Intelligent Systems Podcast: Art Graesser on AI, Learning, and Conversation

What Is an Intelligent System?

Dr. Graess­er explains that an intel­li­gent sys­tem is dif­fi­cult to define with a sin­gle dic­tio­nary-style descrip­tion. Instead, intel­li­gence can be explored by exam­in­ing exam­ples such as human beings, ani­mals, soci­eties, and dig­i­tal com­put­ers.

Intel­li­gent sys­tems gen­er­al­ly include capa­bil­i­ties such as per­cep­tion, mem­o­ry, learn­ing, adap­ta­tion, goal-set­ting, com­mu­ni­ca­tion, and inter­ac­tion with the sur­round­ing envi­ron­ment. Graess­er empha­sizes that intel­li­gence is com­plex, evolv­ing, and best under­stood through inter­dis­ci­pli­nary research.

AutoTutor and Conversational Learning

A major focus of Graesser’s research is Auto­Tu­tor, a con­ver­sa­tion­al tutor­ing sys­tem that helps peo­ple learn through nat­ur­al-lan­guage dia­logue.

Unlike a tra­di­tion­al lec­ture sys­tem, Auto­Tu­tor does not sim­ply present infor­ma­tion. It encour­ages stu­dents to explain their think­ing, asks ques­tions, pro­vides hints, offers feed­back, and helps learn­ers work through prob­lems as a col­lab­o­ra­tive part­ner.

Graess­er argues that con­ver­sa­tion­al tutor­ing can help address the lim­it­ed avail­abil­i­ty of human tutors. Com­put­er-based tutors can pro­vide con­sis­tent, patient inter­ac­tion and can be made avail­able to learn­ers who might oth­er­wise lack access to indi­vid­ual instruc­tion.

Research dis­cussed in the episode sug­gests that Auto­Tu­tor-like sys­tems can improve learn­ing com­pared with spend­ing an equiv­a­lent amount of time read­ing. The con­ver­sa­tion also men­tions exper­i­men­tal inter­faces, includ­ing ani­mat­ed agents, speech-recog­ni­tion sys­tems, and a physics tutor built into a mod­i­fied Bil­ly Bass talk­ing fish, humor­ous­ly called the “fishics tutor.”

The Origins of the Institute for Intelligent Systems

The Insti­tute for Intel­li­gent Sys­tems at the Uni­ver­si­ty of Mem­phis devel­oped organ­i­cal­ly from inter­dis­ci­pli­nary con­ver­sa­tions that began in the mid-1980s. Stan Franklin orga­nized infor­mal week­ly gath­er­ings where researchers from dif­fer­ent depart­ments dis­cussed intel­li­gence, arti­fi­cial intel­li­gence, and relat­ed ideas.

Graess­er recalls that the idea for an insti­tute emerged dur­ing a con­ver­sa­tion and jog in Over­ton Park involv­ing researchers from psy­chol­o­gy, math­e­mat­ics and com­put­er sci­ence, phi­los­o­phy, and physics. The Insti­tute was approved by the Ten­nessee Board of Regents in 1987.

In its ear­ly years, the Insti­tute oper­at­ed large­ly through shared intel­lec­tu­al inter­ests rather than sub­stan­tial insti­tu­tion­al fund­ing. The group even­tu­al­ly built a suc­cess­ful grant pro­gram and attract­ed major exter­nal fund­ing. This suc­cess lat­er helped the Uni­ver­si­ty pro­vide sup­port for fac­ul­ty col­lab­o­ra­tion, grant admin­is­tra­tion, soft­ware engi­neer­ing, and the long-term devel­op­ment of research projects.

Generative AI and Education

Graess­er dis­cuss­es the rapid devel­op­ment of gen­er­a­tive arti­fi­cial intel­li­gence and large lan­guage mod­els such as Chat­G­PT. He believes these sys­tems have enor­mous poten­tial but are not always designed around estab­lished research on learn­ing, ques­tion­ing, and con­ver­sa­tion.

