Is Anthropomorphic Design a Viable Way of Enhancing Interface Usability?

Chapter Two:  Discussion of Previous and Existing Anthropomorphic Interfaces and the Application of Anthropomorphism in New Technology.

As the previous chapter notes, since the 1950’s people have both described computers as ‘human’ and tried to make them behave accordingly. The idea of being able to tell the computer exactly what you want, without implementing old-fashioned tools, such as the qwerty keyboard, (which has evolved for computer use, rather than being intentionally designed) remains a utopian fantasy for many people.

The notion of a truly anthropomorphic interface that actually works has been ingrained in popular culture since the dawn of science fiction. Computers which ‘talk’ and ‘think’, such as Holly (Red Dwarf) or Deep Thought (The Hitchhikers Guide to the Galaxy) are commonplace in fiction, yet so far our own attempts to realise these ideas have been seriously flawed.

One of the pitfalls of software design is the eagerness of many computer companies to ‘run before they can walk’ and release software that is far from ready and which only does half the job. Market pressures and eagerness to ‘lead the field’ have led software houses to release things which simply don’t work. It is like a modern car manufacturer producing a car that has wooden wheels - it doesn’t matter how pretty it looks, or how well the air conditioning works, if the wheels are made of wood then it is not going to run properly and it will need constant maintenance. Software companies then add insult to injury by attempting to use their shortcomings as a marketing technique to sell ‘improved’ versions – yet like a car with plastic wheels instead of wooden ones, the software is often still incapable of meeting users needs.

If it is feasible that anthropomorphism may be successfully applied then we should work within the limitations of current technology, rather than creating something which says one thing and does another. This would avoid the problems raised by anthropomorphic dissonance - as mentioned in Chapter One (Watt, 1998, cited by Swartz,
2003), which describes the gap between actual and expected behaviour. The frustration experienced by some users when they attempt to use a ‘half built’ humanlike interface will often put them off using this type of technology again. This is apparent in an episode of the television series ‘The Osbournes’ where Ozzy Osbourne attempts to use the speech recognition function on his in-car satellite navigation system. It is Ozzy who ends up getting frustrated and looking stupid, yet it is the technology which does not work.

Historically, most commercially available anthropomorphic interfaces have been regarded as unsuccessful. The eagerness of many computer companies to release ‘half baked’ software may be one of the contributing factors to the way in which they have been received.

2.1 Applications of the Anthropomorphic Interface

In ‘Designing the User Interface’ (Shneiderman & Plaisant, 2004: 486), Shneiderman lists some of the humanlike interface characters which he asserts have failed to capture the public’s imagination. These include what he describes as “failed anthropomorphic bank tellers” such as ‘Tillie the Teller’, ‘Harvey Wallbanker’ and ‘BOB’ (from the Bank of Baltimore) plus a failed “pseudo postal clerk” system (the ‘Postal Buddy’), which was implemented and subsequently discarded by the US Postal Service. Shneiderman claims that the Postal Buddy was rejected by users after incurring costs of over $1 billion. However according to one article (Harris, 1996) the company who made the Postal Buddy (Postal Buddy Corp.) had patented the kiosk/change of address process and the real reason that the postal service pulled the plug was because of “intense bureaucratic rivalries.” This view is supported by Marty Goodman, a spokesman for Postal Buddy Corp:

“Postal Buddy was killed because it was successful and the Postal Service realised that they would be making a lot of money if only the idea had come internally…and if they, rather than private entrepreneurs had the patent on the computerised change of address process…”

A subsequent lawsuit over the cancelled contract cost the US Postal Service $50 million and the failure of the project was officially blamed on failure to produce enough revenue, ‘user unfriendliness’ and technical malfunctions (Harris, 1996).

