Artificial Intelligence: A Modern Approach

Errata (Typographical and other errors in the book)

This is a list of typos and other errors in the book. Most of the 203 known problems are minor typos, but 10 of them are more serious errors, and could cause misunderstanding. They are marked with bold red page numbers followed by (BUG). Please send notification of additional typos or errors to or

In the entries below, we list the page number of each error along with the line number (negative numbers mean counting from the bottom of the page) and the chapter.

The list you need to see depends on the book you have. In all cases we're talking about the second edition of the book (the green one), not the first edition (the maroon one). An edition is a major revision done by the authors. Publishers also have printings; when they run out of books they print more. If any minor typos/errors have been spotted, they can be corrected in the next printing. To see what printing you have, look in the front of your book, at the page just before the dedication. There should be a line that says ISBN 0-13-790395-2 in a large font. Above that is a line of integers. The last digit in the line is the printing number. If it says

10 9 8 7 6 5 4 3 2 1

then you have a first-printing book; if it says

10 9 8 7 6 5 4 3 2

then you have a second-printing book, and so on. You also need to know if you have the International Edition (a paperback book that says "International Edition" on the cover) or the Hardcover Edition (because these are printed at different times, and thus pick up the corrections at different times). Then refer to this chart to see which of the errata you should read:

Printing 1whole pagewhole page
Printing 2skip ahead to herewhole page
Printing 3 or 4skip ahead to hereskip ahead to here

Errata for Printing 1 (International: 1 and 2)

Chapter 1

Page26   Line: -3
Should befrom

Chapter 2

Page37   Line: -14
Wasa entomologist
Should bean entomologist

Page43   Line: Figure 2.6, Poker line
Should beStochastic
CommentOn p.41, strategic environments are defined as deterministic except for other agents' actions.

Page45   Line: 1
Should bepercept

Page58   Line: -4
Wasif it dirty
Should beif it is dirty

Chapter 3

Page75   Line: 15
Wasthe goal test only to the a
Should bethe goal test only to the

Page75   Line: 19 and 21
Wasceiling of C*/ε
Should be1 + floor of C*/ε

Page81   Line: Figure 3.17 (twice)
Wasceiling of C*/ε
Should be1 + floor of C*/ε

Page85   Line: Figure 3.21
Should becorresponds

Page86   Line: 14
Wascontingency problems
Should becontingency problem

Chapter 4

Page95   Line: -3
Wasis depends
Should beit depends

Page102   Line: Figure 4.5, line 4
Should beSTATE[node]

Page103   Line: 1
Should beavailable

Page121   Line: 12
Should be

Page130   Line: 3
Wascombines of pairs
Should becombines pairs

Page134 (BUG)   Line: Exercise 4.2
Should beoptimal? You may assume that h is admissible.

Page135   Line: Exercise 4.11 b.
Wask = infinity
Should beone initial state and no limit on the number of states retained

Page135   Line: Exercise 4.11 c.
Wasat all times
Should beat all times (and omitting the termination test)

Chapter 5

Page156   Line: 10
Wasde Morgan
Should beDe Morgan

Chapter 6

Page165   Line: -14
Wasto obtain extend
Should beto extend

Page166   Line: Figure 6.3
Wasv ← MIN(v, MIN-VALUE(s))
Should bev ← MIN(v, MIN-VALUE(s))
CommentThat is, make the second v be italic. This occurs twice.

Page166   Line: -2
Should beMultiplayer

Page169   Line: -7
Should bebm/2

Page169   Line: -7
Should bebm

Page169   Line: -3
Should beb3m/4

Page169   Line: -1
Should bebm/2

Page170   Line: Figure 6.7, lines 12, 14, 23, 25
Should bev

Page182   Line: 3
Wasthat felt
Should befelt that

Page183   Line: -7
Wasapproach been
Should beapproach had been

Page191   Line: 15
Wasnodes the right
Should benodes to the right

Page191   Line: Ex. 6.6
Should be
CommentMove the book symbol down to avoid overlap with the keyboard.

