§ November 16th, 2013 § Filed under AI § No Comments
At MIT’s Artificial Intelligence Lab, a robot baby’s touch points toward consciousness and maybe a sense of soul. On hand is the lab’s resident theologian, Dr. Anne Foerst, to help explore what we can and cannot know about ourselves and the machines that one day may keep us as pets.
“You’d better come look at Cog!” said Matt Williamson, so Anne Foerst, the resident theologian of MIT’s Artificial Intelligence (AI) lab, left her computer and raced down the corridor to see what had made Williamson, an expert on building robot arms, so excited.
At the AI Lab, one can never sure what to expect. Downstairs, in the leg lab, they’ve built stork-like robots that, contrary to conventional wisdom, could run and leap long before they could walk. Along another corridor is Kismet, a remarkably cute robot head with luxuriously long eyelashes that actually does what Furbies only pretend to do: Kismet “learns” from reading human facial expression and reacts “emotionally.” Next door, in the Media Lab, there’s a computer “rabbi” that seems to respond to human problems with the empathy of a true rabbi. But Cog is special: A baby boy — albeit oversized and not yet bouncing — who can make eye contact with humans, play with them, toss a ball, and bobble a Slinky back and forth.
What’s Cog’s secret? Unlike classic AI robots, in which a single computer brain runs separate mechanical systems, Cog is built with “Embodied” AI, meaning that every joint is an independent thinking machine. Each joint in Cog is, by itself, fairly simple, designed to interact in simple ways with the joints around it. But from the series of simple joints the complexity within Cog grows enormously fast. More importantly, each joint in Cog is also “imbedded” in the outside world. In other words, each joint not only interacts with other joints but takes cues directly from the chaos of life. That makes the robot’s behavior unpredictable even to its creators. In short, Cog is designed to interact with the world and learn in the same ways that human babies do. And Cog works, at least a lot better than any conventional robot. That’s what makes Cog so fascinating and so scary.
This particular day with Anne Foerst, Cog would prove to be even more.
Foerst, 33, is the kind of real life character that Michael Crichton might make up on one of his best days. She’s a German theologian (Ph.D., Ruhr-University Bochum) as well as an MIT-level computer whiz. Her position at the lab is part of a seemingly unlikely collaboration between two hugely different venues of higher thinking: Harvard’s comfortingly neo-Gothic Divinity School and MIT’s harshly cubical AI Lab.
One of her colleagues is Harvey Cox, the renowned Professor of Religion at Harvard (and an advisor to this magazine). Cox understands humans as distinct from the rest of creation because of our relationship with God, because of our cognitive abilities, and because of our capability for ethical judgment.
Another colleague is Rodney Brooks, Director of the AI Lab. Brooks feels comfortable understanding humans as “meat machines.” While Brooks doesn’t treat other people like machines, he is confident that AI researchers will eventually build robots that are as conscious and soulful as we think we are — as well as a whole lot smarter. A joke around the lab is that some future generation of Furbies will keep us as pets.
Foerst’s training and position make her an ideal explorer to be smack in the middle of some of the largest questions ever to be explored by science. Her own goal is to get a better handle on what science can and cannot tell us about who we are.
When Foerst came to the lab four years ago, one project she expected to work on was skin and touch for Cog. Why? As both a theologian and a computer scientist, she is “fascinated by the ambiguities of skin, one of the largest organs of the body and a source of sensuality and separation, pleasure and pain. Babies shrivel and even die for lack of caresses. Through touch, babies learn both about their own embodiment and the outside world.” Foerst explains that when you touch a baby’s open palm, it clenches its fist, and when you touch the outside of the baby’s hand, the hand withdraws. These simple actions, “hardwired” into the baby’s brain, allow the baby through trial and error to begin understand both itself and the world around it. So one of her goals at the lab was to help recreate such a learning system in Cog. But, after four years, the project had gotten nowhere. Until that day.
What had happened was this: To help build Cog’s arm coordination, Matt Williamson put “touch” sensors on the robot’s belly — in large part just to give Cog’s hand something to aim for. But Foerst didn’t know that (and even Williamson got a lot more than he expected). So Cog would touch the sensors, a programmed activity. But the appearance was that Cog was exploring his body not as a conscious brain activity or as a visual activity but as an embodied activity — just the way a newborn does.
