While most of the blogosphere is buzz with Twittering, Twining, and Facebooking, a groups of companies are silently building the next generation of intelligent systems, called agents, that act on your intentions to fulfill your (simple) desires in life. These agents or ‘Sevaks ‘, as I like to call them, literally means ‘one who does a service’.
MIT’s Technology Review, in its 10 Emerging Technologies 2009 rated Intelligent Software Assistant as one of the emerging technologies of 2009.
Search is the gateway to the Internet for most people; for many of us, it has become second nature to distill a task into a set of keywords that will lead to the required tools and information. But Adam Cheyer, cofounder of Silicon Valley startup Siri, envisions a new way for people to interact with the services available on the Internet: a “do engine” rather than a search engine. Siri is working on virtual personal-assistant software, which would help users complete tasks rather than just collect information
These ‘do engines’ are not new. In fact, Artificial Intelligence (AI) researchers have talked about and built such limited, task-specific, intelligent software entities for decades now. In fact, the first such Sevak or Intelligent Software Assistant was built to engage in a conversation with its user – like a therapist. Eliza, the first Sevak, was built in 1966 by Joseph Weizenbaum with just 200 lines of computer code. Things have moved a long way from these humble beginnings, but the task of building ‘useful’ Sevaks or ‘doers’ as opposed to just agents that ‘search’ has kept the attention of many researchers and has captured the attention of a few VCs as well.
More recently, such Intelligent Software Assistants are being commercially deployed is Contact Centers. Botego, Virtuoz, MyCyberTwin, NextIT, SitePal, Soliloquy, and Intellichat are some of the firms that specialize in developing voice or chat-based contact center ‘virtual customer service agents’. They are being used by a number of companies, including PayPal, eBay, Buy.com, Continental Airlines, AMP, and National Australia Bank – just to name a few. While comparing these Agents and their applications is a worthwhile (and sometime amusing) exercise, I will reserve that for a future post. Here I would like to expand on five key characteristics that such Sevaks should exhibit.
-
Social – Being able to engage in a conversation with a user that is beyond the immediate task is a critical characteristic of Sevaks. Customers are used to engaging in a conversation beyond their immediate reason for contacting the company. Failing to design an agent that can engage in such conversations can easily expose the agent to abuse and mistrust. For example, one of the sites allows the user to have a conversation with the cyber twin of Paris Hilton. Even a simple question, such as ‘I thought you were a Hotel in Paris?’ results in a incongruent answer, ‘Paris looks like a beautiful city. I’d love to visit the tower’.
-
Purposeful - Sevaks or agents need to be pro-active or purposeful. They should explicitly ask for the user’s goal or infer it, before providing an appropriate response. Failing to have this characteristic, will result in ‘entertaining’ but useless applications from a commercial perspective.
-
Autonomous – These agents must be autonomous or require no human assistance behind the scenes to resolve customer issues. While it might be acceptable to handle Level 1 questions and pass the customer to a ‘Live Agent’ for Level 2 questions, requiring human monitoring of all agent conversations will increase overhead and costs.
-
Rational – The answers provided by agents must follow a logical sequence and be acceptable as adequate explanation for why they answered the way they did. While humans might be irrational when they make decisions, having a Sevak that provides irrational answers will not be conducive to building ‘trust’ that is an essential attribute for success in this domain.
-
Emotional – Finally, having a Sevak emotionally connect with the user or empathize with the users’ concerns and issues will be critical. While a virtual customer service agent might be able to take all the abuse from human customers, being emotionless in its response will be counter-productive as well.
SPARE - is a minimal set of five characteristics that Sevaks in contact centers must exhibit. How does one build such intelligent software agents and how do some of the agents mentioned in this blog stack up against there characteristics will be explored in future blogs. If you are interested in exploring more (and having some fun) why don’t you visit my CyberTwin here or even go to the original Eliza and see how far the technology has evolved (or not!!) since 1966.
Tags: AMP, Autonomous, Botego, contact center, Continental, eBay, Eliza, emerging technologies, Emotional, Intellichat, intelligent agents, intelligent software assistant, MyCyberTwin, National Australia Bank, NextIT, PayPal, Purposeful, Rational, Sevak, SitePal, Social, Soliloquy, Virtuoz
[...] Original post by Connections [...]
Continental Airlines has not implemented their virtual customer service agent yet, however Alaska Airlines has had their Virtual Travel Expert available on their website for over a year now.
Good summary Anand. I like the SPARE characteristics. Obviously the most difficult to achieve will be (appropriate) Emotion. I think that is a long way off.
In fact, generic intelligent virtual assistants are also a long way off. Providing AI-based virtual support by assembling responses on the fly requires a significant effort in identifying and programming for all possible contexts. As a result, most AI-based Virtual Assistants handle only a small set of tasks, even within the customer service setting. Furthermore, expanding the range of tasks is limited by the availability of specialist programmers and the knowledge that they can develop and codify.
A way to bridge this gap is for intelligent virtual assistants to call the large number of less intelligent helpers (wizards) that exist for a large range of activities, and to act as an intelligent interface.
As a Knowledge Management company, we have developed several wizard writers, including gStepOne, a free version, that taps into the existing body of “doing” knowledge from billions of web pages (via Google) or directly from people in business, government, clubs, community organisations or individuals. See the .
People can create wizards to help deliver a service, manage a loan, sell a house, hire staff, conduct an event, adopt a child, take out insurance, apply for a grant – any task where people need an Assistant to help them do things like an expert, without the expert being there.
The logical next step is for intelligent virtual assistants to call and interpret these wizards for their human clients.
gStepOne is free and available from . Please check it out! Virtual Assistants will certainly come to pass, but AI-based Assistants, though they play a key role, are only part of a .