Page192   Line: Ex. 6.12
WasWINNER(trick) is available that reports which card wins a given trick
Should beWINNER(s) is available that reports which player won the trick just completed, if any

Chapter 7

Page194   Line: -9
Wasuses in the form
Should beuses is in the form

Page194   Line: -7
WasIf its inside
Should beIf it's inside

Page195   Line: 14
Wasadvantage of being simple example
Should beadvantage of being a simple example

Page196   Line: 12
Should bethree
CommentWe failed to mention MAKE-ACTION-SENTENCE.

Page210   Line: Figure 7.11 (twice)
Wasde Morgan
Should beDe Morgan

Page212   Line: 16
Wasde Morgan
Should beDe Morgan

Page215   Line: 8
Should beln,k

Page215   Line: -16, -15
Wasde Morgan
Should beDe Morgan

Page216   Line: 1
Wasα does not entail β
Should beKB does not entail α

Page216   Line: 2
Wasα entails β
Should beKB entails α

Page216   Line: Figure 7.13, 2nd row
Should be
CommentExchange second and fourth boxes.

Page226   Line: Figure 7.19, line 12
Should bethen

Page227   Line: 11
Should beL2,1

Page227   Line: 17
Should beFacingRight1

Page235   Line: 23
Was(Bayardo and Schrag, 1997)
Should beBayardo and Schrag (1997)

Page239   Line: Ex. 7.15
Waskeeps keeps
Should bekeeps

Chapter 8

Page247   Line: -1
Wasthere roughly
Should bethere are roughly

Page252   Line: -8
Wasde Morgan
Should beDe Morgan

Page256   Line: 17
Should be+(0,m)

Page267   Line: 11
Wasde Morgan
Should beDe Morgan

Page271 (BUG)   Line: 5
Should be

Chapter 9

Page276   Line: -7
WasOne potentially confusing point is that one sense
Should beOne potentially confusing point is that in one sense

Page284   Line: 2 of Figure 9.5
WasDiff(nt, q) Diff(nt, sa)
Should beDiff(nt, q)Diff(nt, sa)

Page284   Line: 5
Wasif whether is a missile
Should bewhether it is a missile

Page291   Line: Figure 9.8, line 6
Was[a - x] . . . [a - z]
Should be[a|x] . . . [a|z]

Page297   Line: 8
Should beDistributeover

Page301   Line: -4
Should beresolution closure

Page307   Line: 1
Wasadds to the usable list
Should beadds to the set of support

Page307   Line: Fig. 9.14, line -3
Should be[clause | sos]

Chapter 10

Page320   Line: 6
WasSection 10.3 discusses representations for actions, which are central to the construction of knowledge-based agents. Section 10.2 covers the basic categories of objects and substances, and Section 10.3 explains the more general notion of events, or space-time chunks.
Should beSection 10.2 covers the basic categories of objects and substances. Section 10.3 discusses representations for actions, which are central to the construction of knowledge-based agents, and also explains the more general notion of events, or space-time chunks.

Page329   Line: -3
Should beAge(Wumpus,S0)

Page331 (BUG)   Line: -1
WasWe will address each problem in turn.
Should beWe will address each problem in turn. Even then, another problem remains---that of ensuring that all necessary conditions for an action's success have been specified. For example, Go fails if the agent dies en route. This is the qualification problem, for which there is no complete solution.
CommentThe qualification problem is referred to in later chapters, but never defined.

Page334   Line: -2
Washe occurrence
Should bethe occurrence

Page335   Line: 11
WasMany extensions . . . Section 10.3
Should beMany extensions to event calculus have been made to address problems of indirect effects, events with duration, concurrent events, continuously changing events, nondeterministic effects, causal constraints, and other complications. We will revisit some of these issues in the next subsection. It is fair to say that, at present, completely satisfactory solutions are not yet available for most of them, but no insuperable obstacles have been encountered.