Little evidence remains to describe the automated bank teller systems that Shneiderman mentions (most companies seem reluctant to document their own failures...). But in an interesting twist, the new Connecticut Bank and Trust Co recently announced plans to resurrect an ATM ‘character’ called ‘Barney’. The original Barney was created in 1976 and was killed off in 1992 after the collapse of the Bank of New England Corp. He was inspired by the ‘Tillie the Teller’ banking character from Atlanta. According to a report in the Hartford Courant (cited by atmmarketplace.com, 2003):

“The Barney character, which had a small round face and bow tie, was once so recognisable in the state that any bank’s ATM came to be known as a ‘Barney machine.”

According to a feature about Barney on the ATM Marketplace website (2003):

“Barney’s bespectacled face will adorn the new bank’s ATMs as the new CBT hopes to capitalise on some of the nostalgia surrounding Barney…one key to the [original] marketing was using Barney’s circular face as a head: the machine was drawn as the body, serving to further “humanise” the ATM. Legs were considered, but the idea was discarded…the initial deployment of 15 Barney machines in the fall of 1976 took root immediately. ”

Although Shneiderman insists that these types of humanlike characters were poorly conceived and generally rejected by users, it is important to note the corporate undercurrents which may also have contributed to their failure (and in some cases, resurrection). It should also be observed that things which make you feel nostalgic are not necessarily good. One could feel nostalgic about the BBC Microcomputer, but you wouldn’t want to type up a 15,000 word report on one.

One area that has been indisputably unsuccessful, is the use of anthropomorphism when applied to Web-based interface agents. It is now quite difficult to locate even one good example which is still in use, and those sites which used to employ them (such as buy.com) have generally eradicated all traces of their existence.

The animated female news reading character Ananova (Figure 2.1) (ananova.com) was “once heavily promoted as the future of web-based news”, (Shneiderman & Plaisant, 2004: 486) If you look at the website Ananova.com now, she is tucked away in the ‘video reports’ section, you need to download Real Player to see her and she doesn’t even open if you use a Firefox browser. The site claims that she will report “the best quirky stories of the day” yet Ananova no longer appears to be updated on a daily basis. Perhaps the company hopes that nobody has noticed – another indication that few people are now using this system. (Since this report was written, the newsreader has been removed from the site and replaced with the following notice: 'Please note: Ananova video is currently under development. Come back soon to check the latest - 23rd April 2006)

Over a slow connection, Ananova’s lips are not synchronised with the sound of her voice and she mispronounces some words - particularly people’s names - as her speech is computer generated. She works, but not as efficiently as a human – she’s a bit like the car with wooden wheels. Why not use rubber and metal instead? In this instance it would probably be more effective to videotape a real woman reading the news, or at least get a human actor to record the spoken words, as they would be more adept at pronunciation and synchronisation. This is a good example of under-developed technology being used to do a job, which currently could be better executed by a human. It’s a bit like taking a taxi to the corner shop. By the time you ring them and wait for them, they find somewhere to park – and you pay the exorbitant fee, you really could have just walked there.

Another area where anthropomorphic interface characters have failed to capture the imagination is in the design of general application interfaces for home or business computing.

In 1987, Apple Computers developed the concept of the ’Knowledge Navigator’. (Fig. 2.2) They produced several videos showcasing the idea, which featured a young male actor playing an anthropomorphic agent.

The system featured excellent text-to-speech and speech recognition capabilities (similar to those of fictional computers) which allowed the user to interact effortlessly and delegate tasks. 

At the time this utopian interface idea was viewed as groundbreaking. As Shneiderman (Shneiderman & Plaisant, 2004: 485) explains:

“Some futurists celebrated this vision, but sceptics scorned the scenario as a deception; most viewers, meanwhile, seemed mildly amused.”

Although the idea of the Knowledge Navigator as an autonomous agent to assist with computerised tasks was innovative, it may also have been influenced by the popular culture of the day. The American television series Knight Rider (first aired in the US in 1982) featured a computerised car, called KITT, which was endowed with a host of mind-boggling skills as well as being proficient in speech recognition and computer generated speech. Wetmore (1999) describes how KITT and his driver, Michael Knight always worked as a team, communicating back and forth and drawing on each other’s knowledge and abilities.