Says Foerst, “We know that the behavior of human babies is far more primitive that what we project into them. Here the same thing happened. Cog’s programmed activity appeared extremely anthropomorphic. Cog looked so alive! The experience was for me unnerving.”
It turns out, says Foerst, that virtually everyone involved with Cog has experienced similar moments in which the robot did something breathtakingly lifelike — moments that have had the character of real mystery until a closer look at the interactions between the computers processors has provided a scientific explanation for the seemingly “human” behavior. As robots like Cog get more complex, such mysteries will come more often and become more difficult to understand.
But even now, says Foerst, these moments of apparent mystery in Cog’s behavior raise enormous theological questions: Is our respect for human intelligence only caused by our non-understanding of the phenomenon? Do we assign dignity to humans only because we are too complicated to analyze completely? If and when we come to understand ourselves fully, will we lose our self- respect? And, ultimately, if a “conscious” and even “soulful” robot is around the corner, what is a religious person (in her own case, a Lutheran) to believe?
Before we can begin to truly understand such questions, let alone to begin to answer them, the first step, she says, is to take a enormous (and for this article, very brief) step backward.
Sorry, ancient philosophy again…
Since the time of the ancient Greeks, philosophers like Plato have distinguished between two fundamentally different, yet equally important “speech acts:” Logos and Mythos. Logos relates to empirical data and to physical reality. In other words, logos answers the “how” questions. The “how” of anything can be discussed even by widely different groups of people and, eventually, one position will be proven right or wrong by empirical evidence.
Mythos, on the other hand, relates to our interpretation of reality. Mythos answers the “why” questions, and such questions are always answered with reference to a larger symbolic narrative that cannot be verified by scientific evidence. Unlike the universal logos, mythos is personal and cultural. Our myths are a mere fiction or even a lie to those who do not share them. So, Plato would say that Christianity falls into in the realm of mythos, and so does Buddhism, and so do all the new techno-theologies currently emerging from AI and the Internet.
The Greeks understood logos and mythos to be separate and distinct — and that humans need both. They also under stood them to be interrelated. Our mythos shapes the way we perceive the empirical world and, at the same time, our insights into the empirical world gradually change our current mythos. If our mythos stops being convincing, it is because a new narrative has emerged to replace it.
That, explains Foerst, was roughly the way the Greeks separated science from faith, and it held up pretty well until Enlightenment philosophers set about freeing the world from the tyranny of superstition. Even questions of ultimate meaning, they argued, should boil down to hard, empirical evidence — pure logos. Certainly, theologians have not been not immune to the attraction of pure logos. On the contrary, they have often led the charge by trying to use science to prove the existence of God.
But the need for myths cannot be so easily obliterated. So, says Foerst, what has happened, instead, is that our classic religious myths have largely replaced by a series of myths created by science. Right now, technologies have become for many people the new way of expressing myth. The leading edge of this new thought are the various sects of “techno-pagans,” who see computers as models of how human minds work and the Internet as the future of our consciousness.
What’s the point of this all-too-quick history? Seeing the computer as a metaphor for the working of our brains is a very powerful. We all do it w hen we speak of our brains being “hardwired” or “programmed.” But to assume, for example, that Cog’s version of consciousness or the Internet’s have any real meaning for humans is actually a leap of faith — pure mythos. We do not know and perhaps can never know how much these replicas we create actually tell us about who we are. We humans must recognize this leap of faith and be very careful about of how much power we give robots or any other technology over our own self understanding.
That doesn’t mean religious people can’t learn from AI research or that AI researchers can’t learn from our wisdom traditions. The dialogue is important for both sides. But we have to recognize that it takes place in the mythos realm. It’s an interreligious dialogue, and can only be undertaken gently, with open hearts, and good intentions. Demonstrating the importance of this gentle dialogue is what Anne Foerst is trying to accomplish in her bridge between the AI Lab and the Divinity School. And, as it turns out, that’s one reason she’s been pushing all these years to get skin on robots. Why? Because robot skin may be the genesis of a “conscious” robot, and that robot in turn may reveal powerful new insights into the wisdom of Genesis.