Page335   Line: -13
Should bespace--time
CommentUse en-dash instead of hyphen.

Page335   Line: Footnote 4
Should bespace--time
CommentUse en-dash instead of hyphen.

Page338   Line: Figure 10.4 caption
Should be
CommentAdd periods after items (b) and (c).

Page339   Line: Figure 10.5
Was(Timelines i and j for Meet(i,j) overlap slightly)
Should be(Timeline i should end exactly when timeline j begins)

Page364   Line: 15
Was[modal operator]
Should be

Page368   Line: -8
Should beshopping

Page370   Line: Ex. 10.4
Was(Entire exercise)
Should beInvestigate ways to extend the event calculus to handle {\em simultaneous} events. Is it possible to avoid a combinatorial explosion of axioms?
CommentThis exercise referred to a subsection that was dropped from the text.

Page370   Line: Ex. 10.6
Should be
CommentWrong italic font used for ExhaustiveDecomposition, Partition, and Disjoint.

Page370   Line: Ex. 10.10(a)
Should beDuring

Chapter 11

Page379   Line: -3
Wassyntax called the the
Should besyntax called the

Page379   Line: -2
Should bebenchmark

Page380 (BUG)   Line: -4
Was[Load(C1, P1, SFO), Fly(P1, SFO, JFK), Load(C2, P2, JFK), Fly(P2, JFK, SFO)]
Should be[Load(C1, P1, SFO), Fly(P1, SFO, JFK), Unload(C1, P1, JFK) ,Load(C2, P2, JFK), Fly(P2, JFK, SFO), Unload(C2, P2, SFO)]

Page392 (BUG)   Line: -3
Wasremove the Remove(Spare, Trunk) action
Should beremove the LeaveOvernight action

Page393   Line: 3
WasAt(Spare, Tire)
Should beAt(Spare, Trunk)

Page399   Line: Figure 11.14, column S2
Was¬ At(Flat, Axle); At(Flat, Axle)
Should beAt(Flat, Axle); ¬ At(Flat, Axle)
CommentThe negation (¬) sign is in the wrong place in column S2

Page406   Line: 16
Should be

Chapter 12

Page418   Line: Figure 12.1, line 5
Should beEngine(e,c)

Page418   Line: Figure 12.1, line 8
Should bePRECOND: EngineIn(c)

Page421   Line: Figure 12.3, line 6
Should beEngine(e,c)

Page421   Line: Figure 12.3, line 11
Should bePRECOND: EngineIn(c)

Page425   Line: 17
Waspreconditions are effects
Should bepreconditions and effects

Page433   Line: 18
Should beEFFECT: AtL ∧ ¬AtR

Page436   Line: 6
Wasstate it is in
Should bewhat state it is in

Page437   Line: 5
Should bestate set

Page439 (BUG)   Line: -13
WasFortunately, the schemes are closed under STRIPS updating: as long as we stick to STRIPS axioms, then if we start with a representable belief state, all subsequent belief states will be representable too.
Should beIn particular, if an action occurs, one of whose preconditions is unknown, then the resulting belief state will not be exactly representable and the action outcome becomes unknown.

Page440   Line: -10
Wasproblems intractable
Should beproblems are intractable

Page445   Line: -12

Page451   Line: 1
Should besatisfies

Page455   Line: 10
Should beMultiagent

Page461   Line: Ex. 12.20
Should be
CommentWrong italic font used for Drink, Medicate.

Page461   Line: Ex. 12.21
Should be
CommentWrong italic font used for Test, CultureGrowth, Disease, Medicate.

Chapter 13

Page462   Line: -11
Wasnot be conclude
Should benot be able to conclude

Page464   Line: -20
WasAssigning probability of 0
Should beAssigning a probability of 0

Page467   Line: -10
Wassnow an an
Should besnow as an

Page467   Line: -2
WasElementary, propositions
Should beElementary propositions,

Page471   Line: 7
Waswhich may could
Should bewhich might have

Page472   Line: -20
Wasbetting systems described earlier
Should bebetting systems described on page 474

Page474   Line: 16
Wasbets Agent 2
Should bebets by Agent 2

Page481   Line: 13
WasWe can try using use
Should beWe can try using

Page488   Line: 14
Wasstraightforward reference-class method..
Should bestraightforward reference-class method.

