“How can I learn AI?” Lindsay Evett, Alan Battersby, David Brown, SCI NTU Penny Standen, DRA UN
Application-Based Teaching • Relevance • Active enquiry and exploration • Can be case based • Constructivist • Kolb Accommodating (Concrete Experience/Active Experimentation) • Supports more traditional methods (lectures, seminars) – blended learning
Real Applications • Real applications which are publicly and easily available • Some demonstrable success • More convincing than text book toy/engineered examples • While evaluation data often lacking they are braving the open market
Current Practical Work • Chatbots • Topic bridges AI and NLP; so highly suitable for my AI&NLP module • Applications – interfaces in general, IKEA call centre, search engines, Virtual help/assistant, naughty chat lines…… • Mostly work through pattern matching
Chatbot methods? • Available chatbots methods elusive • Jabberwacky says no single recognised AI technique – complex layered heuristics • Others perhaps some form of knowledge bases to produce learning, personalities, knowledge, unclear how
Coursework Requirements • The coursework requires students to use simple forms of AI&NLP techniques to improve conversation of a simple, ELIZA type, pattern matching Chatbot • The Chatbot provided has optional • Speech output • Lexicon • Lots of scope
Learning Outcomes • Use, apply and critically evaluate major techniques used in AI and NLP • Analyse, design and develop AI computer applications • Solve problems using AI and NLP techniques • Use and apply basic algorithmic and design approaches
Future Developments • Intelligent virtual tutors • Could be knowledge based • Could be proactive (need to identify situations and act appropriately) • Could have conversations, discussions, answer questions (Chatbots incorporated) • Plenty of scope
Intelligent Virtual Tutor Agents • Unobtrusive but reactive/proactive tutoring agents • Monitor student actions and visit when need arises, giving advice and/or instructions • Can need knowledge for monitoring
Types of Tutors Deductive tutor agents: give advice on deductive reasoning, (e.g., NDSU Geology Explorer http://oit.ndsu.edu/menu/): a. Equipment Tutor b. Exploration Tutor c. Science Tutor Case-based tutor agents: present relevant cases/experience (e.g., video to demonstrate experimental procedures (Yu et al 2005))
Types of Tutors (contd.) Rule-based tutor agents: a. encode set of rules about domain. b. Monitor student actions for broken rules c. Visit student to provide expert dialogue or tutorial Navigational tutor agents: supply context dependant information to aid navigational tasks (e.g., Quest http://quest.isrg.org.uk)
Develop Tutors • Virtual Health Clinic www.isrg.org.uk/VHC • Currently presents information as text when necessary (information buttons) • Clinic as environment, many opportunities for simple interventions • Receptionist ++…….?
Other Suitable AI Applications? • Game AI – many different methods involved • Speech XML • Tagging, Data Mining • Knowledge tools – search engines, Semantic Web, Ontology tools • Pattern recognition • Robots – toys, prosthetics • NB evaluation of many applications is lacking
Conclusions • Quite a few Weak AI successes • Mostly procedural • Not really intelligent? • A few have a range of methods • Becoming part of the pervasive background • Soft computing type systems as the basis for developing higher order cognition?