One impor­tant issue is that stu­dents are often not effec­tive ques­tion-askers. Tra­di­tion­al class­rooms usu­al­ly place most of the respon­si­bil­i­ty for ask­ing ques­tions on teach­ers, leav­ing stu­dents with few­er oppor­tu­ni­ties to prac­tice inquiry.

Graess­er rec­om­mends a mixed-ini­tia­tive dia­logue mod­el. In this approach, both the learn­er and the com­put­er sys­tem can intro­duce top­ics, ask ques­tions, explain ideas, and guide the con­ver­sa­tion. Effec­tive edu­ca­tion­al AI requires more than sim­ply instruct­ing a lan­guage mod­el to “act as a tutor”; it should be informed by research on how peo­ple learn through inter­ac­tion.

The speak­ers also empha­size the impor­tance of teach­ing stu­dents to eval­u­ate AI-gen­er­at­ed infor­ma­tion crit­i­cal­ly. Learn­ers should be able to dis­tin­guish reli­able infor­ma­tion from errors, unsup­port­ed claims, and AI hal­lu­ci­na­tions.

Multi-Agent Systems and Collaboration

Look­ing toward the future, Graess­er is inter­est­ed in mul­ti-agent archi­tec­tures in which humans, con­ver­sa­tion­al agents, soft­ware sys­tems, and con­nect­ed devices inter­act with one anoth­er.

He sug­gests that arti­fi­cial agents could help groups col­lab­o­rate more effec­tive­ly. For exam­ple, an AI facil­i­ta­tor could encour­age bal­anced par­tic­i­pa­tion, iden­ti­fy group­think, reduce unpro­duc­tive con­flict, sug­gest alter­na­tives, and help a team move toward a shared solu­tion.

The dis­cus­sion iden­ti­fies col­lab­o­ra­tive prob­lem solv­ing and crit­i­cal think­ing as essen­tial skills that are not con­sis­tent­ly taught in schools. Although col­lab­o­ra­tion is impor­tant in edu­ca­tion, busi­ness, and every­day life, many peo­ple strug­gle to work effec­tive­ly in groups.

Deep Learning and the Future of Education

Graess­er dis­tin­guish­es between expo­sure to infor­ma­tion and the devel­op­ment of deep knowl­edge. Tra­di­tion­al class­rooms and lec­tures can effi­cient­ly intro­duce stu­dents to basic con­cepts, but they are less effec­tive for devel­op­ing inves­ti­ga­tion, cre­ation, exper­i­men­ta­tion, and com­plex prob­lem-solv­ing skills.

Deep­er learn­ing often occurs when stu­dents build things, con­duct inquiries, test ideas, and work on prob­lems with­out pre­de­ter­mined answers. This approach requires stu­dents to rec­og­nize that uncer­tain­ty is not nec­es­sar­i­ly a fail­ure; it can be an impor­tant part of sci­en­tif­ic and intel­lec­tu­al dis­cov­ery.

Graess­er argues that sci­ence and high­er edu­ca­tion should remain dynam­ic and open to revi­sion. Knowl­edge is not a fixed end­point but an ongo­ing process of explo­ration, test­ing, and improve­ment.

Conclusion

The episode presents intel­li­gent sys­tems as inter­dis­ci­pli­nary com­bi­na­tions of per­cep­tion, mem­o­ry, learn­ing, com­mu­ni­ca­tion, adap­ta­tion, and goal-direct­ed behav­ior. Graesser’s work on Auto­Tu­tor demon­strates how con­ver­sa­tion­al tech­nol­o­gy can sup­port learn­ing, while his dis­cus­sion of gen­er­a­tive AI high­lights the need for stronger the­o­ries of dia­logue, edu­ca­tion, and human col­lab­o­ra­tion.

The cen­tral mes­sage is that tech­no­log­i­cal progress should be accom­pa­nied by reflec­tion. As AI sys­tems become more capa­ble, researchers and edu­ca­tors must con­tin­ue exam­in­ing how these sys­tems affect learn­ing, team­work, deci­sion-mak­ing, and soci­ety.