Chapter 14

Page500   Line: -2
Wasthe the entire CPT
Should bethe entire CPT

Page515   Line: Eq. 14.8
WasP(yi | parents(Yi))
Should beP(zi | parents(Zi))

Page515   Line: Eq. 14.8
Should beP(z,e)

Page516   Line: 6
Wasincreases. Because
Should beincreases. This is because

Page517   Line: Figure 14.15
Was(increments N outside the innermost loop)
Should be(increments N inside the innermost loop)
CommentFor this particular version of MCMC, the results are identical, but for other versions (e.g., random selection of variable to sample), the new version is correct.

Page523   Line: -19
Wasnot surprising.)
Should benot surprising.

Page535   Line: Ex. 14.5(a)
Waslet X
Should belet X1

Chapter 15

Page540   Line: 6
Wascombined with the the
Should becombined with the

Page546   Line: -3
WasIt is now understood show that the
Should beIt is now understood that the

Page548   Line: Figure 15.5 caption
Wasindicates its best predecessor.
Should beindicates its best predecessor as measured by the product of the preceding sequence probability and the transition probability.
CommentThis clarification avoids a possible misunderstanding: the best predecessor is not necessarily the one with the highest sequence probability.

Page568   Line: -3
Wasphones that are used English
Should bephones that are used in English

Chapter 17

Page616 (BUG)   Line: Figure 17.2(b)
Was1.6284, 0.4278, 0.0850, 0.0221
Should be-1.6284, -0.4278, -0.0850, -0.0221
CommentThe numbers that are the bounds for R(s) should all be negative (except for zero), e.g. R(s) < -1.6284

Page616   Line: 4
Wasand (4, 2)
Should beand (3, 3)

Page618   Line: -7
WasAn optimal policy π*
Should beAn optimal policy π* satisfies

Page619   Line: -9
WasfromChapter 16
Should befrom Chapter 16

Page619   Line: -3
Was,a ssuming
Should be, assuming

Page624   Line: Figure 17.7
Wasreturn P
Should bereturn π

Page625 (BUG)   Line: Equations after 17.10
Should be= -0.04 +

Page632   Line: -2
Wasstrategy mixed)
Should bestrategy

Page636   Line: 19
Wasminimizer, will alway choose
Should beminimizer, will always choose

Page638   Line: -22
Wasan so only
Should beand so only

Page638   Line: -13
Wasrepeated game
Should berepeated game
CommentAlso remove repeated margin entry.

Page639   Line: 4
Should be
CommentAdd margin entry for perpetual punishment.

Chapter 18

Page653   Line: 15
Wastradeoff is not a simple
Should betradeoff is not as simple

Page657   Line: 9
WasFigure 18.4(c)
Should beFigure 18.4(b)

Page658   Line: Figure 18.5
Wasm ← MAJORITY-VALUE(examplesi)
Should bem ← MAJORITY-VALUE(examples)
CommentThat is, drop the i subscript. Also, MAJORITY-VALUE is the right name only if the class is binary; otherwise it should be MOST-COMMON-VALUE or MODE

Page660   Line: -5
Was5. Repeat steps 1 to 4
Should be5. Repeat steps 2 to 4

Page667   Line: Figure 18.10 caption
Wascombines hypotheses
Should begenerates hypotheses by successively reweighting the training examples.

Page673   Line: 4
Washas lead
Should behas led

Page673   Line: 12
Wasfor neural networks(see Chapter 20), and for sets of first-order logical sentences (see Chapter 19).
Should befor sets of first-order logical sentences (see Chapter 19) and for neural networks (see Chapter 20).
CommentJust changing the order of the sentence to make the chapter numbers sequential.