Rabbinical AI Theology 101: Genesis 3:21
Once again, we’ll have to trample roughshod over some beautifully complex thought, but one fairly classic reading of the story of The Fall deals with the ambiguity of knowledge. Adam and Eve have the power to make decisions but lack the perfect knowledge of God, so they make a bad decision: They eat from the Tree, get evicted from the Garden, and are saddled with pain and death. So a great human quest, at least from this interpretation, is to seek complete knowledge. If we could just know enough, we wouldn’t do harmful things, and we could live forever with God.
This fairly classic reading dovetails with the thrust of classic AI. If we could build a powerful enough computer or plug into a large enough Internet, we could know everything for the good of everyone, and maybe even make ourselves immortal. But what’s fascinating is that, while such thinking has created computers that can beat even the best humans at chess, it has so far made lousy robots. Sifting more information at ever higher speeds and in ever larger nets is apparently not what human life or consciousness is about.
As Foerst points out, both this classic reading of Genesis and classic AI are examples of mythos that don’t hold up very well given the current logos. So part of her work has been to focus on a much richer interpretation of Genesis proposed by a fourteenth century rabbi, Abraham Ibn Ezra, and brought to her attention by the modern theologian Norbert Samuelson. The interpretation is rooted in a seemingly innocuous passage: “And God JHWH made garments of skin for the man and his wife and clothed them.” Genesis 3:21.
Meditating on this passage, the rabbi came to believe that God would neither kill animals in the Garden of Eden nor sew clothes for Adam and Eve. To the rabbi, both the killing and the sewing seemed ungodlike activities, so he proposed that what really happened is that God gave Adam and Eve human skin. Thus, the story of the Fall illustrates not just estrangement from God because our knowledge is incomplete but because our knowledge is experienced through our bodies and is, therefore, always ambiguous.
Sexuality, the rabbi noted, symbolizes this ambiguity: For example, the relationship between Adam and Eve in the Garden is never described as an erotic one. However, the term used to describe Eve’s desire for the fruits from the Tree of Knowledge is explicitly sexual. So sexual desire is first aroused in the context of knowledge. Another example is the Hebrew word jada, which means both “to recognize” and “to sleep together,” and hints toward the idea that all human knowledge is embodied. Or to put it in Foerst’s mythos/logos terms, in our most embodied human activity, making love, the logos is that we engage in an act of mutual satisfaction and/ procreation. The mythos is that through our bodies we come to truly know each other and to know God.
Embodied AI now takes the rabbi’s idea of embodied knowledge seriously. With robots like Cog and Kismet this ambiguity of embodied knowledge is being built right in. Says Foerst, as more skin is spread over them, the robots will likely seem more like our own children, and provide more of those breathless moments in which we seem to glimpse who we are and where we came from.
The logos is that future robots will likely be smarter, more capable, and perhaps even more sensitive and sensible than we are. The mythos is what we make of it. We can believe that we are demeaned by machines that seem to replicate us all too well. But we can also see our creations as reminders that, while there are things we cannot know, the most likely path to wisdom is not to become more disembodied like computers but to become more embodied, more human.
Oh, that Jill. she has a special talent for making customers feel all warm and cozy. With her in-depth knowledge, thoughtful questions and almost uncanny ability to predict what kind of laptop a shopper will want, Jill, a service representative on cozone.com Inc.’s e-commerce site, often gets the undying gratitude of online customers tied up like a pink valentine.
In fact, recently Jill received this customer love note:
Your overview about notebooks is AWESOME. I realized over the holidays that I need to buy a new notebook, but felt overwhelmed at the task. Your notes give me confidence that I can figure out and find what I need. Thanks!”
You’d think all this praise would be enough to make Jill blush. And it probably would if Jill were human. But she’s not. Jill is a virtual shopping assistant.
Her face does happen to belong to a member of cozone.com’s ad agency, but that is as lifelike as Jill gets. Behind the pretty face is actually a heady stew of predictive modeling, demographic and psychographic statistics, and Bayesian pattern recognition algorithms.
All the number-crunching of ones and zeros add up to Jill being smart enough to act like any small-town salesperson who would get to know her customers and offer them recommendations. It may be cold statistics churning behind that smile, but the only thing that the online customer experiences is Jill in her get-to-know- you mode as she asks questions such as, “Do you travel a lot?” or, “Will a child be using this computer?” It all adds up to Jill coming across as, well, a Web site visitor’s best friend.
