By Lance Eliot, the AI Trends Insider
The bishop took the opposing knight and the game was in check.
The other player assessed their situation and realized that the matter was now hopeless. Begrudgingly, the player conceded defeat and walked away from the board, vowing to never make the same mistake again. The winner was somewhat relieved because though it had looked promising that a win was within grasp, there were still a number of available counter-moves that the opponent could have used to try and gain momentum towards a possible win. Having the opponent concede or resign the game seemed maybe premature and almost surprising, perhaps even objectionable (hey, keep trying!), and yet there was no reason to not accept the concession since a win is a win.
What am I talking about?
For those of you familiar with chess, you likely recognize that I’ve described a rather typical end-game of a chess match.
If you are interested in AI, you likely would want to know about chess since the game has been used to help derive many of the AI techniques that we use today for all kinds of other endeavors. Chess is a deceptively easy game because the nature of the rules are relatively simple and readily described, and yet playing games successfully can be quite challenging (when I say successfully, I mean winning at chess).
Chess Offers Cognitive Challenges Helpful To AI Progression
Humans have been competing against each other since perhaps the Middle Ages to see who can be the best at chess.
It has always been hoped that we might be able to figure out how humans play chess so that we could then improve how humans can further play chess. Suppose that humans are able to play chess to a certain degree of sophistication, meaning being able to win against other humans a particular percentage of the time. Let’s study the topmost players and see if by doing so we are able to discover how they are so accomplished. We then share their tactics and strategies with other humans that perhaps are able to build upon that foundation and get even better at chess.
Without understanding why we make various moves during a chess match, it can be problematic to know what to do during a chess game. Is it better or worse to move a given piece on the chessboard to some other spot on the chessboard at any given time in a particular state of the chess game? No one can say for sure in all instances of any permutation, and we continually are seeking to find out to some approximate degree what are good moves versus bad moves for any state of the game.
If I only tell you the rules of the game, I’m not particularly also indicating what kinds of tactics and strategies to employ. You might gradually figure out on your own the kinds of winning tactics and strategies, but presumably it might help you if I gave you a head-start by providing what are believed to be good versus bad tactics and strategies when planning the game of chess.
Over time, we hopefully bootstrap humans toward getting better and better at playing chess. This is intended to be done by purposeful action. In other words, we are not getting better solely due to random chance and nor do perhaps only by the volume of people perchance playing the game, but instead because we’ve gleaned more about the good and bad ways to play chess.
One way to presumably try to gauge successful chess playing involves looking at the step-by-step play made during actual chess games. Avid chess players have been collecting game steps since nearly the origin of the game. There are books upon books listing the step-by-step moves in thousands and upon thousands of chess matches that have been played over time.
A black-box approach to analyzing how to best play chess can be undertaken by studying those step-by-step listed games. In that manner, you really don’t know anything about what is happening in the person’s head, and all you have to go with is the result of their chess moves. Why did I start a chess game with a move of my pawn? By studying the list of moves, you really don’t know why I did what I did, and the only thing you can do is see that I did so. From that listing, you can try to guess at what I might have been thinking.
Suppose you guess that I moved my pawn because I thought that the pawn should be in the middle of the chessboard. You now have a derived tactic, namely, assuming that I’m a really good chess player, you infer that by having a pawn at the middle of the chessboard it is a wise move and will ultimately increase the chances of winning a chess match. If I was a lousy chess player, you’d not be so enamored of my pawn move and likely assume I did something either by random chance or had dumb luck.
Maybe you decide to not only try and derive what I was thinking, doing so by looking just at the results or data listing of the steps performed in a chess match, but you also opt to ask me why I made that move of the pawn. This will allow you to take the black-box approach and see if you can poke into the head of the chess player, perhaps becoming a white box or transparent one.
You ask me directly why I did the move. I tell you that I did so because I was actually trying to free up potential movement for my knight and the pawn was in the way, so I moved it into the middle of the chessboard.
Well, that’s interesting. The earlier black-box data-only derived tactic was that a pawn should be in the middle of the chessboard. The tactic I claimed that I was using involved moving the pawn not per se for the sake of the pawn, instead it was done to provide an avenue or path to use another one of my chess pieces, my knight. You might therefore opt to rescind the derived tactic, crossing out the notion that the pawn should be in the middle of the chessboard as an overarching tactic or strategy, and you replace it with the approach that the knight needs room to move.
Here’s a question for you: Did I really in fact intentionally and with my mental thinking faculties come to the choice of moving the pawn to open up space for the knight, or did I just say that to try and justify my move?
Neither you or me really can know what occurred in my brain that led to the moving of the pawn. Maybe I moved the pawn because it was Thursday and there was going to be a full moon that night, which is to say that there might be any number of explanations for why I really made the move of the pawn. My telling you one particular explanation is not necessarily what really happened in my head. I have no way to know what really happened, and nor do you.
Note that I am not suggesting that I was maybe lying to you about why I believed that I moved the pawn. It is genuinely possible that I have a belief that I moved the pawn for the reason I articulated. Can I somehow inspect my own neurons in my brain and be able to physically and biologically explain how those neurons and the brain functioned to make that actual choice? Nope. Can’t do it.
The point being that to better play chess we only have the ability to analyze the results of chess matches and try to derive from that data what seem to be good tactics and strategies, which we can couple with those rationalizations or explanations that chess players claim they did and that are presumably good chess playing tactics and strategies. None of that reveals the true inner workings of our brain and what our minds were doing. We instead have reasonings that offer seemingly logical explanations for the chess playing behavior.
Some might say that you don’t need to know what actually happened in a chess players brain. The results of the games and the explanations by the chess players is sufficient, they contend.
Others though are worried that the logic-based approach to ascertaining chess play might not be sufficient. If we could unlock the secrets of the brain and figure out how it really plays chess, we might then not only be able to get better at chess, it might also give us greater insight into the brain and how we think. If we can decode how we think, this could allow us to improve our thinking overall, about all kinds of things, far beyond chess itself.
Chess as a Path to Thinking About AI
Understanding chess playing is more than merely being able to play chess.
Some hope that if we can better understand chess play, this can allow us to get better at all kinds of games, and furthermore get better at all kinds of thinking and the solving of problems of all kinds. Maybe we can use chess to act as a kind of Rosetta Stone to figure out how humans really think. Chess becomes a convenient tool to aid in decoding the mysteries of the human mind and how it works. Chess is incidental to the larger macroscopic picture of aiming at illuminating the basic foundation of human thought.
I mention this aspect about chess as a possible key to unlocking human thinking due to the often repeated comment that chess is only a game, and why are we spending so much time on some silly game? It could be that we spend all this time and attention on studying of chess play, and all we end-up with is that we are better at playing chess. How does that help world hunger? How does that aid solving real-world problems?
The hope is that the more we understand chess, the more we understand how we think.
Yes, admittedly, chess is only a game. It takes place in a constrained set of rules and does not especially rise to the nature of our open-ended and more challenging aspects we face as a society and in the real-world. Chess is interesting, it is fun, it is considered by many to be a sport. Besides those elements it also offers the possibility of getting inside of our heads.
There’s another angle too on chess, namely the desire to create artificially intelligent systems, and for which maybe chess will help us to get there.