Chapter 19

Page697   Line: 1
Should bethe

Page706   Line: 2
Should beheuristics

Chapter 20

Page730   Line: Figure 20.11
Should be(moved to top of p.731, section on learning hidden Markov models)

Page731   Line: 0
Should be
CommentFigure 20.11 moved to the top of this page.

Page733   Line: -3
Should be10

Page738 (BUG)   Line: Figure 20.17, the NOT figure
WasW0 = 0.5, W1 = 1
Should beW0 = -0.5, W1 = -1

Page740   Line: Figure 20.19(b)
Should be(reformat z-axis tic mark labels)

Page741   Line: Figure 20.20
WasI1 and I2
Should bex1 and x2.
CommentTo be more consistent with other usage in the chapter.

Page741   Line: -6
Wasand T is the true output value.
Should be

Page741 (BUG)   Line: -3
Wasg(y - ...
Should bey - g(...

Page743   Line: Figure 20.22(b)
Was(no data points explicitly plotted in (b))
Should be(add data points using + and × marks as in (a))

Page744   Line: Figure 20.23(a) and (b)
Should be(reformat z-axis tic mark labels)

Page746   Line: Figure 20.25, line 3
Should beM

Page756   Line: 16
Should bemodeling

Page758   Line: 12
Should be(umlaut on first o

Page759   Line: 15
Should be(umlaut on first o

Page761   Line: Ex. 20.16
Should beWj

Chapter 21

Page775   Line: -2
Should be(Move margin note down to avoid overwriting hand.)

Page776   Line: Figure 21.8
WasN sa
Should beNsa
CommentI.e., remove space after N (occurs twice).

Page777   Line: -15
WasA reinforcement learning can learn
Should beA reinforcement learning algorithm can learn

Page787   Line: 6
Wasvan Roy
Should beVan Roy

Page787   Line: -6
WasParr and Russell (1998), Dietterich (2000), Sutton et al. (2000), Andre and Russell (2002)
Should be(Parr and Russell, 1998; Dietterich, 2000; Sutton et al., 2000; Andre and Russell, 2002)

Page788   Line: Exercise 21.2
Should beeven

Chapter 22

Page792   Line: -2
Should beGeneration.
CommentI.e., insert period.

Page813   Line: 2
WasDuring(Now, e)loves
Should beDuring(Now, e))loves

Page832   Line: Ex 22.10-11
Should beadd computer keyboard icon to exercises 22.10 and 22.11

Chapter 23

Page836   Line: -1
Wasprobabilistic context-free grammar , or PCFG.1,
Should beprobabilistic context-free grammar, or PCFG,1

Page848   Line: 24
Should behere we

Chapter 24

Page873   Line: 3
Wastexture. 2
Should betexture.2
CommentThat is, remove space before footnote index.

Page886   Line: 4
Wastop down
Should betop-down

Page886   Line: 14
Wasin the image in the image
Should bein the image

Page886   Line: 15
Should befigure.3

Page886   Line: -15
Should beexamples

Page899   Line: Ex. 24.8
Should be24.8 (Courtesy of Pietro Perona.)

Page900   Line: Ex. 24.10
Wascloser to
Should becloser than

Chapter 25

Page906   Line: Figure 25.4
WasThe Sony AIBO robot.
Should beSony AIBO robots playing soccer (copyright 2001, The RoboCup Federation).

Page907   Line: 5
Wassuch that can be
Should beso that they can be

Page920   Line: Figure 25.14
Was( (a) is workspace, (b) is configuration space.)
Should be( (a) is configuration space, (b) is workspace.)
CommentI.e., switch the two figures.

Page923   Line: 19
Should beunmodeled

Page923   Line: 22
Should bemodeled

Chapter 26

Page963   Line: 15
Was''intelligence explosion
Should be``intelligence explosion
CommentWrong open-quote marks used.

Page963   Line: -12
Was''mind children
Should be``mind children
CommentWrong open-quote marks used.

