As Innovation Lab work across industries, public/private, small and large on challenges ranging from digital product development and business models to organizational redesigns, really understanding the problem is imperative to us. To accomodate this, we’ve set up a chat interface in the bottom right corner on the ilab.dk website with humans (at least that’s what they told us) replying and we are launching a bot-driven conversational agent in a few weeks to provide answers to your requests. To put this application of artificial intelligence into context, here’s an introduction to what is known as conversational agents or chatbots.
In 1990 Hugh Loebner established the “Loebner Prize” of 100.000 USD as well as a solid medal of gold to the group or individual who could produce a chat-bot that was able to convince its partner it was a human. This test is called a Turing-test and was proposed by Alan Turing in 1950 as a test to identify when computers would be on-par with humans. After this point, the computer may be many, many times more intelligent than a human, and we will never know as we could not possibly comprehend what it was doing.
Alas, there is still a long way to go. New developments from Apple, Google, Amazon, IBM and other tech leaders have not shown comprehensive proof of any significant progress. Well, granted, when you gather the brightest heads in data science and add unlimited computational resources to the pool progress is bound to render new “improved” conversational skills or other small iterative improvements.
However, when you look at the tremendous amounts of resources that go into AI these days, in comparison with the results we get out of it, we are in a situation of diminishing returns on investment.
According to the research done at Innovation Lab, there is no true solution to the problem, and we are not a lot closer to passing a Turing-test with a chatbot, than we were 20 years ago.
There are claims (Source : Quantum Mechanics and Artificial Intelligence, Subhash Kak) that AI will not truly work in a computer, as our human intelligence and organic nature of the brain interacts with the quantum realm in order to work. Others suggest (How to Create a Mind: The Secret of Human Thought Revealed, R.Kurzweil) that we may decode the functions of the human brain though real-time high-resolution brain scans and create a digital equal inside a sufficiently powerful computer.
Given that we, for the time being, may need to make do with chatbots that are not truly intelligent, there is a need to identify the areas of operation where chatbots may be of use to humans, but first, let’s have a look at the different flavours of chatbots
There are two main types, closed domain and open domain.
Closed domain bots are used for answering specific questions in relation to a product, service og consultancy. This provides value if you want your customers to get answers on specific questions about your products, e.g. through an automated chat function on your website. It would also work great for ordering pizza or making arrangements for getting a haircut or car-repair. In terms of consultancy, IBM has recently showed off an AI that can act as a medical advisor for people with specific heart conditions and has also made a general (medical) showcases for identifying cancer, and have thereby defined the professional medical AI market, a market that is “blue ocean” at the moment.
Open domain, are the ones we talk about when we want to find a strong candidate to the Loebner prize. These are general intelligences, that will have the combined knowledge of Wikipeidia, will be able to use slang, swear, have humor, be moody at times. These are also the hardest to make. A open domain intelligence may converse on news, daily tasks, family relations, and perhaps answer mails for you, but will initially (the next 10 years) be limited to perform simple tasks. So either, they are closed domain, and nothing more than puppetries guided by complex statistics, or open domain and appear as coherent as 3-year old trying to explain what a prime minister does.
However, let us dig a bit deeper into the status of present day chatbots.
Chatbot applications could be placed along an evolutionary spectrum from bot-aided humans (E.g. back office bots for Q&A in customer service), to human-aided bots (E.g. human curation of product suggestions by bots) to pure artificial intelligence passing the Turing test (E.g. conversations indistinguishable from humans). There are available applications of the first two steps and several claims for the third level but in general the use of chatbots is only slowly gaining mass market appeal despite the 1.4 billion chatbot interactions in the US last year (Source: mobilemarketer.com).
Bots have already found application in various domains but mostly within conversational commerce, productivity and news.
One example is the conversational commerce bots of Sephora and H&M on Kik that drives discovery with simple decision trees. Chatbots allow user-driven two-way conversations that, at least in theory, lead to more engaged and vested users (Source: mobilemarketer.com).
Amy or Andrew, depending on your preference, is the meeting schedule bot of x.ai that recently received a $23M series B investment to take the technology from a closed beta to a commercial product. By Cc’ing Amy/Andrew into emails regarding the nuisance of scheduling or rescheduling meetings, she/he will reply and schedule based on calendar availability (Source: x.ai).
In a different domain, Quartz released a conversational news app that will provide you with the latest or breaking news allowing you to interact to get further information. The news are human-curated by an editorial team and will adapt to your preferences over time (Source: qz.com).
Some news agencies such as Associated Press, Fox and Yahoo are using artificial intelligence to auto-generate stories on business and sports giving the journalist more time to add in-depth perspectives. Bot there’s still some ground left to cover for auto-generated news content (Source: ap.org).
As with most new technologies there’s a tendency among tech pundits to overestimate its impact in the short term and underestimate it on the long term. And chatbots are no exceptions. What companies can do about it is to seek out the lowest hanging fruits or jobs-to-be-done of their users and apply artificial intelligence to it.
As user expectations of a lean and convenient experience continues to rise at an ever-accelerating pace, early chatbot applications have one big advantage: they exist within the mostly used messaging apps. Widely-adopted and highly-used apps such as Facebook Messenger, Kik, WeChat, Line, WhatsApp, Snapchat, Telegram Viber and Skype all have or are planning bot store and/or API-enabled external functionality.
Digital platforms leverage data from user behavior and chatbots will continue this as the conversation with users will allow for more fine-grained profiling and recommendations as the adoption and intelligence grows. The downside of the proliferation of chatbots and the resulting recommendation engines running in the background is an even narrower range of choices presented to the user. The better the chatbot knows you, the better it will match products and services to fulfill your needs but the less “out-of-scope” results you will see. In time, this means that we will see less news and meet less people that don’t fit our profile.Thus, the validation and reinforcement mechanism of most services online inhibits the serendipitous exploration that helps us avoid confirmation bias (Source: nytimes.com).
If you’re interested in building your own bot, there are some things to consider (adapted from intercom.io):
1. DON’T PRETEND TO BE A HUMAN
Bot personality is ok but don’t try to fool users into thinking that your bot is human
2. KEEP IT INCREDIBLY SIMPLE
Delineate subjects and create linear conversation rules to avoid dead ends. Explain boundaries to users.
3. RESPECT THE CHAT MEDIUM
Don’t use the chat medium footprint for all app functionality - follow the conversation.
4. OPTIMISE FOR THE END USER
All eyes on UX - don’t introduce bots where humans do a better job.
5. USE SPARINGLY
Less is more. KISS - keep interactions short & simple.
6. PROVIDE AN ESCAPE HATCH
Bots in the front - humans in the back. Humans should be reachable at all times.
7. USE STRUCTURED INPUT WHEN POSSIBLE
Streamline over complexity. E.g. tailored keyboards for specific users or use cases.
8. EVERYONE SEES THE SAME THING
Keep all data available to improve UX.
So please go ahead and start chatting away - in a very short time we will publish the Innovation Lab chatbot which is currently creating havoc on Twitter + we're building a pizza delivery bot (Don't worry - we won't be making the pizzas).
We're down here -->