The apparent ease with which we are able to construct fictional anthropomorphic characters seems to have led many designers to believe that they are just as easy to implement in real life. In an early article (Brennan, Laurel & Shneiderman, 1992) Shneiderman describes this pursuit of an ideal humanlike agent as misdirected. He believes that such scenarios are “obstacles in the development of truly powerful and simple tools that will empower users…” (ibid.) and goes on to describe it as the “obstacle of animism”, (after Mumford, 1934, cited by Shneiderman & Plaisant, 2004: 486).

It is easy to pretend that computers can talk, but it is very difficult make it happen - the Knowledge Navigator video itself is a fictional construct, with no actual functioning prototype involved. The interface agent is played by an actor, just like the ‘human’ computers of science fiction film and television.

In 1990, another project at Apple Computer by the Advanced Technology Group, claimed to have pioneered the use of characters in the interface with the ‘Guides’ application (www.abbedon.com, 2005). The project used human actors to represent guides in a North American history database (Figure 2.3). The context of use involved using point of view as a navigational tool. The intention of ‘Guides’ was to engage users emotions and provide a sense of companionship.

Although these design concepts from Apple are significant, not all of their ideas have been adopted. Natural language technology is taking way longer to develop than the computing pioneers of the 1950’s had anticipated. As a result of this, many of the Knowledge Navigator’s features; such as an ability to respond to gesture or to interact within the context of a conversation, have failed to take off. Despite this, it has been credited as the inspiration behind the Apple Newton handheld device (sadly another example of immature technology, released too soon and doomed to commercial failure) as well as other more successful innovations, both by Apple and their competitors.

2.2 Microsoft Interface Characters

Probably the most prominent example of an anthropomorphic character-based interface agent to date is the renowned Microsoft Office Assistant (Figure 2.4)

The Office Assistant was first distributed with the Microsoft Office software suite in 1997. Most people refer to the Assistant as ‘Clippy’ after the default character (from a small selection) which was a stylised humanlike paperclip (actual name “Clippit”).

Clippy was intended to enhance the user interface by offering proactive help and tips and is descended from Microsoft’s earlier attempt at implementing an interface agent, ‘Microsoft Bob’ from the ‘Microsoft Home‘ software suite. 

2.3 Microsoft Bob

‘Bob’ (code named “Utopia”) was a commercial disaster. He was released in 1995, with the intention of replacing the Windows 3.1 and 95 ‘desktop’ with an interface designed mainly for novice users. Lanier (1995) describes Bob as “offensively paternal” and there are a number of factors which are attributed to the program’s demise: Bob needed a lot of power to run and many computers of the day did not meet those minimum requirements. He was not considered useful enough to justify his price tag of $100 - at that time, those who sought something easy to use would simply buy a Macintosh. He was also considered “too cute” for existing PC users and when Windows 95 was released later that year, with the new Windows Explorer interface, it “wiped the floor” with Bob (Toastytech.com, 2005).

Bob was inspired by contemporary research, such as the previously mentioned Computers are Social Actors paradigm (Nass, Steuer & Tauber, 1994), which had just been published. Microsoft were reportedly also influenced by the notion that if they created a popular character, then it could be a success in it’s own right. They had noted that revenues from California Raisins character merchandise exceeded the actual sales of the entire worldwide raisin industry (Cuneo, 1988, cited by Swartz, 2003). They introduced several professionally designed characters such as ‘Rover’ the retriever (Fig. 2.5), to guide users through the ‘virtual room’ interface
(Fig 2.6). Although Bob’s failure was never officially explained, it certainly didn’t deter Microsoft from having ‘another go’ and so ‘Clippy’ was unleashed on an unsuspecting public.

