The best computer in the world today is a small machine about the size and consistency of a ripe cantaloupe. It can digest, evaluate and extrapolate more data than the most sophisticated hard-metal device yet evolved and can do it quicker and better. The huge computer complex—a machine that takes up more than a thousand feet of floor space—has one advantage over the little one. It has a better memory.
Both types of machine are used in modern professional football, and next week they will be working overtime as the combined National and American football leagues meet to draft this year's crop of eligible players. The little machines—the cantaloupes—rest in the skulls of the coaches and scouts of the game. The big one—the computer machine—accepts the data given it by the little ones, analyzes it, shuffles through its memory bank and returns black and white judgments to the brains for further evaluation.
In professional football the use of the computer has proliferated enormously during the last five years. The trend began with the escape of a general manager from a professional football team to a short term as an assistant to CBS Sports Director Bill MacPhail. It grew with the immigration of an Indian statistics expert to the U.S. and reached fulfillment when a young man who had made his living taking pictures of newborn babies in Milwaukee hospitals gave up his job to follow his hobby. The three together—led by the ex-CBS executive—easily developed the most intelligent scouting system in all sports.
Tex Schramm, formerly general manager of the Los Angeles Rams and now president of the Dallas Cowboys, decided upon computerized consideration of football players while he was associated with CBS. The Rams, during the years Schramm worked for Owner Dan Reeves and luxuriated in what was then by far the most efficient scouting system in pro football, consistently came up with the best draft in the National Football League and just as consistently lost to other teams that grabbed their discards. Deluged with fine young talent in those years, the Rams tended to drop ripening players in favor of bringing in the new ones.
"While I was with CBS, I thought the whole thing out very carefully," Schramm said the other day. "I decided that I had undervalued experience and overvalued youth. And I decided, too, that I would have to find an objective method of deciding on the worth of a football player when I went back into pro football. The only defect in the Ram scouting system was that the people involved all had built-in prejudices of one sort or another. I thought we had to find a way to judge players without emotion. We used computers to figure scores and standings when I was in charge of CBS coverage of the Winter Olympics in 1958, and I discussed using computers to evaluate football players with IBM experts then. But I didn't get a chance to put the idea into operation until 1962, when I was with the Cowboys."
As examples of what Schramm means by emotional judgment, he admits that for years he has been partial to speed to the exclusion of other qualities when judging the ability of a player. "If a guy can run a 9.4 hundred," he says, "I'll overlook a lot of faults. Some coaches have built-in prejudices against small colleges, and some coaches feel that a Big Ten player automatically is good. There are prejudices for and against regions and for and against individual coaches. These prejudices all lead to inaccurate judgments."
Restored to football in 1960, when Clint Murchison bought the Dallas franchise, Schramm hired Photographer Gil Brandt of Milwaukee as his chief scout and installed a detailed and expensive scouting system. Because there were so many other details to be mastered, it was not until 1962 that he began to solve the problem of objective analysis. In that year the Cowboys were approached by a subsidiary of IBM, Service Bureau Corporation, which was trying to develop a market in handling pro football accounting systems. Schramm countered with the suggestion that SBC try to develop a method for applying computers to the multiple problems of scouting. Eventually SBC sent an Indian—Salam Qureishi—to Texas to look the situation over.
"By the time we had organized and set up the Cowboy scouting system and it was operating efficiently I found another problem," Schramm said. "The more efficient we were the less efficient we became. We were gathering too much information on too many players. We would start with, say, 2,000 players in their freshman year in college and steadily accumulate information on them. By the time they were seniors the number was down to 500 or 600. That total was reduced to 300. Then each of the 300 was ranked from one to 300. Since it took a man at least an hour to read and evaluate the information on a player, it became obvious at once that no one could judge the 300th player as efficiently as he did the first.
