Chapter Eleven Artificial Intelligence II: Operational Perspective What is AI? From one perspective, AI is the study of automata (machines) that can learn, understand, interpret, and arrive at conclusions in a manner considered intelligent, just as if it were being carried out by a human. Some Approaches To AI “Top Down” (Abstract thinking and logical processes) Formal Logic Deduction Induction Abduction Neural Net Fuzzy Logic “Bottom Up” (Build a machine that is a “copy” of the brain and let it “think.”) A Sampling of Applications Management: Cost estimates, scheduling; intelligent document retrieval. Science & Engineering: prediction of chemical reactions; chemical identifications; equipment configuration; system troubleshooting; circuit design. Industrial: process control; mfg. quality control. Financial/legal: investment strategies; prediction of financial trends; loan application analysis; real estate price evaluation; estate planning. Medical: image processing; diagnosis; rehabilitation. Military and Space: classification of fingerprints; computer security; signal/target recognition. Other: language (natural language processing); speech recognition; prediction of sporting events; handwriting recognition; optical character recognition Architecture of the Knowledge-Based System INFERENCE ENGINE KNOWLEDGE BASE INTERFACE USER Interface: Allows user to access the system (questions, answers). Inference Engine: Includes reasoning (Production rules, Logic). Knowledge Base: Facts and abstract representation of the worldview. Logic-Based Reasoning Systems • See example of SNePS Expert Systems Operate in domains in which There are human novices. There are human experts. There are no well-defined “correct” answers. Novices can become experts. Novices are trained by experts. Novices are declared experts by experts. Production Rule technology often used. Fuzzy Logic • Replaces two-valued (True or False) logic. Belief in Fuzzy Logic belief that the person is old. - 1.0 0.8 0.6 0.4 0.2 our ‘confidence’ that an individual aged 30 is old is only 0.2. 0.0 age 0 10 20 30 40 50 60 70 80 90 100 Fuzzy Rules of Logic A and B = min (µA, µB) A or B = max (µA, µB) Not A = 1 - µA A Fuzzy Example Dieting—We all know that one has to have proper diet and exercise. In this case we will consider dieting alone. What we measure are the size of a person’s waist and the person’s weight; these are the "real world" variables. Our FL controller is going to recommend the kind of diet that the person should undertake. Waist Fuzzy Inference Weight Engine Diet Fuzzy Rules for the Example Rule 1: If (waist is “fat”) and (weight is “heavy”) then (recommend weight loss diet). Rule 2: If (waist is “normal”) and (weight is “normal”) then (recommend maintenance diet). (A diet index value of 0 means “stuff your face” and a diet index value of 100 means “prisoner’s starvation.”) Waist Membership Classes for the Fuzzy Example NA = normal waist NA F F = fat 1 32 34 36 38 40 42 44 waist Weight Membership Classes for the Fuzzy Example NW= normal weight NW H H = heavy 1 100 120 140 160 180 200 220 240 weight Membership Classes for the Rules of the Fuzzy Example M = maintenance WL = weight loss M (Rule 2) WL (Rule 1) 1 0.4 0.3 20 30 40 50 60 70 80 90 100 diet index Assessing the Facts for the Waist in the Fuzzy Example A person comes to our (very profitable) diet clinic with the following facts: waist = 37 inches weight = 170 pounds What diet should we advise? NA F 1 F=0.7 N=0.3 32 waist 34 36 38 40 waist = 37 42 44 Assessing the Facts for the Weight in the Fuzzy Example H = Heavy NW = Normal weight NW µNW=o.8 H 1 µH=0.4 100 120 weight 140 160 180 200 220 weight = 170 240 Reasoning in Words for the Fuzzy Example • Applying Rule 1 (Waist is fat and weight is heavy) The µ of the combination = min (µH, F ) = min (0.4, 0.7) = 0.4 We apply this to weight loss and this tells us to recommend a weight loss diet level index of 55 (see earlier membership curve). • Applying Rule 2 (waist is normal and weight is normal) The µ of the combination is min (µ[normal waste], µ[normal weight]) = min(0.3, 0.8) = 0.3 We apply this to the maintenance diet membership class that tells us to recommend a maintenance diet level index of 28 (see earlier membership curve). We appear to be confronted with two “conflicting” recommendations: Recommend dieting index of 55 and recommend maintenance diet of 28. We must resolve this and produce “crisp” results. Finding a Recommendation for the Fuzzy Example We must combine the recommendations of Rule 1 and Rule 2 into a single result. There are several ways to do this; one method is to generate a weighted average. The weight of each rule action is weighted by the corresponding membership of its condition and the result is then averaged. (28)(0.3) + (55)(0.4) Final dietary recommendation = (0.4 + 0.3) 43 43 represents a “moderate” diet somewhere between free range and starvation. In the real world this could be directly translated into daily caloric intake. Evaluation of Fuzzy Logic • Haack argues that there are very few true candidates for which Fuzzy Logic is useful. Most problems can be solved using principles drawn from probability. The computer programs are much too complicated and thus Fuzzy Logic serves no useful purpose. • Fox has rebutted this line of reasoning by noting that FL is effective when we need to describe real-world relationships that are “fuzzy.”