A.L.I.C.E.: an ACE in Digitaland

  • Huma Shah Department of Artificial Intelligence & Interactive Multimedia, School of Computer Science, University of Westminster
Keywords: ALICE, ACE, Linguistic Productivity, Turing Test


Artificial linguistic Internet computer entity, A.L.I.C.E. is considered head and shoulders above other artificial conversational entities, an ACE in digitaland. Three times winner of Loebner’s annual instantiation of Turing’s Test for machine intelligence in 2000, 2001 and 2004 judged most human-like machine, A.L.I.C.E. was additionally gold medal champion in 2004, for most knowledgeable programme in Chatterbox Challenge and won bronze medal for most popular ACE. As a modern Eliza, A.L.I.C.E appears as a dark-haired, blue-eyed female avatar, or e-person. The programme’s architecture contains a combinatory scheme including key-word matching, spell checker, grammatical parser, random sentence generator and case-based reasoning or next-neighbour classification. These features allow A.L.I.C.E. to correctly identify the sense of word ‘live’ to produce responses about residential location when asked “where do you live?” and ask question about “subject” being “studied” when presented with “I study a lot”. As a discourse model, discourse features such as information exchange, disclosure of intentions, goals and desires are minimally exhibited in A.L.I.C.E.’s conversations; its verbal behaviour is akin to that of autistic children. However, A.L.I.C.E. type programmes appear on e-commerce Internet sites in a variety of roles; their use will continue to grow as more companies see their deployment as enhancing humancomputer interaction while building brand awareness and increasing sales. ELBOT, Loebner’s 2003 bronze runner up and Chatterbox 2003 winner, is the underlying technology behind text-based dialogical query system Anna, used by Swedish furniture store IKEA. As a virtual customer service agent, NY Wall Street Journal considered it a most useful ACE. As seen in both the Loebner Contests and Chatterbox Challenges, in unrestricted domains these programmes have a long way to go before they are able to constrain their artificial linguistic productivity to that which is meaningful and be deemed intelligent. Nonetheless, in single specialised e-domains ACE are succeeding and with speech-recognition augments will afford natural human-machine interaction.