That may seem like a touchy-feely goal for online transactions. But establishing that kind of deep connection with customers is exactly what many next-generation e-businesses are after. The problem is that it has been difficult to establish a level of trust between a consumer or business partner when all they have to relate to is a Web site built on CRM (customer relationship management) applications that flip through cold, hard data in the background. To humanize the interaction, enterprises are giving their Web sites face lifts-or more accurately, face consolidations.
The Jill persona is just the tip of the iceberg. Behind that bright, friendly image, savvy e-businesses are integrating customer information stored in CRM and ERP (enterprise resource planning) back-end applications. But building a Web site with personality isn’t easy. It requires stitching together multiple tools such as Silknet Software Inc.’s Trusted Advisor-the technology behind Jill-as well as intelligent search engines, e- mail management, Web chat and call-back buttons to give the site its own charm. Because there is no one software suite that does it all, it usually falls to the IT department to write the code that brings it together.
And as companies such as Sun Microsystems Inc., DaimlerChrysler Corp., Proflowers.com and Porsche Cars of North America Inc. work on creating a single corporate identity for customers and partners to relate to, they’re running into another challenge: persuading division managers-who are sometimes reluctant to relinquish control of customer information-to buy in to the idea.
Ultimately, however, creating a friendly corporate persona on the front end and pooling data on the back end means customers avoid having to deal with a corporation’s separate business divisions, which can all have distinct Web personalities, log-ins and passwords. They get more self-serve options and quicker answers, and e-businesses get tighter customer relationships and lower support costs.
Experts say it’s crucial for businesses to determine what technology is needed to put those relationships together now, before competition eats their lunch. “I do believe that although we’re on the leading edge now, and [some of this technology] is rudimentary, a year from now things will be significantly different,” said Peggy Menconi, an analyst with AMR Research Inc., in Boston. “Personalization will be past making the screen look the way you want it to. You’ll see more and more of these virtual assistants.”
Some companies, such as DaimlerChrysler, are working toward such intimate customer relationships. The company is installing online technologies based on Ask Jeeves, Ask Jeeves Inc.’s natural language search technology. The tool will be customized into a corporate persona similar to Dell Computer Corp.’s “Ask Dudley,” which allows users to type technical service and support questions in using plain English to receive instant answers.
It’s all about creating a customer relationship based on trust. “It’s a relationship with the Web,” said AMR’s Menconi. “It doesn’t surprise me that [Jill] gets mail. Rather than being a cold search engine where you put in Boolean logic, she asks questions a real person would ask you.”
Just another pretty, virtual face?
Proflowers.com, an online florist, is now in the process of evaluating Silknet’s Trusted Advisor-a set of technologies and services that can be used to produce a virtual online persona. Jonathan Sills, vice president of strategy and product development at Proflowers.com, in San Diego, said a virtual salesperson makes sense, particularly in the flower business, where a purchase isn’t just a transaction-it’s an intimate experience. “We ship an emotion, not a flower,” Sills said. “Every element of our customer experience should reflect that touchy-feeliness.”
A key part of creating a Web site with personality is doing the research it takes to imbue virtual sales personas with the kind of expertise that will keep customers coming back. An example of that research is Silknet’s Trusted Advisor division’s visits to CompUSA Inc.’s stores, where researchers eavesdropped on exchanges among experienced salespeople and customers. The result was Jill, a persona they developed specifically for cozone, which is CompUSA’s Internet spinoff. For Proflowers.com, the research will involve visiting florists to do similar research.
Based on technology developed through research at MIT’s Sloan School of Management, Trusted Advisor uses mathematical curves that assign value to shopper responses, such as whether the site visitor has children and how the visitor self-ranks his or her knowledge. An online persona-aka “bot”-such as Jill then presents products tailored to a customer’s needs, along with third-party reviews and articles.
So do sites with these personas actually manage to sell all that much more than a straight configurator or search engine-enabled site? Jill’s creators say yes.
“We’ve seen healthy conversion rates” in converting browsers to buyers, said David Fowler, vice president of marketing for Silknet. “It’s doubled in Version 2.2 of Jill.”
But comprehensive, next-generation CRM has been a tough row to hoe for many companies. Despite the current consolidation trend among CRM vendors seeking to gain more product functionality, no CRM suite today can deliver both the front-end virtual persona and integrated customer information on the back end.