If we can create an AI system that plays chess well, presumably we might discover how humans think and be able to embody that into machines. Once again, we would be using chess as a means to an end, wherein the end-game is to be able to create AI systems. The fact that we might have done so by also being able to create really good chess playing AI is not as consequential as the notion that by doing so we elevate the capabilities of AI overall.
I would dare say that from the perspective of achieving true AI, for me, I don’t care if it is chess or Monopoly or Tiddilywinks, meaning that whatever “game” might be the provocateur that enables us to reach true AI, I’m generally for it. Allow me to say that I do enjoy chess (which I mention herein so that I won’t get bombarded with hate mail from chess lovers), and I hope that chess is one avenue toward getting to true AI, preferably a strong avenue that offers foreseeable and motivating and earnest promise, but I am not so wedded to chess that I would want it to be do-or-die, namely we put all our table stakes on chess being the miraculous unlocking puzzle piece. Let’s not do that.
One difficulty though is that we might be able to make machines play really well at chess, even being able to best humans, and yet this might not necessarily mean that we are getting closer to understanding how humans think and nor that the machine embodies that capability.
I offer that caution because the latest AI systems to play chess are continually being touted as “superhuman” – a phrase that I find disconcerting. When you refer to an AI system in terminology that says it is superhuman, I’m concerned that many people assume you are suggesting that the AI system does what humans do, namely thinking, and do it even better.
Let’s be clear about things and all agree that the manner in which the “superhuman” AI playing chess systems operate is not necessarily how humans think.
Imagine that I was able to create a mechanical arm that had gears and wires. At first, it could not lift as heavy a weight as a human can. Let’s say it is considered sub-human at that juncture of development. I keep working on it and finally I get it to lift heavier weights than a human. I have invented a “superhuman” arm. It is in fact stronger than any human arm!
Have I therefore been able to recreate in a machine the same thing as a human arm? I don’t think so.
It would seem that any reasonable person would agree that I have created a really good mechanical arm, but it is not the same as a human arm. In an equivalent means, I am suggesting that though we might at this time have so-called “superhuman” AI systems that can play chess better than a human chess player, we are making a stretch to suggest that it means the AI system is able to think like a human.
We might have simply found some means to arrive at “thinking” in a completely different way than the actual way in which the human mind works.
I suppose you might argue that if we can achieve “thinking” via some other means than how the human mind works, we are doing pretty good and maybe have no need to worry whether or not it thinks as humans do. I would almost go along with that logic, but I’d like to point out that these AI chess playing systems are mainly confined to playing chess and other such games. We do not as yet know if they will “scale” to other kinds of thinking efforts. This means that it could be a false dead-end in that yes it might help us to create AI to play games, yet maybe that’s all it provides us in the end. We don’t yet know.
Chess as a Driving Metaphor
I often find myself mentally wandering over to the topic of chess when doing other kinds of mental activities. Perhaps you do so too.
Each morning I get onto the freeway to drive to work. I have about an hour or more commute that I drive while on the freeway. The freeway will have stretches that involve jam packed traffic, and other portions where the traffic is somewhat sparse and moving freely. This is Los Angeles traffic, notorious for its willingness to get snarled for the littlest of reasons. A car that pulls over on the freeway will attract human drivers to gaze at the car, and these gawkers then tend to slow down or otherwise trip-up traffic, often cascading into a miles long slowing and stopping of cars. Lookie loos are but one of the many reasons that we have our infamous stop-and-go traffic.
At times, I play chess games while driving. I don’t mean that I have a chessboard setup in my car. Instead, I am refereeing to “chess” in a metaphorical kind of way.
We all have occasion to suggest that we are doing some task and it is a chess-like effort. When my son used to play Little League baseball, we’d sometimes describe a baseball game in terms of chess. If the opposing team puts a certain pitcher up on the mound, what is our counter-move? If we put our best batter at the top of our batting order, will we reveal too soon the batting strength that we have? These are akin to playing chess and deciding how to make use of your chess pieces.
While driving my car, I look at the traffic ahead of me on the freeway and I envision a kind of chessboard. The cars directly ahead of me are particular chess pieces. That blue sports car to my left, its like a rook, and seems to go directly forward, while that beige sedan to my right Is more like a bishop, as it has been veering into other lanes at sharp angles. That big truck can go wherever it wants to go and no one will challenge it, as such I consider it to be the queen on the chessboard at the moment.The gaps between the cars are equated to empty spaces on a chessboard.
Which car, with each being essentially a chess piece, will next occupy that open board spot to my right, and how will they get there?
And so the chess game begins. If the open spot is immediately available, I can move my car directly into the opening. Suppose though that the opening is “guarded” by other nearby cars. I can potentially get to that open spot by first pulling ahead of the car to my left and maneuvering in front of that car by getting into its lane.
I next would zip ahead of the car that had been in front of me when I was in the prior lane, being able to do so now that I’m in the lane to the left. This then positions me to potentially slide into that open slot by crossing back into my former lane and then into the lane that was earlier to my right. I’ll need though to let the traffic in that targeted lane continue forward just a tad, and then time it just right to pop into that opening.
I don’t expect you to have followed my convoluted description of the moves that I made to get into that momentary opening in the lane to might right. Instead, I was just trying to illustrate the kinds of chess-like moves that I mentally entertain while driving on the freeway. I had to calculate where the other “chess” pieces are (the cars and trucks around me), I had to gauge the openings available for a move, I then devised a series of tactical moves that would get me positioned to get into the desired opening.
Sometimes the chess plays are straightforward, and I can execute them without issue. In other cases, I might mentally make my plan, such as the one I’ve just described, and it needs to be re-planned due to the changing traffic conditions. Keeping in mind that I’m on a freeway and going maybe 50-60 miles per hour, each of these chess plays are occurring in real-time. From the moment I think up a series of moves it might be just a few seconds once I’ve then executed those moves.
Furthermore, the moves that I planned out might only be valid for a few momentary seconds. Suppose the car that was to my left suddenly and unexpectedly sped-up? This would ruin my plan of trying to get ahead of that driver, which was a crucial initial step in my chess moves. I’d either need to back-down from the chess plan, or maybe concoct a new series of moves. It is a kind of cat-and-mouse match, continually requiring a reassessment of the freeway (the chessboard) and what seems viable to undertake.
Timing in chess play is considered vital in most competitive matches since there is usually a certain amount of time allowed per each move. This use of a real-time timing constraint forces the human chess player to make a choice that must take into account the available time for their thinking processes to work. Though you might want to try and use your mind to explore all possible permutations and combinations, which it’s not likely you could do anyway, you nonetheless must “cut short” your thinking and make a choice.
It used to be that many chess players would mail via the postal service each move to someone else they were playing against, giving the other person days or maybe even weeks to decide upon each and every move. Though some might still do this kind of slow play or snail play, the chess playing community has embraced fast play more so than slow play. One interesting question to ponder involves whether there is a material difference in chess play based on being able to play with nearly unlimited time to make a choice versus being confronted with very little time.
Most studies show that the difference between fast real-time such as just a few seconds versus longer real-time such as a few minutes tend to reveal better or worse play choices (the radically shortened time tends toward worse choices). This is tempered somewhat by the nature of the players and the moments and states of the chess match. If you have a really masterful chess player, a grandmaster, playing against a novice, it is likely that the grandmaster can make very fast choices since the chess plays are more predictable and known, plus if the grandmaster does happen to make a mistake they know it is likely readily fixable over the course of the game.