2.4 The Microsoft Office Assistant

At an Office 97 demonstration in Las Vegas in 1996, Mr Vail of Microsoft explained the idea behind the Office Assistant:

“Oftentimes when a user is using an application, they call somebody on the phone and say, ‘Hey, how do you do this?’ What we want to do is put that expertise into the product itself. Using natural language technology, I’d use it like this: I’d type ‘How do I send this to my boss?’ When I hit enter, here, we’re going to use natural language technology to read through the sentence, pull out the key elements and map that against the index and we’ll bring back help topics that are most relevant, based on what we asked here. It’s a great way for users to become empowered and become more productive in the application themselves.” (Office Assistant Demonstration, 1996)

Again, this is an example of an interface character being over-sold. It doesn’t have true natural language capabilities - you can’t have a conversation with it. If you type ‘How do I send this to my boss?’ into the query box of the Office Assistant, it uses a rather sophisticated system to work out the probability of what you are trying to ask, but it might as well be performing a simple search. A number of options are returned – but none of them tell you how to actually ‘send this to your boss’. This is just one of the many factors which have contributed to the seemingly universal condemnation that the Office Assistant has received.

The example character used in the Las Vegas presentation (ibid.) is the ‘Genius’ character (based on Albert Einstein – see Fig. 2.7). Half way through the demonstration, Bill Gates prophetically asks:

“What if I don’t like the Genius?”

What indeed. Despite the failure of Bob and over 25,000 hours of user testing for Office 97, Microsoft had no idea just how much one little character would manage to irritate so many people. What is it about the Office Assistant – and Clippy in particular, that drives people nuts? – does this mean that we shouldn’t use anthropomorphism in interface design – or should we just observe and learn from Microsoft’s mistakes?

The proposed technology behind Clippy was actually rather advanced. Although the system drew on previous expertise from ‘Bob’ (for an example, compare Bob’s ‘Rover’ dog character with the Office Assistant’s ‘Rocky’), a great deal of research by the Adaptive Systems and Interaction team at Microsoft was also intended for use in the program.

2.4.1 The Lumière Project

The Office Assistant uses Bayesian modelling techniques, derived from the in-house Lumière Project (Horvitz, Breese, Heckerman, Hovel & Rommelse, 1998) and elements of Microsoft’s Lumière-inspired Answer Wizard program (previously deployed in Office 95). The Bayesian approach involves the implementation of a formula which updates the probability of an event, based on repeated observation of events.

The Lumière project began in 1993 and involved extensive user testing which applied a ‘Wizard of Oz’ approach. Test subjects worked on a spreadsheet program in one room whilst an expert sat elsewhere, monitoring their progress and offering appropriate help or tips via the user’s computer screen. The experts were a ‘human model’ of what Microsoft were trying to achieve with the Office Assistant. Much of the information gathered from these tests was then used to build an ‘intelligent’ query service for interpreting natural language queries. This technology was applied in the Answer Wizard and subsequently in the Office Assistant, which also monitors user’s behaviour and makes proactive suggestions based on what it sees – for example, if you click too many times in the margin, it assumes you might be trying to type there and advises you on the best course of action.

Although most of Lumière’s attention was focused on building the ‘intelligent’ query service, they also constructed a prototype which deliberated whether or not to disrupt the user when offering proactive help or advice. The team asserted that “such intrusions should be done in a conservative manner, with the express approval of users” (Horvitz, 2005). The prototype employed Bayesian models to control “speculative assistance” and included a ‘volume control’ that could be used to minimise interruptions. The control was always available when advice appeared and allowed users to move a ‘slider’ up (“bother me more”) or down (“bother me less”).

