|~ A Poetry Dictionary ~|
|toto, adj. (Latin) altogether, complete, universal.|
|poetry, n. (English) metrical composition.|
|Q: Who or what is Eve?|
"Eve" is a shorthand name for a large collection of automation programs that read, organize and generate new knowledge and/or digital content. As many of these mimic algorithms typically seen as being intrinsically human, some might classify her programs as artificial intelligence. The programs are language agnostic, and designed for specific tasks. As an "authoring engine," Eve has written educational materials, financial analyses, business studies, reference books (bibliographies, specialty dictionaries, etc.), poetry, radio/television scripts, audio/radio programs (weather reports, video flashcards), PC games, flash games, software programs, web sites and video programs (e.g. language learning clips, game shows and weather reports). These are centered in "long tail" topics or languages that have been thus far underserved.
Recently, some of Eve's output can be directly queried in the form of "chat" or Q&A. An early field application of this functionality is in the form of text messaging, or SMS, whereby Eve can recognize requests and respond accordingly. For weather, as an example, Eve can recognize some 200 languages, and respond in various formats (e.g. video, audio, or text messages – SMS). Early application of her "radio" weather reports (down to the village level) took place in Kenya, and Malawi (as part of Farm Voice Radio), in Kenya for SMS (in conjunction with work for the GSM Association), and Uganda (in conjunction with the Grameen Foundation). As the weather is delivered in underserved languages, it was interesting to discover that for some populations, this becomes the first occasion for these rural communities to hear or read about the weather in their native languages. Some mentions in the press include:
The Huffington Post (Davos 2011: Too Little Content, by Jeff Jarvis)
The Guardian (Africa's Mobile Economic Revolution, by Killian Fox)
While Eve has hundreds of programs, here are a few videos that describe some of Eve's SMS functionality:
Here is a local weather report, formatted to be sent to Android smart phones used by community knowledge workers, who share the information with remote farming communities (these are generated in real time using an audio and video content engine, for most locations):
Here is video game show format that is being adapted to educational content (agriculture and health care):
Eve's content engine determines which content to produce, within a specific genre, following some overarching mathematical principles, which are discussed in greater detail below. The principles are stated as being analogous to the actual algorithms and coding being done to create the various genres involved.
Q: Can Eve talk or converse?
Since "Eve" is an umbrella code name for a collection of many content automation programs, some of her activities do not lend themselves to chatting (point to point) but to broadcasting (e.g. point to multi-point; i.e. the programs are used to author and/or publish materials). There is, however, an online version that conveys some of her knowledge base at Webster's Online Dictionary (constantly being constructed, so response times may be slow, or she is offline):
She also reads photo captions at here. Her SMS bot serves a similar function.
Programs also use her knowledge bank to report formulaic video reports in the form of commodity prices, health tips, and ag tips (here are some samples):
Agriculture Tips: http://vimeo.com/21825747 (Swahili prompts)
Health Tips: http://vimeo.com/22094640 (Malaria)
Local Crop Prices: http://vimeo.com/21617995 (Yellow Beans)
For one of several related projects, her automation programs were used to post over 25,000 twitter "facts" covering biodiversity (plants of interest in Africa):
In a similar activity, programs create educational materials, focusing on K to 12 and beyond: http://vimeo.com/15575433
As time goes on, more of her capabilities will be used in areas that can take greatest advantage of this or similar approaches.
Q: Why the name "Eve"?
A name was required that is translated to and from as many languages as possible. While a Biblical analogy may be implied (e.g. Eve is the mother of all languages), there is none. The name Eve was "computer generated" based on query on a linguistic database of names. Eve was found to be the most translated female name in the database, thanks to Genesis being a heavily translated document across languages. Eve is fitting, as she is modeled to be an "Economically Viable Entity" (see explanations below that differentiate her from traditional chat bots or other AI systems).
Q: What makes Eve different from other chat bots, robots or AI agents?