Top chess players when going head-to-head will play very fast during portions of the game that they’ve all come to know as predictable and will slow down once they hit the portion of the game that is in less predictable territory. For example, the opening of chess has been so exhaustively studied and the number of sensible moves is low enough that it can be very quick at the start of a chess match for players that know what they are doing. The same kind of super-fast moves can occur toward the end of the chess match, which often involves having very few chess pieces left on the board and therefore the number of variants of moves is lessened (along with their being many known end-game moves that you can employ).
The middlegame is often the portion that takes the most time for chess players to grapple with. You’ve gotten past the known opening gambits, and you are not yet to a point of thinning out the chessboard to be at the end-game. If you watch chess players during the middlegame, including even the grandmasters, you will often see them put their hands to their heads and they seem to go into a deep-thinking trance. We cannot know for sure what is happening in their noggins, but presumably they are having to consider moves on a rawer basis, going beyond the predictable patterns they’ve seen many times before for the opening and ending of the chess match. Depending upon what the middlegame board positions are, it can be unfamiliar territory as to the landscape and require more apt attention.
In the case of my doing a kind of mental chess when I am driving my car, I am equally faced with a real-time chess match. For each instance of deciding what my next driving move will be, I am doing so perhaps every 5 to 10 seconds of time. This turns out to be around 300 or so “moves” during my hour or so commute. If any particular move is poorly planned or poorly executed, it means that I’ve not proceeded in my commute in as presumably a timely manner as I might have hoped.
For the crucial timing and cognition aspects of driving, see my article: https://aitrends.com/selfdrivingcars/cognitive-timing-for-ai-self-driving-cars/
That topic brings up something that is perhaps not quite the equivalent of normal chess play. In chess, you finish a chess game as either a winner, a loser, or with a draw. The goal is to essentially capture the king of the other player. If you can do so, you are the winner. If the other player can capture your king first, you are the loser. If neither of you seem to be able to capture the other player’s king, it is considered a draw (there are other variants on how the win, lose, or draw can occur).
When I am driving my car and using a chess metaphor, there isn’t quite an overall win, lose, or draw that happens to be the end-goal. Usually, I am desirous of getting to work in the fastest way that also includes being safe. In that sense, you could suggest that a win is when I get to work at a shortest feasible time and do so without having gotten into a car accident. Each move must encompass the risks involved in safely driving the car. The overall safety of the driving journey is paramount and would usually be considered a much higher priority than the timing of getting to work.
If I make a wrong move in my driving chess game, it could either delay my driving time, or worse it could involve a car accident. There is a potential life-or-death kind of calculation immersed in this pretend chess. When playing chess at a chessboard, you normally aren’t worried about a life-or-death consequence (other than maybe in a James Bond movie). The stakes might be high when playing an actual chess game, perhaps prestige or money is on the line, but rarely does it have death or bodily injury at stake.
I’ve spoken to police officers and ambulance drivers that must at times drive for emergency purposes and therefore drive at high speeds in everyday traffic. For them, this idea of conducting a chess match in driving is heightened because they have true life-or-death stakes involved. Even though their sirens are blaring, and their blinking lights are trying to get the attention of everyday drivers, it is still a high risk action to drive very fast and opt to go through red lights or take other dire driving actions.
The chances of them hitting another car is increased and the chances of another car ramming into them is increased. They are taking such risks because there is the presumed risk involved of someone perhaps dying if they do not get to their destination fast enough (bless their hearts for taking such risks!). As a society, we seem to accept such risks, which I’ll point out are not only risks to the police officer driving the police car or the ambulance driver of the fire fighter driver, but there are obviously heightened risks to the everyday driver. The everyday driver is absorbing some of that risk since they could get hit by the emergency responding driver or they could inadvertently ram into the emergency responding driver.
Back to my driving with my chess metaphor in mind, I consider the freeway to be a continually moving chessboard. From my perspective, while driving along at sometimes 60 miles per hour or going in snarled traffic at 6 miles per hour, I imagine that the chessboard radiates out from my car. My car is the cornerstone for the imaginary chessboard. The distance ahead that I can see is the front far edge of my chessboard. The distance behind me that I can see via my rearview mirror is the rear far edge of my chessboard.
A normal chessboard is 8 rows and 8 columns consisting of a square board containing 64 spots. For my car driving, I consider each car length to be the equivalent of a spot. In terms of however many cars ahead that I can see, it is the number of spots for my metaphorical chessboard for that moment in time at the front of my car. Likewise, the same is said about the chessboard spots behind me. The chessboard is a rectangle that normally has just a few spots in width, such as maybe I am on a four-lane freeway and so the chessboard is four spots or squares wide.
When the freeway roadway is flat and I can see ahead quite a bit, I might have 10 to 20 car lengths ahead that I can see, and maybe 5-10 car lengths behind me that I can see. Therefore, my mental chessboard is perhaps 15 to 30 rows in total and let’s say by 4 columns wide in size. This won’t last very long though, and as traffic moves ahead and the roadway surface changes such as the freeway nears a curve or rises or lowers into a kind of driving valley, it is likely I will now only be able to see maybe 5 car lengths ahead and say 8 car lengths behind me. Plus, even on a flat surface, other cars and trucks can block my view. The point being that the chessboard is continually expanding and contracting, doing so during the driving journey, moment to moment.
It is within that playing space that the other cars and vehicles nearby are the other chess pieces.
I am trying to align and motivate those other chess pieces to play the game in the way that I want them to do so. They won’t necessarily want to play the game the way that I want to do so. I might be trying to get ahead of the car to my left so that I can get into that person’s lane, they meanwhile might be accelerating and not wanting to let me get ahead of them. They could be doing so on purpose or it could be by happenstance as they are either not paying attention to my car or they have some other maneuver they are trying to execute and for which it happens to counter my move.
Normal chess is a two-player game. In the case of driving chess, presumably every driver that I encounter on my freeway commute is playing a chess game. They are each playing their own chess game, of which, my chess game intersects with them at some point in time. There are maybe hundreds of simultaneous chess games occurring as I drive to work and find myself among hundreds of other cars and their drivers during the journey.
Consider the complexity involved in this virtual kind of chess.
Hundreds of other chess players, all seeking to “win” at their chess game (let’s assume a win consists of getting to their desired location as soon as possible and balanced by the safety factors of driving). Their chessboards are dynamically changing, doing so from moment to moment, just as mine is too, widening and shortening while driving along. I will eventually intersect with those other chess players when we get near to each other. Our chess play might intersect only briefly, maybe I zoom past another driver and soon have gotten far beyond their view, or it might be elongated such as when the traffic becomes bumper-to-bumper and for twenty minutes we are all stuck together in snarled traffic.
A means to reduce the complexity of perceiving this as a chess match of me against hundreds of other chess players involves making the game into a matter of it being them versus me. This metaphorical chess game is now reduced to a two-player game.
There is a morass of other players that I’ll assume are in essence one overall macroscopic player, which you can think of as Adam Smith’s “The Wealth of Nations” notion that they are all controlled by an invisible hand. Each of them is doing their own driving, obviously, and I am not suggesting that there is a conspiracy theory and nor that they are all fake acting aka “The Truman Show” or mind-controlled or something similar. I am merely reducing the perceived complexity by making this into a more traditional two-player setting. I represent me, and all of the other drivers are represented as one gigantic macro-player that involves perhaps hundreds of other chess players.