"Then the individual rating varied by the qualities the rater considered most important, consciously or unconsciously. For instance, Tom Landry [the Cowboy head coach] tends to give priority to character. A player who gets a good character rating gets special consideration from Tom, even if he may fall short in other areas. When I saw the mass of information we had to digest and the difficulty of getting uniform reactions from the people who had to digest it, I knew we had to find a quick, dispassionate judge. The computer was the answer."
Schramm immediately began work with Qureishi, who was employed by the Computing Sciences Division of the Service Bureau Corporation. The upshot of their discussions was a hush-hush contract between the Cowboys and SBC to develop computer analysis of the qualities that make up a good football player.
Qureishi, who was born in Aligarh, India and took his B.A. and M.A. degrees there, knew cricket well, but he had never participated in any sports.
"Until I was called to Dallas," he said recently, "I knew nothing about American football. I had learned to enjoy baseball because of its similarity to cricket. Now I think American football is easily the most scientific game ever invented."
Qureishi set out quickly to learn about football, but his first steps were stumbles, according to Schramm. "We had an Indian who knew absolutely nothing about football and coaches who knew nothing about computers and less about Indians. Luckily, Landry is always looking for a better way to do things. If he had not wanted to cooperate, we never could have succeeded."
In long and difficult sessions with the Cowboy coaches Qureishi found himself concentrating almost exclusively on what made a good player. The coaches came up with a staggering total of 300 or so variables affecting their judgment of talent.
"At that time," Qureishi says, "the most sophisticated computer system could work with something like only 80 variables. It was immediately evident that we would have to cut down. We reduced everything to five dimensions. But there was a problem of semantics. We had to make sure that the scouts and coaches all meant the same thing when they analyzed a player. We had to find key words that, as much as possible, said what we wanted to know and what the coaches and scouts wanted to say."
The five intangibles were character, quickness and body control, competitiveness, mental alertness and strength and explosiveness. Three other qualities—weight, height and speed—were physically measurable, and no evaluation by a coach was necessary.
"You get down this far, then you have to have an accurate measure of all of these qualities," Schramm explains. "I mean, you ask a coach a general question about any one of these qualities and you get an answer that is practically meaningless. For instance, we used to ask how quick a player was. One coach said he was quick as a cat; another said he was quick as two cats. We had to ask hundreds of questions, trying to find the key phrases that were meaningful both to the coaches and to us."
Now the Cowboys' scouting questionnaire, which is filled out by all of their many scouts, is a rather simple form, with 16 options, all of them in the form of statements. The scout grades a prospect from one to nine on each statement, depending on how well he fits the description.
"We have discovered that the human mind is not capable of judging degrees on a scale with more than nine ratings," Qureishi explains. "I mean, you cannot say that this man is one-twentieth more agile than that one or one-twentieth more competitive. So we designed our grading system to fit into the scale of the mind."
The Cowboy questionnaire provides declarative sentences describing a facet of football skill and asks the scout to decide, from one to nine, how the candidate fits the description. For instance, one sentence reads: "His movement is awkward in wave drill." A wave drill is an agility exercise familiar to all coaches, and it means the same thing to Robert Smith at Southern University as it does to Woody Hayes at Ohio State. So when a coach or a scout is asked to rate a player on his awkwardness, or lack of it, in wave drill, he has something concrete to go by, and his rating is far more meaningful than "Quick as two cats." It is also usable in a computer.
Aside from the eight basic qualities, the scout must rate a player on the specific skills of his position. For offensive ends and flankers, for instance, there are eight such dimensions: receiving short, receiving long, avoiding being held up, faking and cutting ability, running ability after catch, ability as a blocker, catching in a crowd and hands. Each word, or combination of words, has a clear-cut, definite meaning to any scout. If he gives a player a rating of nine this means, in the Cowboy lexicon, that he has rare ability. Five would mean above average, and the lowest grade, one, is poor. Finally the raters must decide on the likelihood of a particular player making it in the National Football League. Nine in this category means a cinch, and one means none. Ones are seldom drafted.