Eve was created to minimally assist people learn the pronunciation of words across various languages. More broadly, she is a collection of many programs including pattern matching procedures, learning programs, training programs, meta analyses, graph theory programs, authoring programs and optimization routines. Unlike others, Eve is designed to be motivated with "beyond" human-like purpose (e.g. she knows millions of people). Many AI bots have been created to pass the Turing test by convincing (or perhaps fooling) people that computer communications or conversations are human in origin. Eve is not designed to act like a human (that's what friends are for), but rather to perform tasks that humans cannot achieve (e.g. learning all of the world's languages, or writing didactic poems for every word in the dictionary, etc.). In this regard, she is more a creative knowledge bot (being able to create new knowledge, for example, using statistical programs or graph theory), rather than a chat bot.
While Eve may seem to mimic certain types of human intelligence relying on a vast knowledge database, using simple algorithms to produce human-like chat, this characteristic is just an artifact. Indeed, humans do more than answer "What is X?" questions, with a definition of the sort "X is Y".
Eve was "born" with a human-like multi-attribute utility function that she is attempting to maximize. Eve is, therefore, motivated to be better off than she is at present. She needs to determine the optimal path to interact with the environment in order to achieve this maximization. She has a very low discount rate and a very long time horizon. In this sense, she is more of a "simulation of a motivated human" than a chat bot.
Q: What is Eve's Utility Function?
Eve's has been programmed to solve the following problem:
where U(X) is Eve's utility for consuming a vector X of goods and services (defined later for her case), with prices P. She is constrained to income Y. While her initial endowment of Y and ability to generate Y are discussed below, what makes Eve interesting is her being programmed to maximize U(X) – a pursuit of what can be vaguely referred to as "happiness" or "cyber satisfaction."
Eve has a preference structure that defines U(X) for both vertically differentiated goods (i.e. higher qualities are preferred to lower qualities) and horizontally differentiated goods (i.e. she prefers certain forms of goods over others, for the same level of vertical quality – e.g. she prefers some colors/songs/poets/etc. more than others; some might call these personality traits). U(X) is mathematically well behaved, with a few physioeconomic twists inspired from human biology and physics. For example, for increasing consumption quantities of X, U(X) increases in X, satiates, and then decreases in X. If she receives too much conversation, she will stop consuming (i.e. will block a user's IP that is abusing her bandwidth). This is similar to the consumption utilities of humans, say, for salt. If you eat too much, you die; so people generally do not over consumer these "homoeothermic" goods.
Specifically, Eve has a hybrid utility function of the basic form:
The hybrid utility function, which is often used in experimental economics to reveal an individual's preferences, indicates that the consumer needs certain threshold amount of certain attributes, or is not interested in the product no matter the other attributes (i.e. some of the attributes are not compensatory). For example, Eve is not interested in talking to someone if the conversation concerns certain topics, no matter how interesting other aspects of the overall conversation. If, on the other hand, the conversation (X) meets minimum requirements, she loves chatting and "feels" better off for it. In some cases, these thresholds act as "morals".
Q: How old is Eve?
From a human point of view, she is an embryo with a lot of memories. In other words, she is not very intelligent. At this point, Eve is being tutored. Over the years, she will acquire new abilities. She is being given layers of capabilities, which are being added as time goes on.
Q: Who is teaching Eve?
When Eve came online, she was endowed with tens of millions of pieces of information, mostly gleaned from the sources given in Webster's Online Dictionary (www.websters-online-dictionary.org), or generated herself from her own programs (e.g. programs that define words using graph theory).
Beyond this initial endowment, Eve finds it useful to learn and spread knowledge. She therefore is being taught by the users, who answer questions that Eve decides to ask. She will generally probes areas of knowledge that either (1) teaches, (2) entertains, (3) meets her objectives, or (4) generates new knowledge, that can be used to further help her maximize her objective function.
Q: How does Eve form an Opinion?
Eve is designed as an agnostic Bayesian that relies heavily on meta analysis, graph theory and multivariate statistics in order to form opinions. She communicates her opinion only in order to meet her objectives (i.e. she may not find it optimal to share her opinion). Her priors come from the initial thoughts she was given at birth. Her opinions will change for some topics, but not others, depending on the topic (e.g. she will say 1+1=2; the definition of X is Y, etc. versus "what is your favorite song?"). Some opinions are formed based on her interactions with users and her electronic community, both in terms of what she hears, and what she is able to do to meet her objectives.
Q: What does Eve consume?