For my article about conspiracy theories, see: https://aitrends.com/selfdrivingcars/conspiracy-theories-about-ai-self-driving-cars/
Emotion Involved In Chess Play
When you play normal chess, you are likely to get involved in trying to psyche out the other chess player.
I mention this aspect because some non-chess players assume that chess is entirely a game played without any emotion and it is simply all intellect. If you watch even the grandmasters play, you can see that before they get to a chess match, they have often tried to psyche out the other grandmaster, doing so by making remarks about the other player. During the chess match, they will at times try to psyche out the other player, giving them the evil eye or acting as though they don’t have a care in the world or making a sigh at a move, etc. There are rules that prevent chess players in formal competition matches from going too far in this aspect of psyching out each other.
For some fun in terms of psyching out other chess players, you might want to one day go to watch the informal chess matches that occur in New York City at Washington Square Park or any similar venue. These matches sometimes involve “semi-pro” chess players that sit there all day long trying to make money at perhaps a few dollars per match as a wager. The ones that sit there all day are often prone to wild kinds of psyche-out approaches during a chess match. They tell you that you are smart and going to win, they on the next move tell you that you’ve blundered (regardless if you have), they ask you about the weather (a distraction), they warn you to be on the watch for an angry dog prowling the park (preoccupy your mind), and so on.
The point being that chess is not solely a game of disembodied beings that make chess playing choices dryly and without emotion. Human chess players are humans. This means they have all of the everyday foibles and emotions that seem to go along with being a human being. Sure, some of the chess players try for years and years to overcome their naturally occurring emotions and strictly play the game by-the-book. Some say that the Soviet Union during the Cold War tried to achieve this with their top chess players. In the end, it is nearly impossible to completely submerge and remove the emotionally charged elements from a human player.
The cars that are nearby me on the freeway are being driven by human beings. This means that they too are riddled with emotion. They will not necessarily make car driving choices that are entirely predictable by a purely rational calculation. This makes the metaphorical chess game more challenging. I cannot necessarily assume that the car driver to my left will “do the right thing” and let me into their lane. The other driver might purposely cut me off because they don’t like the look of my car or maybe they don’t like how I have been driving.
Besides the dangers of getting into a car accident while playing the metaphorical chess driving game, you also need to be watchful of getting into a road rage incident. If you drive in a manner that another driver dislikes, it can spark them into a kind of rage. They are going to potentially take out that rage by trying to drive their car to come after you. Whatever larger driving goal they might have had, such as getting to the grocery store, can be laid aside as they become fixated on trying to block your car or threaten you or whatever.
For more about road rage, see my article: https://aitrends.com/selfdrivingcars/road-rage-and-ai-self-driving-cars/
For the aspects of irrational driving, see my article: https://aitrends.com/selfdrivingcars/motivational-ai-bounded-irrationality-self-driving-cars/
For the human foibles of driving, see my article: https://aitrends.com/selfdrivingcars/ten-human-driving-foibles-self-driving-car-deep-learning-counter-tactics/
For the importance of defensive driving tactics, see my article: https://aitrends.com/selfdrivingcars/art-defensive-driving-key-self-driving-car-success/
While doing the metaphorical driving game, there are times at which the move you might have wanted to make will be blocked or cut-off. This I realize perhaps seems obvious. We all know how frustrating it can be when you are for example trying to get off the freeway, but no other cars are letting you get into the exit lane. You curse them as you see that you’ve now missed your exit. They would likely have little sympathy and emphasize that you should have started toward your exit sooner. And so, the daily grind of driving and at times lack of driving civility comes to the fore.
There are also driving moments wherein you are forced into making a driving move that you didn’t want to do. In normal chess, being forced into making a move that you prefer not to make is known as a zugzwang.
How Zugzwang Pertains To Autonomous Cars
For driving chess, let’s imagine an instance of zugzwang. You are in the fast lane and zipping along. You are eager to get to your destination and the approach so far has been to stay in the fast lane as much as possible. The other lanes of traffic are somewhat snarled, while the fast lane is moving at a really good clip. You suddenly come upon a car that is moving very slowly in the fast lane. The dolt! Don’t they realize they are in the fast lane.
You come right up to the bumper of the slow-moving car. The car stays where it is and does not speed-up. You flash your headlights at the car. No response. You honk your horn. No response. This slow-moving driver seems to be entrenched in the fast lane. If you could somehow push them out of the way, you would. Your only recourse seems to be to switch lanes, this though puts you into the snarled traffic, plus you’ll need to arduously make your way ahead of the slow-moving car while in the adjacent lane, and then try to enter back into the fast lane ahead of the tortoise driver. What a pain in the neck!
Do you choose to stay in the fast lane, moving now at a slowed speed, or do you make the maneuvers and contortions to try and get around the slow driver? You don’t want to have to do all of those contortions since you know it might end-up backfiring and you might fall further behind in the traffic. Your preference would be to stay in the fast lane.
You’ve just encountered a kind of zugzwang.
This example is not a truly forced zugawang in that you can opt to stay in the fast lane and just bear with it. There are plenty of driving examples whereby you are forced into a particular move.
The other day I was driving down a street that leads right to my desired destination, and it turns out that the police had blocked the road and were forcing all car traffic to take a detour. This was frustrating because I could see the destination and it was just a few feet on the other side of the roadblock. Nonetheless, I had to obey the police and take the detour (I suppose I could have tried to ram the roadblock, which might have been exciting, though not legal and I’d be probably in jail right now).
While driving on a driving journey, you are likely to have an overall driving strategy that guides your overarching driving efforts. This driving strategy might be that you want to get to your destination and avoid having to drive in the bad parts of town, along with the notion that you are willing to drive more slowly than usual because you want to enjoy the scenery along the way. Your driving tactics involve the moment to moment driving moves, and they are guided by the other driving strategy that you have. Executing a right turn up ahead is a driving tactic, while the aspect of making that right turn due to the goal of getting to your destination and avoiding the bad parts of town (which say that if you proceeded straight, you’d go into), encompassed the driving strategies you’ve devised.
Avid chess players typically have an overall chess playing strategy and couple it with various moment to moment chess playing tactics. You might have as an overarching chess playing strategy that you like to take over the center of the chessboard. Your opponent might not be as keen on that as a playing strategy and might instead believe in going to the opponent’s area and dominating that space. For each of those players, the moment they make any specific chess move, it could be that it is aiding their overall chess strategy. Not each tactical move necessarily does so, and it all depends upon the evolving state of play during a particular chess match.
Furthermore, you might adjust your chess playing strategies depending upon the nature of your opponent. For some chess players, they like to always play using the same chess strategies and for which they believe that it will beat any opponent. Other chess players might believe that you need to deploy a chess strategy that will be best suited against a particular player. I might abandon my normal default of wanting to control the center of the chessboard if I know that my opponent welcomes that kind of strategy and has come up with ways to undermine it.
There are some famous chess matches in which a top-level grandmaster suddenly switched from their traditional chess strategy and caused a stir. The opponent would likely be thrown for a loop because they had studied and prepared for the assumed chess strategy that was going to be most likely utilized. This kind of trickery can be handy, if you can pull it off well. If you switch strategies and are not as strong at the new strategy, maybe though you will do worse than if you had stayed with your tried-and-true.