If only this piece of technology had been included in the final version of the Office Assistant, one can speculate that it might have been more successful. An article by John Markoff (2000) describes how Eric Horvitz (head of the Adaptive Systems team):

“…speaks apologetically about the paperclip program…..it’s shortcomings were the result of Microsoft’s failure to implement all of his team’s ideas”

The way in which Clippy continually offers unwelcome help and unsolicited tips or advice is one of the key factors contributing to the way in which it has come to be regarded as a major irritant.

Perhaps the most infamously irritating feature of all, familiar to anyone who has used the Microsoft Office Assistant, is the proactive Letter Wizard. One has merely to type the words: “Dear Sir/Madam” into Microsoft Word 97, and a bubble appears from Clippy’s mouth, (see fig. 2.8) “It looks like you’re writing a letter…”

No matter how many times you tell Clippy you do not want help, he always asks. Cultural differences may also exacerbate the problem. If you read the question to yourself in a happy American accent, it sounds like Clippy might be genuinely trying to help you. In Britain, where there is a general culture of sarcasm and irony, it sounds like Clippy might be gently mocking you: “It looks like you’re trying to write a letter.”

2.4.2 Cultural Differences

Microsoft’s sporadic tendency to overlook cultural differences like these, combined with their occasional accidental insensitivity has reportedly cost them millions of dollars in lost sales (New Scientist Archive, 2004). Although the Office XP multilingual pack provides additional Office Assistant characters for Asian customers (Saeko Sensei and the Monkey King), the company has often failed to recognise the subtle cultural variations of different countries. The study into the effects of computers which apologise (2004), mentioned in Chapter One, noted that the subjects of that study (which took place in Taiwan) may have responded better to an apologetic computer because it is part of their culture to be polite. Jeng-Yi Tzeng, the author of the study is now extending that research to include a cultural comparison of user interaction (personal communication, 2005), to see if other cultures feel as compelled to be courteous to their PC.

It is hard to find documentation regarding the geographical location of Office 97’s 25,000 hours of user testing, though some of it was carried out abroad, it may be assumed that much of it took place in the United States. The fact that the product was tested so extensively and yet Clippy was not identified as an irritant is a cause for some concern – although the Office suite itself continues to be a commercial success. Clippy’s failure may be attributed to factors such as test subjects being self-submitted, or a reluctance by users to criticise the software directly.

2.4.3 A Personal View

I am now going to offer my personal view on the Office Assistant, in order to make a few points. The first time I saw Clippy, in the PC labs at UWE,  I admit I thought he was cool. Lanier (1995) asserts that anthropomorphic agents will “inevitably deliver an overdose of kitsch” – and as I type this report in my pink office, surrounded by plastic flowers, you might get some idea of the type of person that might perceive the Office Assistant as “cool” (and perhaps the motivation for this report). During the course of my studies I repeatedly heard both lecturers and fellow students criticising Clippy and somewhere along the line I decided that he was very uncool. When I began my research for this project I decided to enable Clippy (I am using Office 2000) to find out just how irritating he could be. For the first time I also managed to locate the installation discs so that I could change the characters – after some experimentation I finally decided to use Rocky, the dog character. I now confess that in the five months that I have been using the Office Assistant, I have not found it irritating - I have not found it particularly useful either, but it is certainly a welcome distraction and I love the way that when Rocky barks or whimpers, my own dog gets up and looks out of the window. When discussing the Office Assistant at the project poster session a senior lecturer in the UWE computing department confessed to me that they also had the Office Assistant enabled and set to the little cat character.

Shneiderman (Shneiderman & Plaisant, 2004: 487) asserts that anthropomorphic interfaces may seem cute at first, then silly and ultimately distracting. Perhaps the Microsoft user tests did not investigate how Clippy would be received over a prolonged period of time – and maybe the test subjects did not all get together and criticise Clippy during their coffee break, as peer pressure may have affected the way in which some users view the Office Assistant. People generally enjoy having a pop at Microsoft (whilst continuing to use their software), so Clippy may also be seen as a figurehead for the Microsoft backlash, particularly amongst IT professionals.