“There really is a lot of concern about whether you go with a point solution that manages one aspect of CRM, such as something that deals with e-mail management really well, or do you go with one that covers the whole spectrum,” said Proflowers.com’s Sills. “I haven’t seen one out there that can do everything we need to do with customer relationships at the same level of proficiency as the point solutions.”
That’s meant that Proflowers.com has had to write custom code to tie together its CRM environment, which includes Kana Communications Inc.’s Kana Customer Messaging System e-mail management tool as well as LivePerson Inc. technology, which enables “call me now” requests to be typed online alerting a company representative to phone the customer back (see story, below).
Proflowers.com also uses Personify Inc.’s Personify Essentials, a system that links visitor or behavioral data with tracked customer data, such as items purchased, amount of purchase, length of time since previous purchase and items viewed.
Tinkering with the n-tier
At DaimlerChrysler, getting cozy with customers has meant that Mike Mortan, senior manager of interactive communications, has had to get his hands dirty tinkering with every application to retrofit them into state-of-the-art n-tier architecture, a setup that allows Web site user interfaces to be changed without having to fiddle with a long string of code.
Standard site architectures have coding all the way down into the database; when one word in the interface is changed, the code has to be reworked back to the database. N-tier architecture separates codes into levels, each of which can be modified separately. What this means, Mortan said, is content can be changed quicker and is therefore more nimble at responding to customer needs.
The outcome of the tinkering will be a Web site that will not only greet visitors by name but will get downright chummy, asking, for example, “Hey, how’s your Jeep running? You know, you’re due for a tuneup. You want me to set up an appointment now?” Mortan said this ability to schedule service online will be launched later this year, even though it’s not something customers want yet-but Mortan says they soon will.
“As we know, things change quickly on the Web,” pointed out Mortan, in Auburn Hills, Mich. “As we move to the future, a big part of personalizing customer experience on the Web will be knowing [customers] well and knowing when they’ll be ready to shop again, plus their buying habits and shopping and lifestyle preferences. We’ll be able to push to a consumer relevant info they’re interested in, as opposed to making a Web site that’s meant for the masses.”
DaimlerChrysler is also plugging in the friendliness fix of a natural language search engine. The company’s Ask Chrysler is a tool that’s based on Ask Jeeves’ Ask Jeeves Corporate Service. Why not just put in a simple search engine? They just don’t have any personality, Mortan said.
“This [involved] humans sitting down and looking at the Web site as a potential consumer and really writing answers and pointing the questions to the actual Web site where the answer would be found,” Mortan said. “[Where] other products use fuzzy logic to do that … we wanted the ability to sit down as if you had another person on the other end of that scene and open up dialogue through natural language.”
A virtual foot in a virtual mouth
Despite the lure of this type of customer self-service, e- businesses should be aware that they can easily backfire and blow up the site’s credibility. That’s because such transactions require data in ERP and CRM systems to be available so that online visitors actually touch the data in enterprises’ back ends. Unfortunately, many businesses aren’t at liberty to tweak long-standing back-end systems at will.
That’s the case at Porsche Cars North America, which is piloting a relational vehicle/customer database that would provide transparency from a browser into the AS/400 back end to access information such as a car’s service history and a customer’s lifestyle, including how many vehicles a customer has and even how many children.
What hamstrings an effort like that is that the legacy systems- including the half-dozen databases through which information flows sluggishly-can’t be changed, since they’ve been adopted by Dr. Ing. h. c. F. Porsche AG as worldwide standards, according to John Jacobs, manager of Porsche Cars North America’s dealer and field systems, in Atlanta.
“There’s a lot of one-to-one relationship marketing aspects we can leverage with a CRM initiative,” Jacobs said. “[For example,] we could provide service reminders, [set up] service appointments. … [But] we can’t touch that legacy code and modify it for our own use because it upsets the apple cart for the international implementation.”
To get that information to flow more freely, at least for dealers, Porsche has licensed Jacada Ltd.’s Jacada for Java, a tool that applies graphical interfaces to huge databases of information and ERP applications, eliminating the need to rewrite code or move data off the mainframes.