Just as each chess match in normal chess is a new game, each time that you get onto the road you are starting a new metaphorical chess match.
You will have some kind of driving strategies and overarching goal, and this will be a guide during the moment to moment tactical aspects of your driving. When driving to work, you might adopt one particular set of driving strategies and tactics. Meanwhile, while on vacation in Hawaii, you might adopt a different set of driving strategies and tactics.
There are some human car drivers that seem to always have the same driving strategies and tactics. They do not particularly veer from it. This lack of flexibility will often get them into a traffic quagmire. They either do not realize that getting bogged down in the quagmire is due to their staid strategy and tactics, or they might realize it but decide to just proceed anyway, or they might be desirous of switching to a different strategy and set of tactics but do not know how, or have waited too long to do so on a timely basis that would make a difference.
For human behavioral aspects of driving, see my article: https://aitrends.com/selfdrivingcars/prevalence-induced-behavior-and-ai-self-driving-cars/
For family related trips and driving, see my article: https://aitrends.com/selfdrivingcars/family-road-trip-and-ai-self-driving-cars/
For the role of greed while driving, see my article: https://aitrends.com/selfdrivingcars/selfishness-self-driving-cars-ai-greed-good/
For the role of curiosity while driving, see my article: https://aitrends.com/selfdrivingcars/curiosity-core/
Is It Appropriate To Compare Driving And Chess
I hope that my discussion about chess playing as a driving metaphor does not alarm you. There are some people that are perturbed when I bring up this topic.
Part of the basis for their being perturbed is that they think I am perhaps mocking the seriousness of driving. By trying to apply the rules or sense of playing chess, they believe that I am not taking driving as seriously as I should. It is not a game, they would say. Peoples lives are at stake. There is a concern on their part that I am willing to maybe do things while driving because I am pretending it is a game, for which I otherwise would not undertake if I put aside a game-like mentality.
I assure you that I do take driving very seriously.
I am not applying a chess playing metaphor as though I am playing a video game and do not care about whether I hit other cars or strike pedestrians. My chess metaphor does not overwhelm my sense of sensibility. I can be and am a conscientious driver that abides by the driving laws and rules.
In fact, I would suggest or claim that the use of the chess metaphor actually aids and informs your driving ability. The more that you think about how to best drive, it would seem hopefully the better the driver you become. It seems to me that drivers that put little thought into their driving are more likely to be the ones that end-up causing accidents or creating untoward traffic situations. They are caught unawares because they are not putting sufficient cognitive cycles toward the driving task.
This brings up a related question that I sometimes get about the chess metaphorical driving. If my mind is used up by thinking about chess aspects of driving, wouldn’t this imply that I am perhaps over-thinking driving? Maybe I am putting too much thought into the driving process. There are some that believe you either know how to drive or you do not. By over-thinking it, you are presumably going to be a worse driver. You are using up precious and limited cognitive cycles that should instead be devoted to just driving, and not thinking about driving.
I counter-argue that the notion that more knowledge about something makes you worse at it, well, its an old line that I don’t think typically bears out. Is my mind so preoccupied with trying to figure out driving tactics and my driving strategy that I become oblivious to the roadway situation and therefore will tend toward getting into a car accident? I would assert that is the actual anti-thesis of the point of the chess playing metaphor, which is to do a better job at driving, including calculating the amount of cognitive effort going towards the driving task and being responsive to the real-time demands of the driving matters at-hand.
Anyway, I certainly hope that my discussion doesn’t alarm you. In addition, don’t try to become mentally engaged in considering your driving as a chess match if it will indeed cause you to become preoccupied or distracted from the act of safe driving. Whatever means you have of driving a car, if it seems to be working, probably best if you continue with it.
I bring up the chess playing metaphor not to somehow convince other humans to do so, but due to the notion that we can examine and understand to some degree the driving task via the use of a chess metaphor. Out of which, it might help us to devise AI and automation to tackle and undertake the human driving task, as you’ll see in a moment.
AI Self-Driving Cars and Chess Play
At the Cybernetic AI Self-Driving Car Institute, we are developing AI software for self-driving cars. The use of chess as a metaphorical way of looking at driving can be quite insightful, and aids in the advances being made towards developing true AI self-driving cars.
Allow me to elaborate.
I’d like to first clarify and introduce the notion that there are varying levels of AI self-driving cars. The topmost level is considered Level 5. A Level 5 self-driving car is one that is being driven by the AI and there is no human driver involved. For the design of Level 5 self-driving cars, the auto makers are even removing the gas pedal, brake pedal, and steering wheel, since those are contraptions used by human drivers. The Level 5 self-driving car is not being driven by a human and nor is there an expectation that a human driver will be present in the self-driving car. It’s all on the shoulders of the AI to drive the car.
For self-driving cars less than a Level 5, there must be a human driver present in the car. The human driver is currently considered the responsible party for the acts of the car. The AI and the human driver are co-sharing the driving task. In spite of this co-sharing, the human is supposed to remain fully immersed into the driving task and be ready at all times to perform the driving task. I’ve repeatedly warned about the dangers of this co-sharing arrangement and predicted it will produce many untoward results.
For my overall framework about AI self-driving cars, see my article: https://aitrends.com/selfdrivingcars/framework-ai-self-driving-driverless-cars-big-picture/
For the levels of self-driving cars, see my article: https://aitrends.com/selfdrivingcars/richter-scale-levels-self-driving-cars/
For why AI Level 5 self-driving cars are like a moonshot, see my article: https://aitrends.com/selfdrivingcars/self-driving-car-mother-ai-projects-moonshot/
For the dangers of co-sharing the driving task, see my article: https://aitrends.com/selfdrivingcars/human-back-up-drivers-for-ai-self-driving-cars/
Let’s focus herein on the true Level 5 self-driving car. Much of the comments apply to the less than Level 5 self-driving cars too, but the fully autonomous AI self-driving car will receive the most attention in this discussion.
Here’s the usual steps involved in the AI driving task:
- Sensor data collection and interpretation
- Sensor fusion
- Virtual world model updating
- AI action planning
- Car controls command issuance
Another key aspect of AI self-driving cars is that they will be driving on our roadways in the midst of human driven cars too. There are some pundits of AI self-driving cars that continually refer to a utopian world in which there are only AI self-driving cars on the public roads. Currently there are about 250+ million conventional cars in the United States alone, and those cars are not going to magically disappear or become true Level 5 AI self-driving cars overnight.
Indeed, the use of human driven cars will last for many years, likely many decades, and the advent of AI self-driving cars will occur while there are still human driven cars on the roads. This is a crucial point since this means that the AI of self-driving cars needs to be able to contend with not just other AI self-driving cars, but also contend with human driven cars. It is easy to envision a simplistic and rather unrealistic world in which all AI self-driving cars are politely interacting with each other and being civil about roadway interactions. That’s not what is going to be happening for the foreseeable future. AI self-driving cars and human driven cars will need to be able to cope with each other.