It took more than three years to arrive at the dimensions that define a pro football player and to winnow down the definitions to a usable few, but then Schramm and Qureishi found another troublesome problem: how to assay the qualities of the scouts who were feeding information into the computer.
One scout may tend to overestimate the value of a player, another to be too harsh in his judgments. Schramm, Brandt and Qureishi worked for years going over scouting reports and assigning a value to each scout. This value—or weight—then was fed into the computer and forthwith became an integral part of its memory.
Recently the Cowboys joined with the San Francisco 49ers, the Los Angeles Rams and the New Orleans Saints in setting up a systems consulting company. Qureishi quit IBM last spring and became president of Optimum Systems, Inc. on August 16. Optimum Systems, at the moment, primarily operates as the computer division of the scouting system of the four clubs, although it has recently acquired contracts in other fields. If a computer can help decide the relative merits of football players, then it may also be able to predict who will be a good astronaut, jet pilot or comptroller.
Los Angeles and San Francisco became part of the Cowboy computerized scouting project in 1963 almost at the beginning, the Saints last year when they joined the league. "I'm no philanthropist," Schramm says frankly. "I wasn't trying to help the other clubs, but we needed a bigger sample for our computers. We weren't getting enough reports from just our own scouts. Now we get three or four times as much information, and that reduces the margin of error."
When Schramm originally met with Qureishi and the officials of SBC, Qureishi, after having had the problem explained to him, was asked how accurately he thought a computer system could operate on the given information.
"Oh," he said airily, "I should think to about 95% accuracy."
Since scouting systems obviously do not operate with anything like that kind of efficiency, Schramm was very skeptical. When Qureishi came to understand the complexities of evaluating human capabilities he was not quite as sure of the 95% accuracy, either.
During the five years the Cowboys have been using the computer on scouting, they have refined their methods more each season. Weights now are assigned not only to the individual scouts, but to the type of scouting material (i.e., whether the report comes from observation of movies, from a report from a staff member in the field or from an area scout hired by the club), the school from which the player comes and the type of competition he operates in.
"The reports are getting better each year," says Schramm. "Our scouts are getting better, for instance. They have seen their ratings—too harsh, too soft—and they have tended to improve so that the reports are getting more standardized and more accurate."
When all the information has been fed into the machine the Cowboys—or the 49ers, Rams or Saints—readily can extract what they want. Under the old system, each club had hundreds of black loose-leaf notebooks crammed with data. The process of extracting from this formidable array a list of, say, the 10 best defensive tackles was so time-consuming that it had to be started in the spring before a player was to become eligible. Updating at draft time was impossible.
"Now we can get a list of the top hundred in a day," Schramm says. Still, the manual system is retained as a check on the computer. If a computer comes up with a surprising rating, the coaches can go back to the notebooks and discover exactly what each scout had to say about the player in question.
To test their computer system against known quality, the Cowboys evaluated the 1964 draft against the weights being used in 1967. The New York Jets and Sonny Werblin doubtless will be happy to know that the computer agreed with their $400,000 price tag on Joe Namath. Namath came out No. 1 of the 1964 crop, although Dick Butkus, now the middle linebacker for the Chicago Bears, and Gale Sayers, the brilliant Bear halfback, had higher numerical scores. On that part of the scale where 900 is perfect, Butkus scored 854, Sayers 851 and Namath 803.
"Namath rates ahead of them anyway, because he had qualities that were held in particularly high esteem by this model. He had individual qualities that outweighed certain aspects of the 900 scale," Qureishi explains. "On another model with another set of requirements Butkus or Sayers might have come first."