But even after businesses figure out how to hook up legacy systems for next-generation CRM, they still have to make sure their back-end data is squeaky clean, experts say. After all, even Jill gets the occasional less-than-charming e-mail, such as one from a customer who found her technical knowledge lacking. “Several things can make [these personas] annoying,” AMR’s Menconi said. “[Companies] put up bad info, or they don’t update it, so they have old info that’s no longer relevant.”
A company such as Sun knows all about exposing its problems publicly. Like many large companies, Sun initially lurched onto the Web with all its divisions dangling. Fixing that situation so that customers could access one Sun and get what they need without getting shunted from division to division is the reason that Sun created its eSun initiative a year ago.
Uniting disparate divisions will spare Sun’s customers pain and save the company money. For example, an order status check costs about $20 when a customer calls Sun, but just pennies when done by a customer online, according to Bobbi Burns, director of Web commerce at Sun. First, however, Burns had to be able to present customers with information from multiple back-end systems. And that meant she had to fight to get divisions’ upper management to let go of their online fiefdoms. “It was a nightmare,” said Burns, in Chelmsford, Mass. “We were asking divisions that were traditionally very independent and used to making their own choices to do something new. It’s perceived as a threat and a taking away of power.”
It took the eSun initiative six months, even with upper management’s support, to get buy-in to the new deal. Now the challenge is to present Jill-style customer recognition in the world of business-to-business, which is far more complex than business-to-consumer relationships.
For example, presenting customized product catalogs so that customers don’t buy products that would be incompatible with their networks can involve five levels of employee access and buying authority. A service contract may be relevant only to a subset at a company, and information may be off-limits to some employees.
“Having multiple customer management systems gets difficult,” Burns admitted. “We have to make sure we can tie systems together with common identifiers and not have to replace them all at once.”
So why go through the angst? Because customers won’t be putting up with online enterprise schizophrenia much longer. “Our customers know what they want and don’t want to spend time calling someone,” Burns said. “They want access right away … and they want to work with one Sun.”
So where should businesses start to give their online personalities some charm? AMR’s Menconi said the first thing to do is to look at a site from the customer’s point of view and design CRM features accordingly. That means letting customers do whatever they want to do quickly and easily-forget the sticky site syndrome; get them out quickly and only upsell subtlely.
Next, make sure you clean up before you open up. Dirty data is just an embarrassment. Last, don’t wait, because “there’s too much at stake,” Menconi said.
After all, you don’t want to catch Jill winning the hearts of customers on a competitor’s site.
Don’t let them drop before they shop
Consumers want self-service on e-commerce sites that allow them to complete as much as they can, as fast as they can. But businesses have to stitch together a crazy quilt of applications to cover all their visitors’ needs, since no one suite offers all the coddling required for a good customer relationship. Here are some of the functions that will satisfy CRM needs:
Research products: There are many solutions to choose from, but some businesses are using natural-language technology to make sure Internet newbies don’t get scared off by Boolean logic. *(products include: Ask Jeeves, Edify and iPlanet)
Create order: Let customers view products before dumping them in shopping carts, or your site will be littered with abandoned carts. *(products include: Calico, FirePond and Trilogy)
Search solutions: Smart technology takes note of what a visitor is researching and will then use predictive modeling to determine what they might buy. *(products include: Ventix, Silknet and Inference)
Check order status: Link-ups to delivery channels such as FedEx are de rigueur. *(products include: netDialog and Silknet)
Dispatch engineers: The ability to call for service help online is rare but will soon be a must-have feature. *(products include: IET, eDispatch.com and PointServe)
Online buddies vs. big brothers
The Catch-22 in today’s e-economy is that on one hand, to build a deep and lasting relationship with customers or partners, you need to know as much about them as possible. On the other hand, many people are fed up with their personal information being bandied about, since it often results in an e-mail box full of spam. Here is what e-businesses need to collect and analyze to turn the Web site into a visitor’s trusted assistant:
Basic profile. Name, e-mail address, billing and shipping addresses, phone number(s), whether they’re part of a household (for business-to-consumer interactions) or employees of a business (for business-to-business).
Preferred method of communicating. Do visitors prefer communicating via phone, e-mail, Web chat or face to face? Do they prefer different channels at different points in their transactions?
Quality of interactions. If Web site visitors are going away frustrated, you better know about it, and you also better know when and why it’s happening in a transaction.