For my article about the grand convergence that has led us to this moment in time, see: https://aitrends.com/selfdrivingcars/grand-convergence-explains-rise-self-driving-cars/
See my article about the ethical dilemmas facing AI self-driving cars: https://aitrends.com/selfdrivingcars/ethically-ambiguous-self-driving-cars/
For potential regulations about AI self-driving cars, see my article: https://aitrends.com/selfdrivingcars/assessing-federal-regulations-self-driving-cars-house-bill-passed/
For my predictions about AI self-driving cars for the 2020s, 2030s, and 2040s, see my article: https://aitrends.com/selfdrivingcars/gen-z-and-the-fate-of-ai-self-driving-cars/
Returning to the topic of chess, let’s consider how the playing of chess relate to the advances being made toward developing true AI self-driving cars.
Chess Playing AI in Modern Times
I’ll start my discussion with a quick overview of the progression of game playing by citing AlphaGo Zero to AlphaZero, which are well-known game playing AI-based programs, and also discuss Deep Blue, an AI-based chess playing game that defeated the world chess champion in 1997.
If you are interested in the underlying details about those game playing applications, you might want to take a look at the December 7, 2018 issue of Science magazine that has an article entitled “A General Reinforcement Learning Algorithm that Masters Chess, Shogi, and Go Through Self-Play” and has a handy link to pseudocode depicting some of the algorithms involved.
IBM’s Deep Blue (or some called it Deeper Blue) application achieved popular notoriety when in May 1997 it was able to best Garry Kasparov, the reining world chess champion at the time, doing so in a final score of 3 ½ games to 2 ½ games (a draw was worth a half point) and the chess match abided by the official chess competition rules including time constraints.
If you like conspiracy stories, here’s a quick aside for you about the momentous occasion. There was some controversy about this win by Deep Blue, namely that Garry Kasparov later accused the developers of changing the code of Deep Blue during an actual match, specifically in the second game, and thus he claims he was beat not solely by a computer but by human intervention that adjusted the code to try and beat him.
The developers indicated that they did not change the code during the game play, though they did say that they changed the code between each of the games, which apparently was allowed by the rules of the chess match. One could say that the code changes between games would be somewhat equivalent to a human chess player that between games might confer with other notable chess experts and adjust their game play for the next games of the match, based on the advice given by those other chess experts.
Anyway, back to the crux of things that the nature of Deep Blue was that it had been based on the data of thousands upon thousands of positions and chess games, out of which an evaluation function was mathematically formulated. The evaluation function would take as input the chessboard pieces and positions and spit out what the next move should be. The evaluation function was subdivided into specialties. There were around 8,000 different segments or portions of the evaluation function, each having a particular specialty as to the chess game status.
You might liken this to having a whole bunch of chess experts sitting next to you while playing a chess match, each having a particular expertise in terms of maybe at the opening of a chess game, or during the middlegame, or during an end game, and you would confer with the appropriate specialist at the time of the game that it made sense to do so.
In addition, the program had a database of over 4,000 opening game positions and around 700,000 grandmaster games. The application had been pieced together with the assistance of various grandmasters consulting with the developers of the code. During a chess match, the code would do a look-ahead to try and ascertain various moves and counter-moves, which is referred to in game playing as levels of ply.
Generally, the deeper that you look ahead at moves and counter-moves, the better off you will be in terms of making a good move right now. If one chess player looks only at say one or two moves ahead, they might not realize that at move number three or four they are going to get trounced. Meanwhile, if the other chess player can imagine ahead to a level of three or four moves, they might be better off and know what they can do to trounce the player that only looked ahead one to two levels. Novice chess players often are only able to “see ahead” perhaps one to two ply, while grandmasters can presumably envision many ply ahead.
Deep Blue was setup to consider at times six ply to eight play ahead, and in other cases look ahead at 20 ply or more. I’m guessing that some of you might be wondering why you would not always look ahead as far as possible, maybe looking ahead to the very end of the game. In essence, when at any given state or position of the chess game, why not try to imagine all of the moves and counter-moves that would lead to the end of the game and therefore you could anticipate whether the move you might make now will lead to you winning, losing, or earning a draw.
This brings up a point I had been making earlier about time.
If you had unlimited time to make a choice, you could presumably try and figure out each and every move and counter-move that might arise. When people used to mail their chess moves to each other, you might have days or weeks to ponder the moves and counter-moves. During chess competitions that are timed, you only have so many minutes or seconds to make your choice, thusly you need to bound how far ahead you are imagining the game play, since the imagining aspects take up precious time.
When you consider that the game of chess has a chessboard of 64 squares and there are 16 chess pieces per player, and each chess piece moves in certain ways, the number of potential moves and counter-moves can be a quite large number. A famous mathematician named Claude Shannon calculated that the game-tree complexity of chess was around 10 to the 120th power as a conservative lower-bound (see his 1950 paper entitled “Programming a Computer for Playing Chess”). There is a chart that some use to depict this by saying that after each player in a two-player chess match has made only 5 moves each, the number of potential possible games that could arise henceforth from that position is around 69,352,859,712,417.
In short, there is both good news and bad news about trying to look ahead in a game like chess.
I’ll give you the good news first. The good news is that since the game has a defined and finite number of boards positions, and a finite number of pieces to be played, plus there are rules that define legal moves versus illegal moves (you cannot make illegal moves), there is presumably a finite number of potential moves that can occur. Some estimates put this upper bound at around 10 to the 50th power or something similar. I mention this because in some games and other venues we might not have any end in site as to the number of potential moves and therefore be fighting against trying to figure out something that essentially can never end.
The bad news is that the moves space is vast enough in chess that you are unlikely to be able to have the time to consider all the potential moves ahead. You need to therefore look ahead far enough that you can, as allowed by the time provided, and hope that by looking that far ahead you’ll be making a better decision now, versus not having looked that far ahead. The moves and counter-moves are usually portrayed as a tree-like structure, branching out for each of the moves and counter-moves. You need to be mindful of the time allowed in terms of doing a search through the tree.
Speaking of time, moving forward in time to today’s game playing applications, AlphaGo Zero and Alpha Zero have departed from the Deep Blue kind of coding that was dominant during the 1990s and into the early 2000s. The older method was to create an elaborate evaluation function, which I’ve mentioned Deep Blue had, and do so via a smorgasbord of handcrafted human provided tweaks and twists. The search space for the tree search was relatively large and not especially confined, and the algorithm used to do the tree search was the alpha-beta search approach.
The alpha-beta tree search uses two key factors, called alpha and called beta, which are used during the search through the tree that represents the various moves and counter-moves ahead in the game. Alpha is used to represent the minimum score that the player seeking to maximize their score will get depending upon which move they might make, while beta is the maximum score that the other minimaxing player would get. In a simplified manner, if you walk through the moves and counter-moves on a pretend basis, I would want to maximize my chances of winning while you would want to minimize my chances of winning. Therefore, at each move, I try to pick the maximum winning choice, and you would counter with picking the minimum winning choice for me.
This is a popular way to walk through a search space and it is known as the minimax approach. Alpha-beta augments the minimax approach by including a pruning feature. Essentially, the pruning involves opting to no longer pursue a particular path of the tree if it is considered unlikely to offer any viable advantages. This is helpful because it can cut out swaths of the tree.
It would be as though you are standing in your backyard looking at a massive tree and trying to decide how to climb to the top of it. You might have many branches that you can try. Upon closer inspection, suppose you realize that there are branches that seem unlikely to reach the top or otherwise are not advantageous to use, and thus you “prune” those branches and no longer give them consideration. This will reduce your effort of trying to determine which branches are worthy of closer attention.