Good summary Anand. I like the SPARE characteristics. Obviously the most difficult to achieve will be (appropriate) Emotion. I think that is a long way off.
In fact, generic intelligent virtual assistants are also a long way off. Providing AI-based virtual support by assembling responses on the fly requires a significant effort in identifying and programming for all possible contexts. As a result, most AI-based Virtual Assistants handle only a small set of tasks, even within the customer service setting. Furthermore, expanding the range of tasks is limited by the availability of specialist programmers and the knowledge that they can develop and codify.
A way to bridge this gap is for intelligent virtual assistants to call the large number of less intelligent helpers (wizards) that exist for a large range of activities, and to act as an intelligent interface.
As a Knowledge Management company, we have developed several wizard writers, including gStepOne, a free version, that taps into the existing body of “doing” knowledge from billions of web pages (via Google) or directly from people in business, government, clubs, community organisations or individuals. See the press release.
People can create wizards to help deliver a service, manage a loan, sell a house, hire staff, conduct an event, adopt a child, take out insurance, apply for a grant – any task where people need an Assistant to help them do things like an expert, without the expert being there.
The logical next step is for intelligent virtual assistants to call and interpret these wizards for their human clients.
gStepOne is free and available from http://www.gstepone.com. Please check it out! Virtual Assistants will certainly come to pass, but AI-based Assistants, though they play a key role, are only part of a bigger picture.
We should talk more about this and give it a try!
Just because you don’t see all of these characteristics in Intelligent Software Assistants yet, does not mean they do not exist.
Our advanced robots freely evince purposeful, autonomous and rational behavior. (Our standard ones are set up so that anyone can build their own AI, in ten minutes flat, with no skills. They do not have the same capabilities as advanced robots.)
As to Social and Emotional behavior:
- We often find that clients do not want their enterprise robots to engage in social or emotional behavior. Some of our robots know exactly what is going on socially and emotionally, and what an appropriate humanized response would be, but are not permitted to make that response for various ethical and legal reasons.
- We have also had large media clients deploy MyCybertwin robots without telling the audience they are robots. 95% of the audience do not work it out, they assume they are talking to a human; and nothing in the interaction changes that assumption. These robots evidently have full SPARE characteristics, but the client has chosen not to disclose their inherent robot-nature. We are also not able to disclose them, so that journalists can go test them; because they belong to our client, not us.
I believe a set of standards around this will evolve in our industry:
- Some people will want every robot they encounter to act like a robot, not engage in human-like behaviour, and be very obviously a robot.
- Others will want the delusion – the “willing suspension of disbelief” to quote Coleridge. The quasi-comfort of talking to a being who cares enough to listen and respond endlessly.
- Others will want a hybrid: Obvious robotness for new robots, but known AI’s (for example, your personal assistant/home organiser/personal banker AI) achieve “friend” status of various levels.
I would love to set up an industry body to start monitoring these matters, and welcome contact via MyCybertwin from anyone skilled in the field who is interested.
Thanks Jen.
Yes – Alaska Airlines has been using a Virtual Travel Expert for some time. It is the kind domain one would think is quite suited for Intelligent Agents. I tend to ask the same questions to my Human Travel Agent all the time – Can you get me a cheaper fare if there are stop-overs? Is the fare cheaper if I fly out of neighboring airports? What if i change my departure by a couple of hours – does it alter the fare? etc. Unfortunately, even my ‘Live Human Expert’ does not offer me the options without me explicitly asking each one of the above. Being able to program my virtual travel agent to ask these questions all the time would be great indeed.
Thanks Liesl for your comments. You make an interesting point about revealing the identity of virtual agents. I agree that companies would like to have the option of deploying their virtual agents across a spectrum of complete anonymity to complete transparency. In addition, one may want to switch certain features/characteristics ON or OFF based on the situation – e.g., emotional filter on/off or identity on/off based on company/user desires. I think a system of rating different virtual agents or even a contest like the ‘Turing Test’ contest for the Loebner Prize (http://www.loebner.net/Prizef/loebner-prize.html) might make sense.
Thanks Greg for your post and the pointer to GStepone. Ability to take a process diagram and convert it into a wizard is really great. This reminds me of the Procedural Reasoning System (PRS) and the family of BDI (Belief Desire Intention) architectures that I used to work with. In PRS you have the ability of defining Hierarchical state charts (or process diagrams) with a conditional trigger. The visual language could be interpreted or compiled and run to exhibit a variety of intelligent behavior. Check out some of my papers in http://citeseer.ist.psu.edu/Agents/BDI/ for more details.
I really liked this post. Can I copy it to my site? Thank you in advance.
Hi, interest post. I’ll write you later about few questions!
The article is ver good. Write please more
Hi! I like your srticle and I would like very much to read some more information on this issue. Will you post some more?
Some of us even don’t realize the importance of this information. What a pity.
[...] Top 5 Characteristics to Look for in Intelligent Software … [...]
Thx for information
This is a great post. We have to have these discussions constantly with our clients to reinforce just how important it is that users “trust” the virtual agent.
You should check out Continental Airlines site now. They have now launched their new virtual expert “Alex”.
Thanks very much for posting such a great article!
http://www.workingforgoogle.com – free guide to making money online from home.
[...] one of my earlier posts on ‘Top 5 Characteristics to Look for In Intelligent Software Agents’ I had mentioned Siri – a silicon-vallye start-up focusing on building virtual software [...]
[...] one of my earlier posts on ‘Top 5 Characteristics to Look for In Intelligent Software Agents’ I had mentioned Siri – a silicon-vallye start-up focusing on building virtual software [...];. All the best!!