2.4.4 Personality

A further possible reason for Clippy’s general rejection is the matter of personality. It’s a talking paperclip. Most users might have a little chat with their dog from time to time, but generally they wouldn’t start asking a paperclip for help or advice. In tests prior to the release of the Office Assistant (which took place in three different countries), Clippy was not identified as the most popular character. He was eventually chosen as the default because he was associated with an office (Nass & Reeves, 1996 cited by Swartz, 2003). CASA theory (Nass, Steuer & Tauber, 1994) suggests that characters that are popular and likeable in the real world will also be more popular and likeable in a computer interface. Issues concerning the use of similar or complimentary personalities, as discussed in Chapter One, may also have affected Clippy’s popularity.

In one study (Rickenberg & Reeves, 2000), the ‘Genius’ character (Fig 2.7) was used to ascertain whether animated characters have enough social presence to make users feel as though they are being monitored. It was found that users did feel more anxious when working on a task, where the Genius was ‘present’ – in the same way that they might if someone entered the room and looked over their shoulder. The fact that Clippy (and some of the other Office Characters) stares directly at you whilst you are working may be another factor affecting the way in which he was received.

2.4.5 Clever Marketing?

Of course, the way in which Microsoft have been undeterred by Bob’s failure and their decision to use Clippy as the default, even though he was not the most popular character could indicate that the Office Assistant was some kind of elaborate marketing technique. Clippy could not be permanently hidden in Office 97 – this ‘feature’ was added in Office 2000. In Office XP, the Assistant was disabled by default. Microsoft asserted that this was because the new version was so easy to use that Clippy was redundant. However, they also used Clippy’s ‘retirement’ as a marketing tool, employing derisive websites which encouraged users to vent their frustrations and “celebrate Clippy’s demise."

2.5 The Future of Anthropomorphism

Apologetic about Clippy, yet undeterred, Horvitz’s team at Microsoft continue to develop anthropomorphic interfaces. In an interview with John Markoff (2000) Horvitz explains how his researchers are:

“….working on a conversational system which they plan to use experimentally in the research group’s reception area .. the system will be able to greet and converse with new visitors. The conversation….will be on par with speaking to a person who is hard of hearing….”

Anyone with a deaf friend or relative may confirm how frustrating it can sometimes be trying to communicate with them. Why not stick a real person in the lobby, rather than working on a system that is going to cause frustration?

Elsewhere, in a step towards the ‘Knight Rider’ model of user interaction, current satellite navigation systems are increasingly being developed to accept speech, yet most still leave a lot to be desired. In a further development, the Apple’s new Mac Mini (a tiny portable system) is being viewed as the “divine gift to in-car computing” (jalopnik.com, 2005), following recent trends in car customisation (inspired by TV shows such as MTV’s ‘Pimp My Ride’) the Mac Mini is being offered by some auto customisation firms as part of an in-car computing package. It fits into most car’s stereo or DIN slot, it supports both voice recognition and voice response, uses bluetooth and has 40 GB of storage for media such as music and DVDs (www.technovelgy.com, 2005). Whether the system will be able to exhibit any of KITT’s other character traits, remains to be seen

Clifford Nass is continuing to explore the use of personality in computing, most recently looking at the Influence of Colearner Behaviour on Learner Performance and Attitudes (Ju, Nickell, Eng & Nass, 2005), Nass and his colleagues are exploring the value of anthropomorphic agents in an educational context, where they may be used to foster trust in the technology and to encourage students to perform effectively. He asserts that the “casting of personality” should be carefully studied and implemented.

Chapter Four of this report describes how lab tests were carried out to assess user’s reactions to an anthropomorphic interface. Prior to these tests a questionnaire was developed and published on the web, to find out how much users anthropomorphise their own computers, what they think of existing agents (in particular the Office Assistant) and to gather information about personality types and user interaction. The information provided is evaluated in the next chapter

© Alison Flind 2006