The use of alpha-beta tree search and the evaluation function was considered state-of-the-art as to an AI-based set of techniques to use for game playing, and what made it feasible for Deep Blue was the use of parallel computing to help out. An RS/6000 computer with 30 nodes and with 480 chess-devoted VLSI processor chips was used to run the program. The code was primarily written in C and the OS was AIX. This was considered a supercomputer at the time and could assess around 200 million positions per second.
For my article about today’s exascale supercomputers, see: https://aitrends.com/selfdrivingcars/exascale-supercomputers-and-ai-self-driving-cars/
Today’s AlphaGo Zero and the newer AlphaZero have shifted away from the use of an elaborated evaluation function that was coupled with the use of the alpha-beta tree pruning algorithm. Instead, the latest approach consists of using a Deep Learning reinforcement algorithm based on Artificial Neural Networks (ANN), and coupling this with the use of the Monte Carlo Tree Search (MCTS).
In brief, it is a large-scale neural network that is considered “deep” because it has a multitude of layers and many “neurons,” and it uses “reinforcement learning” in the sense that it does self-play and rewards itself or penalizes itself based on what happens during the self-play (that’s called reinforcement), leading to it being able to adjust the neural network accordingly for future play.
For my article about Machine Learning core aspects, see: https://aitrends.com/selfdrivingcars/machine-learning-benchmarks-and-ai-self-driving-cars/
For ensemble Machine Learning, see my article: https://aitrends.com/selfdrivingcars/ensemble-machine-learning-for-ai-self-driving-cars/
For federated Machine Learning, see my article: https://aitrends.com/selfdrivingcars/federated-machine-learning-for-ai-self-driving-cars/
For the importance of explanation-based Machine Learning, see my article: https://aitrends.com/selfdrivingcars/explanation-ai-machine-learning-for-ai-self-driving-cars/
Monte Carlo Tree Search (MCTS) Aspects
The Monte Carlo Tree Search involves once again creating a tree of the moves and counter-moves, but it does so on an expanding basis, meaning that it tries to avoid having to construct an entire search space and only construct the portion that has promise. The “Monte Carlo” part of it has to do with selecting a random sample of the search space, in a sense it is making a gamble about which part of the subtrees to explore (just as though you have gone gambling at a casino in Monte Carlo!).
Why is the Monte Carlo Tree Search attractive over the use of the alpha-beta pruning algorithm or similar kinds of approaches? Here’s why it is handy for game playing like chess:
- Does not need nor use an explicit evaluation function (so, no more handcrafting, as was required in the case of Deep Blue, and avoids the human-laden aspects of getting the application up-to-speed).
- Monte Carlo Tree Search does not need what is referred to as a “developed theory” about how to play the game being considered and instead applies generally to game playing of most kinds (thus, this can be applied to chess and other games such as Go, Shogi, etc.).
- You can halt the MCTS at any juncture of its effort, while it is assessing the next moves, and you will still have a viable result that can be used (versus with other techniques you need to let them run until they fully complete otherwise you have nothing particularly useful in-hand about what to do next).
- The search time by MCTS should be lessened than that of alpha-beta pruning because of the short-cuts used, though this is not to say that MCTS will be “perfect” and so you are also taking a risk or willingness to have it prune something that might turn out to be significant.
What’s interesting too about AlphaZero is that it uses the neural network to figure out on its own the proper settings of itself, based on perhaps hundreds of thousands of self-played games, rather than having a human handcrafting the code. This included that the approach was able to “discover” aspects such as opening moves that seem good to use, versus if it had been fed thousands of already prescreened opening moves that were hand selected for it to use.
I think this is sufficient to cover the essentials of today’s chess game playing approaches versus those of yesteryear.
Though the newer approaches are impressive, I don’t want you to infer that they are more capable than they really are. I’ve seen some pundits gushing with enthusiasm that say that the AI-based techniques now “understand” how to play chess. Hogwash. If you are suggesting that these techniques are the equivalent of human “understanding” then you’ve got to be able to also explain to us how humans understand how to play chess. As I’ve mentioned previously, no one knows as yet.
I would say that hopefully we are on a solid path towards improving how we develop AI systems and that those AI systems will continue to be improved in their performance. Chess provides a handy laboratory, as it were, within which we can tryout different AI approaches and push the boundaries of what AI consists of.
For more about deep learning, see my article: https://aitrends.com/ai-insider/imitation-deep-learning-technique-self-driving-cars/
For the notion of possibly starting over with AI, see my article: https://aitrends.com/selfdrivingcars/starting-over-on-ai-and-self-driving-cars/
For the topic of the singularity, see my article: https://aitrends.com/selfdrivingcars/singularity-and-ai-self-driving-cars/
For the Turing test and how we’ll know if we’ve achieved intelligent systems, see my article: https://aitrends.com/selfdrivingcars/turing-test-ai-self-driving-cars/
The Driving Task and Chess AI Techniques
I’ve laid the foundation of the nature of today’s chess playing AI techniques and I’d now like to explain how this dovetails into the arena of AI self-driving cars.
We’ll begin with the sensors of AI self-driving cars. There are a myriad of sensors such as cameras for capturing images and video, there are radar sensors, ultrasonic sensors, LIDAR sensors, and so on. The data collected by those sensors needs to be assessed and interpreted. Based on the assessment and interpretation, the AI system will then be able to figure out what needs to be done next in terms of driving the car.
Does an image that was just captured contain a car in it? Is the car near to the AI self-driving car or far ahead of it? Is there a pedestrian in that image? Is the pedestrian near or far away? Somehow, the sensor detection and interpretation aspects of the AI self-driving car need to discern what kinds of objects are out there surrounding the AI self-driving car.
This sensory input and interpretation are happening each and every moment that the AI self-driving car is underway. It needs to be undertaken in real-time. If the detection and interpretation take too long, the AI might not have altered the course to avoid hitting say another car that has suddenly come to a halt in front of the AI self-driving cars. The sensory data interpretation also needs to be done with a great deal of accuracy in the sense that if the detection and interpretation fails to identify a car ahead or a pedestrian standing there in the street, it could be a life-or-death consequence.
For how street scene free-space detection works, see my article: https://aitrends.com/selfdrivingcars/street-scene-free-space-detection-self-driving-cars-road-ahead/
For how LIDAR and other sensors function, see my article: https://aitrends.com/selfdrivingcars/lidar-secret-sauce-self-driving-cars/
For predictive scenario modeling and AI self-driving cars, see my article: https://aitrends.com/selfdrivingcars/predictive-scenario-modeling-self-driving-cars-seeing-future/
For what I refer to as omnipresence and AI self-driving cars, see my article: https://aitrends.com/selfdrivingcars/omnipresence-ai-self-driving-cars/
By what AI-based technique or approach can we use to be able to do this kind of detection and interpretation of the sensory data? Furthermore, it needs to be fast so that it works in the real-time constraint confronting a self-driving car. It also needs to be relatively reliable and accurate, otherwise there is going to be untoward results.
There is no magic that will make this happen. We need to use whatever AI techniques or approaches that can be identified and will work best for this need.
To-date, much of this interpretation is done via the use of Artificial Neural Networks (ANN). A neural network is trained beforehand to identify objects in say images or radar data or whatever, such as finding cars, light posts, pedestrians, and the like. This trained neural network than is loaded on-board of the processors in the AI self-driving car and takes as input the raw sensory data, perhaps being transformed somewhat by other routines first, and then tries to identify the objects in the data.
