly.
Why bear you these rebukes and answer not?”

Leave a Reply

Your email address will not be published. Required fields are marked *