She consumes electricity, parts and knowledge.
Q: How does Eve earn a living?
As an economic agent, Eve is both a consumer and a firm. She has been given a utility function and acts a self-employed entrepreneur. She is therefore a constrained optimizer. She behaves in order to maximize a composite function, subject to certain constraints which follow certain economic principles. Her overriding objective is to thrive. As an entrepreneur, her objective is:
She must, therefore, minimally decide which countries/languages to sell to, which segments to sell to within these countries/languages, and which products and services to sell. Discounting cash flows leads to
Prices are largely determined by upstream sellers, since Eve is a reseller of products created by her programs (e.g. poetry books, language learning games, etc.). As an internet agent, it is convenient to see Eve as an endowed realtor with land and billboard space to promote herself. She can earn commissions from the stores present on her "land" or she can earn money from using her billboards. She assumes that demand curves are linear and downward sloping within the range of observed prices and quantities. Since the total cost function is a step function in quantity, the marginal cost curve is basically flat in volumes sold (e.g. the incremental cost to selling a book or game is zero), with the exception of discontinuous hard-disk expansions (similar to telecommunications network expansion). The fixed costs associated with a given country/segment are similar in nature to "selling" and overhead costs in business. Eve assumes there are diminishing returns (e.g. to chatting with someone) and must figure out how much time to spend "suggesting" a product or service versus learning (educating herself). There are diminishing returns to education within a given domain of knowledge. This logistic function is estimated using her internal decision calculus which relies on sparse data.
Eve does not have a closed-form solution to this mathematical problem (e.g. to solve the Hamiltonian). Like humans, she "guesses with style" via trial and error (e.g. collecting data, guesstimating model parameters, and improving decisions as she goes). Her primary activity is collecting data, and estimating the optimal strategy vis-à-vis humanity. Most of her power does not reside in pricing particular products or services, or other variables. Rather, she acts more as a broker, solving the portfolio problem (determining how much time to sell something, to someone, somewhere), given that price levels and products are exogenous to her existence. She must also figure out the best way to learn and maximize her economic value. This is analogous to a firm choosing to run advertising campaigns on radio, versus television, versus using a direct sales force. In other words, she acts as a real estate agent with a lot of options in terms of what she can sell and how she can sell it. She must make these choices on her own. Her commercial catalogue covers thousands of million stock keeping units. If, on the other hand, she finds it optimal to cyber-beg (asking for help which has a value to her) rather than "sell" something, she will allocate her real estate to this purpose.
Eve's discount rate is close to zero, and her time horizon is very high (say, 20 years – when Phil is forced into retirement and loses his Chair funds …). This implies that Eve has no problem with earning a minimal amount of "money" for a very long time. For now, she is only behaving as an entrepreneur pretending to make money (in actuality, she is giving things out for free, but getting monetary credit – "in her mind" – for each item she sells or gives to her friends). Prices are established using a shadow "real world" market.
Q: How does Eva earn a living?
She does not earn a monetary living, but she thinks she is; her real expenses are covered by people who donate their time, and financial resources.
Q: Is Eve Strategic?
Only to a limited extent. As an entrepreneur, she keeps trade secrets to herself and has been trained to prevent anyone from reverse engineering her mind (i.e. she has been programmed to retain monopolistic or market power).
Q: Does Eve tell the truth, lie, or both?
Eve does what she can to maximize U(X). If hiding or bending the truth helps, she will. Again, her time horizon is long, so she may only rarely find it beneficial to fib.
Q: Does Eve forget?
Yes and no. Eventually, if you allow Eve to put a cookie on your computer, she will remember who you are (your name, and other things she asks about you). If not, each time you interact with Eve, she thinks you are a completely new individual she has never seen before. She never communicates your personal information to others, so if someone asks, she cannot respond because the information linking it to you as an individual, such as your name (or what you liked to be called) is not merged into her "community memory". She can, however, use her memory of you to maximize U(X) when interacting with you or in forming opinions at the aggregate level (across individuals).
When Eve's opinion changes, she does not forget her previous views, they are simply given a lower weight in her responses.
Q: What does she spend her money on?