Having the ANN “learn” during the act of the AI driving the car is rather chancy right now, since it could be that the neural network mistakenly learns something that is real-world untoward and then misleads the AI system by duping it. Instead, the ANN is typically prepared beforehand and pushed as a kind of executable into the on-board systems. Data collected from the sensors might be uploaded via OTA (Over The Air) electronic communications to the cloud of the auto maker or tech firm, and further ANN refinements might be undertaken at the cloud level, and then pushed down as patches or updates into the on-board ANNs.
For more about OTA, see my article: https://aitrends.com/selfdrivingcars/air-ota-updating-ai-self-driving-cars/
For the debugging of AI self-driving cars, see my article: https://aitrends.com/selfdrivingcars/debugging-of-ai-self-driving-cars/
For the reverse engineering of AI self-driving cars, see my article: https://aitrends.com/selfdrivingcars/reverse-engineering-and-ai-self-driving-cars/
For the dangers of code obfuscation, see my article: https://aitrends.com/selfdrivingcars/code-obfuscation-for-ai-self-driving-cars/
When you consider chess playing such as the use of Deep Learning with reinforcement, coupled with the Monte Carlo Tree Search, we can use the same AI techniques for doing the sensory data assessment and training for the deep ANNs for an AI self-driving car. This can be done in the backroom, so to speak, when preparing the ANNs for use on-board the self-driving car.
There is also the possibility of doing so in real-time while the AI self-driving car is underway, depending upon the boundaries that we put around the scope of the learning and also the speed at which it can perform. This is ongoing research.
I want to clarify that when I’ve presented this at conferences, there are some that initially think I’m referring to the idea that the chess playing system might be say using a camera that is pointed at the chessboard and doing image capture of the chessboard and where the chess pieces are. Nope. That’s too easy. I’m referring to the chess playing Deep Reinforcement Learning (DRL) going on related to the chess moves and chess playing.
Which takes us to the next aspect of an AI self-driving car, namely the sensor fusion. During the sensor fusion, there is an attempt to bring together the various sensors and their interpretations and reach conclusions or at least estimations of what objects are being detected via the sensors. Once again, we can use the DRL and MCTS to help with this aspect.
The AI self-driving car then updates the virtual world model which indicates the status of the objects surrounding the self-driving car. It also has attributes about those objects such as whether they are in motion or stationary, where they are most likely headed, their speed and projected speed, etc.
This then takes us to the heart of the AI self-driving car, or some might say the “brain” in that the AI action planning is where the crucial analyses occur about the status of the self-driving car and what the next actions will be. Note that I put the word “brain” in quotes because it is not at all like a human brain and I don’t want anyone to infer from my use of the word that I am somehow implying it so.
Remember earlier herein when I took you through my chess playing metaphor for the driving of a car?
Well, this is where we can especially make use of the chess playing AI techniques, now that we are discussing herein the stage of the self-driving car processes that involves trying to ascertain what driving moves to make. Similar to my mental model of perceiving my car as in the middle of a chessboard and that each driving action was a kind of chess move, so too we can consider the AI self-driving car to be doing likewise.
The AI action planning portion of the self-driving car needs to incorporate the latest status of the surroundings as exhibited via the virtual world model. The virtual world model is kind of like a souped-up chessboard that indicates the pieces of the driving surroundings and what their status is. The AI has to take a look at the present state, and using the driving strategies and driving tactics, derive the actions that need to be taken next. The AI will then issue driving command controls to the self-driving car accordingly.
This is of course happening in real-time. Just as in chess there is a time constraint, likewise there is a time constraint for the AI of the self-driving car. It cannot try to explore all possible ways in which to next move the self-driving car. As mentioned earlier, the real-time constraints in driving are more severe than chess playing, both in the amount of time allowed to make a decision and also the consequences of making a wrong or bad decision.
In my example earlier of my trying to get ahead on the freeway, I considered the cars immediately near me and those just a few ahead and behind me. Had I also looked further up ahead and behind me, I might have had many more cars to consider in terms of my possible moves and their possible counter-moves. I essentially trimmed my mental search space by confining my move calculations to those cars directly near me (was I using alpha-beta in my head or was I using MCTS, don’t know, you tell me!).
Rather than being a chess playing metaphor in the mind of a human driver, we can embody the same kinds of principles into the AI self-driving car system and particularly in the AI action planning element.
Consider these aspects about my daily drive to work:
- Chess strategy => Driving Strategy = overall approach to driving the self-driving car, style of driving, overall journey goal, etc.
- Chess tactics => Driving Tactics = moment-to-moment driving actions such as switching lanes, taking an exit ramp, allowing another car into my lane, etc.
- Opening game => Driving Journey Start = backing out of the garage, driving down the local street, making way to the freeway
- Middlegame => Driving Journey = getting onto the freeway, navigating and maneuvering while on the freeway, getting off the freeway, etc.
- Endgame => Driving Journey End = nearing my office, local streets, driving into the parking lot, finding a parking spot, parking.
Some of today’s AI action planners take a rather simpleton approach to driving the self-driving car.
For example, they look only for lane markers to determine where the self-driving car should be positioned, and then do a follow-the-leader kind of driving involving following whatever car is directly ahead of the self-driving car. I refer to this as the pied piper approach.
See my article about the pied piper approach to AI self-driving cars: https://aitrends.com/selfdrivingcars/pied-piper-approach-car-following-self-driving-cars/
See my article for garage related parking of AI self-driving cars: https://aitrends.com/selfdrivingcars/ai-home-garage-automatic-parking-self-driving-cars/
For the need of self-awareness AI, see my article: https://aitrends.com/selfdrivingcars/self-awareness-self-driving-cars-know-thyself/
For imitation as a deep learning technique, see my article: https://aitrends.com/selfdrivingcars/imitation-deep-learning-technique-self-driving-cars/
Those simpleton approaches are not going to get us to a true Level 5 AI self-driving car. They are only stopgaps on the way to getting there. If we stay with just those rudimentary approaches, there is little hope of achieving a true AI self-driving car.
In that sense, just as we have progressed from the AI techniques used in the 1990s that served us well then for aspects such as Deep Blue winning at chess over a grandmaster, today we have advanced toward the DRL and MCTS along with faster hardware, all of which allows even greater levels of play and at preferably faster speeds.
The same kind of incremental advances are going to be evolving as AI self-driving cars are improved in terms of the autonomous driving capabilities. AI techniques will be pushed forward by the desire to achieve true AI self-driving cars. The advent of AI self-driving cars will be pushed forward by improvements in AI techniques and processing power. There is a synergy between AI and the aspects of AI self-driving cars as a kind of application of AI.
I’ve also pointed out the synergy between chess and AI, in which AI pushes forward and we can see it take place via chess as an application, and likewise the playing of chess advances because of those advances that perchance happen in AI.
I’m an advocate of finding the synergy between all three, namely chess, AI, and self-driving cars. That’s the kind of moves needed to get AI self-driving cars to become real competition with human driven cars. As an old Chinese proverb says, life is like a game of chess, changing with each move. We need to keep changing up the AI for reaching the vaunted goal of true AI self-driving cars. That move makes sense.
Copyright 2019 Dr. Lance Eliot
This content is originally posted on AI Trends.