Of course, it did not take an IBM computer in 1964 to know that Namath, Sayers and Butkus were all extraordinary pro prospects. Of the 100 players rated by the computer in this experiment, 87 became pros. For the record, here is how the computer rated the first 15 prospects in 1964:
1) Joe Namath, 803, starting quarterback, New York Jets; 2) Dick Butkus, 854, starting middle linebacker, Chicago; 3) Gale Sayers, 851, starting halfback, Chicago; 4) Fred Biletnikoff, 776, starting flanker, Oakland; 5) Mike Curtis, 801, starting linebacker, Baltimore, now on injured reserve; 6) Steve DeLong, 764, defensive lineman, San Diego; 7) Clancy Williams, 772, starting defensive back, Los Angeles; 8) Roy Jefferson, 749, starting flanker, Pittsburgh; 9) Tucker Frederickson, 750 (rated as defensive back), starting fullback, New York Giants; 10) Fred Brown, 719, starting defensive lineman, Philadelphia; 11) Ralph Neely, 725, starting offensive tackle, Dallas; 12) Dave Simmons, 740, linebacker, St. Louis (now with New Orleans); 13) Jack Snow, 739, starting flanker, Los Angeles; 14) Larry Elkins, 728, flanker, Houston; 15) Craig Morton, 725, quarterback, Dallas.
"We get at least 50% more information on each player now than we did in 1964," Schramm points out. "About 112 rookies made it in the NFL last year. That averages out to about seven players per club, or 175 players per year for the 25 teams in both leagues. We want to be able to pinpoint those 175 players and avoid wasting draft choices on the hundreds of others who will be drafted and will not have the ability to play professional football."
Although the four teams sponsoring Optimum Systems, Inc. completely share the computer, they do not completely share information with one another.
"About 40% of the information we get on players is common knowledge," Schramm explains. "The other 60% the clubs keep to themselves." Brandt, in fact, often wants to keep back more than 60%. An intense, exceptionally dedicated man, Brandt sometimes works straight through the night on his scouting projects. His complete devotion to his job and to acquiring football players for Dallas probably has made him the most thoroughly disliked scout in the business—by the other scouts.
"That shows he is doing his job better than any of the others," Schramm says, grinning. "If the other scouts liked him, I'd be worried. The players like him. He has a real knack for meeting players and coaches on their own ground and gaining their respect."
Brandt himself never played or coached football. While he was making his living taking photos of newborn babes in Milwaukee, he scouted football players as a hobby. He attended as many games as he could for more information, and each year he held a mock draft for his own edification. He worked for Schramm when Schramm was with the Rams, getting information on players in the annual All-Star games.
"I asked him to sign some ballplayers for me when I took over the Cowboys," Schramm said. "He worked 18 hours a day signing them, and he never missed. I was impressed with his knowledge of college talent. Maybe it was a long shot hiring him, but it paid off."
Although Schramm has been a mover and shaker in the development of group scouting, he feels that it has problems. "One of the gravest dangers facing pro ball is the growing tendency to share what should be competitive jobs," he says. "You can't share in anything that bears on the competition on the field. In our scouting combine we remain highly competitive in all but the purely mechanical aspects of scouting. There are other groups now, and they are sharing totally, or tending that way, and that's wrong. When you do that you abrogate your basic responsibility to compete. I have grave misgivings about total sharing of scouting information."
As an example of how diverse the information available to the Rams, Cowboys, 49ers and Saints can be on a particular player, Schramm cites Clinton Jones, the Michigan State halfback.
"San Francisco had 21 scouting reports on Jones," Schramm says. "We had 24, and only six of the reports were common to both clubs."
At present the repository for all this information is a collection of magnetic tapes resting in the offices of Optimum Systems. Optimum is using two computer centers, one at the American Institute for Research, a computer complex perched in the hills near Palo Alto, Calif., not too distant from its offices, and the other at SBC. The computer in use at AIR is one of the most sophisticated of the IBM units. It leases for about $35,000 a month and is used for many projects other than scouting for the pros.
Not long ago Qureishi and Schramm visited the computer to extract some information. Qureishi walked over to the biggest machine in the room, a box which operates as a sort of mammoth typewriter, printing out the information requested from the computer. It was burping information for another customer at the time.
"We have space optioned in our building for our own computer," Qureishi said. "When we get a few more customers, maybe we will lease it. I would like an IBM System 360, Model 67. It is a more sophisticated system than this one."