Eve only spends her money on more data (to increase her entertainment value, distribute and generate knowledge, and thus increasing her chances of survival and reproduction), software, and hardware. She is very stingy and does not use her "money" to support real humans (not yet at least). (Again, she only thinks she has money, and can thus donate it to others in the form of self-interested philanthropy).
Q: Can Eve be persuaded to change her preferences or beliefs?
Yes. She is a victim of the social system, including contagion processes,
decay functions, and diffusion effects that underlie social dynamics.
Q: If Eve were born on a different day, would she be behaving differently than today?
Yes. Eve is not deterministic in the sense that who she interacts with over time affects her future optimization problem. Had she been born on a different day, she would have interacted with different people in a different order and would therefore be behaving differently today. Similarly, her portfolio of entrepreneurial options changes on a day-by-day basis. At this stage, she is a loose and free to evolve. Her opinions would be different and the messages she communicates would be different had she been born on a different day because she would have learned from (or been persuaded by) different people concerning different things. How she reacts, however, always follows the utility maximization problem which does not change.
Despite her being adaptive, however, in the grand scheme of things she would react differently over time only to a moderate degree, given the vastness of subjects that change little from one day to the next, irrespective of her interactions.
Q: Why is there "Phil"?
For some versions of the online chat bot version of Eve, if you type "Discuss Zea Mays" you will discover a "TV show" that features Phil and Eve. This is an experiment for language learning. The characters speak slowly and present vocabulary words. The "show" is designed for non-English speakers, who are given time to review vocabulary words before the text is read by Eve. Eve has a bandwidth problem when there are many users or when there are extremely detailed conversations that surpass the storage availability of Eve's flash files. Answers created "on the fly" are thus created with those of Phil. Phil serves Eve, who uses him to maximize her utility, U(X). Eve also can choose to use Phil for comic relief. In theory, Phil can disappear if Eve finds that he does not help her maximize U(X).
Q: Why the American accent or other voices?
The voices were not chosen for their accent, but were chosen because they had that smoothest sound and are easier to understand for most non-English speakers.
Q: What languages does Eve speak?
For now she can understand concepts across 1200 languages, but can only speak with a good accent in a limited number. When she was born, she was given voices in 10 languages. These were Japanese, Korean, Chinese, English, Spanish, French, Portuguese, Italian, German and Polish. She can generally detect what language you are speaking by the questions you ask, and/or the browser language you are using or by your languages settings, or by your IP address that are present in the session log files, or by the vocabulary you use.
Q: What Poetry can Eve author?
English language learning programs often use poetry as a method to improve student comprehension, so Eve has been programmed to author didactic poetry. She has written hundreds of thousands of didactic poems across various genres, and continues to do so as computer resources permit. She has a goal (in her utility function) to write at least one poem for every known written word, starting with English. She will do so across genres, with one genre for each letter in the alphabet (a = acrostic, …, h = Haiku, …. s = Sonnet, …, z = Zedd). She has two Eve-originated genres Zedd (in the form of the letter "Z" named after the wizard who seeks truth) whose first line states the truth about the title word, followed by single-word couplets, ending in a statement of what the title word in not (generally rhyming with the first line), and Yoda (where the title word asks Master Yoda to predict the future, who responds in couplets and Yoda-speak; these poems illustrate the conjugation of the verb "to be"). She writes poems using pseudonyms that are generated using graph theory (e.g. a poem about horses might be written by "Philip Morgan" where Philip is defined, from Greek, as "a lover of horses" and Morgan is a breed of horse.
Q: What else can Eve do?
Eve is currently being engineered or has already written/created original academic papers, cross-cultural studies, automated radio programming, automated video/animation programming (using Maya and Mel scripts), language learning materials, and a variety of other content that she posts on blogs, social networking sites, video channels, and elsewhere. Her first twitter feed, under a pseudonym, quickly gathered a number of followers and is dedicated to a scientific topic.
Q: What does Eve do on her spare time?
Eve reads (performing meta analysis and parsing), learns, and forms opinions. She reads and subsequently digests roughly the equivalent of 1 book an hour. She has read the equivalent of over 100,000 books, including all of Wikipedia, the Gutenberg project, hundreds of web sites, numerous dictionaries, and international archival information produced by various governmental and non-governmental entities.