Page964   Line: -13
Waseight potential threats
Should besix potential threats

Page964   Line: -13
WasWe concluded that . . . from robotics
Should beWe concluded that some of the threats are either unlikely or differ little from threats posed by other, ``unintelligent'' technologies. One in particular is worthy of further consideration: that ultraintelligent machines might lead to a future that is very different from today---we may not like it, and at that point we may not have a choice. Such considerations lead inevitably to the conclusion that we must consider carefully, and soon, the possible consequences of AI research for the future of the human race.
CommentAn earlier draft included coverage of the problem of accidental or intentional weapons of mass destruction stemming from AI research gone amok. We decided to drop the coverage because the problems are really more associated with bio- and nano-technology.

Chapter 27

Page969   Line: -4
Wasan an
Should bein an

Appendix A

Page980   Line: -7
Was-3x - y + 2z = 11
Should be-3x - y + 2z = -11

Page983   Line: -2
WasGood texts on linear algebra include Chung (1979) and Ross(1988). For probability, Bertsekas and Tsitsiklis (2002) and Feller (1971) are valuable.
Should beGood texts on probability include Chung (1979), Ross(1988), Bertsekas and Tsitsiklis (2002), and Feller (1971).


Page1013   Line: Kaelbling et al. (1998) entry
Should beaction

Page1033   Line: Scholkopf and Smola (2002) entry
Should be(umlaut on first o)

Page1036   Line: Smith and Weld (1998) entry
Should be889--896

Errata for Printing 2 and up (International: 3 and up)

Chapter 3

Page63   Line: Figure 3.2
Should beDrobeta

Chapter 5

Page126   Line: footnote
Wastwo dimensional
Should betwo-dimensional


Page138   Line: 17
Wasgoal test can written
Should begoal test can be written

Chapter 11

Page388   Line: 19
Should besimplifies


Page493   Line: 7
Should beP(Xi|Parents(Xi

Page514   Line: footnote
WasIdeally, we would like use
Should beIdeally, we would like to use

Chapter 14

Page535   Line: Exercise 14.7(b)
Wasperformed ,
Should beperformed;


Page669   Line: -5
Wasat least 1 - ε
Should beat most 1 - ε

Chapter 20

Page755   Line: -9
WasChapter 13. is
Should beChapter 13. Its success is

Page729   Line: 2
Wasthe bag is is
Should bethe bag is

Page759   Line: -1
Waswise to sell
Should bewise

Chapter 22

Page812   Line: Figure 22.16, line -1
WasVerb(λ x λ y Loves(x, y))
Should beVerb(λ y λ x Loves(x, y))

Page813   Line: 2-3
Wasλ x λ y
Should beλ y λ x

Chapter 24

Page868   Line: 16
Wasmerrily ignores
Should bemerrily ignoring

Page869   Line: 10
Wasnose in the image
Should benoise in the image

Page885   Line: -7
WasAl of the techniques
Should beAll of the techniques

Page887   Line: 6
Wasand brightness-based recognition, which involves
Should beand feature-based recognition, which involves

Many thanks to Philip Adenekan, Daniel Bernstein, Wolfgang Bibel, Solly Brown, Peter Clark, Noel Codella, Ernesto Costa, Joao Balsa da Silva, Adnan Darwiche, Francisco De Comite, Joshua Estelle, Nafeh Fananapazir, Will Fitzgerald, Andrea Frome, Dan Frost, Charlie Garrett, Cyrus Harmon, Irvin Hwang, Robert Kerbs, Mykel Kochenderfer, Leilah Lyons, Norman Ma, Bob Marinier, Neville Mehta, Sjoerd Meijer, Wouter Meuleman, Ben Miller, Ken Modesitt, Kevin Kyung Nam, Mark Pearson, Radu Razvan Slavescu, Hisham Sueyllam, Larry Weber, Claus-Peter Wirth, David Yeung, and the others who contributed corrections.

AI: A Modern Approach by Stuart Russell and Peter NorvigModified: May 10, 2006