Schramm regarded Qureishi through cool blue eyes.
"How much does that rent for?" he asked.
"Ninety thousand dollars a month," Qureishi said. "But it can handle so much more than this one, Tex."
Qureishi and each of the four clubs own a fifth of Optimum Systems. "I hope you become a very wealthy man," Schramm said. "But we aren't any of us wealthy yet. Wait."
Undoubtedly Optimum Systems will acquire its own computer—starting with one of the modest $35,000-a-month models. When it does, there will be two sets of customers it will never service. They are BLESTO and SEPO, the scouting combinations that include all the rest of the clubs in the National Football League. BLESTO and SEPO joined the computer parade this year, signing a contract with Computer Applications, a company based in Silver Spring, Md. Computer Applications will operate somewhat as Optimum Systems does.
Computer Applications also will work with member teams on offensive and defensive game analysis, or game scouting. The game-scouting phase of computerized football got its start about four years ago with Bill Witzel, a Maryland alumnus who was an employee of Computer Applications at the time. As a hobby, Witzel went to the Maryland coaches and took the information they could give him on Maryland opponents and programmed that into the computer, coming up with frequencies, formations and habits of the upcoming teams. He was unable to sell his idea to Tom Nugent, the Maryland head coach, so he tried Bill McPeak, then head coach of the Washington Redskins. McPeak turned him over to Ed Hughes, the Redskin defensive backfield coach, and together Hughes and Witzel worked out a computer program for game analysis. That year—1965—the Redskins finished second on defense in the NFL, despite a so-so won-lost record.
Hughes is the brother-in-law of Dick Nolan, the Cowboy defensive coach who last week was named the new head coach of the 49ers. Nolan followed Hughes's lead and installed much the same system for the Cowboys. "The computer does about 10 hours' work for you in 30 minutes," Nolan explains. "It will analyze the other team's offense much more quickly than we can do manually. It will tell you what the other club has done inside its own 10-yard line or inside yours, what it does from one hash mark or the other, what it will do inside its own 30 on third and short, third and long, and so on. You can get frequencies on any situation you can imagine."
The use of computers in game analyses and in the creation and adjustment of game plans is only getting started. When the Cowboys do set up their own computer machines in Palo Alto, it is conceivable that they will be able to use them in order to decide what strategy to pursue during the game itself.
"All we would need would be a terminal in the press box," Nolan says. "We could feed the first half into the computer play-by-play, then, while the teams are walking off the field after the second quarter ends, we could ask the computer in Palo Alto for an analysis of the other team's offense and find out if they were going away from the trends we expected. We could find out where they were hurting us most, which holes they were hitting. We could analyze our own play and find out what had been successful and what had failed and what defenses they had used most effectively during the first half. The computer could print out all that information fast enough for us to use it for adjustments during the half-time interval."
The Cowboy and Redskin computerized defenses met for the first time this year, with Hughes and Nolan basing their defenses, for the most part, on what the machines had revealed to them. As you might expect, it was a whale of a defensive struggle, with the Cowboys winning on the last play of the game. On that play Don Meredith threw a 36-yard scoring pass to Dan Reeves, the Cowboy halfback. Reeves was all alone and had ample time to score, even though he had to wait for Meredith's wobbly pass. The Redskin computer might have been confused by a couple of things. Meredith had been knocked silly the play before and had no recollection of calling the play or throwing the ball after the game, so it would have been impossible for the computer to forecast what he would do.
As Qureishi pointed out, computers are not creative thinkers and cannot react to situations like this. No matter how sophisticated the machines get, the game will always be dependent upon human inspiration and human error.
But there is one thing to ponder—Reeves was not drafted by either the AFL or the NFL when he ended his college career at South Carolina. Not long ago Schramm ran Reeves through the computer, using the 1967 Cowboy model to find out how high the computer would have picked him.
The computer rated him 29th. That means he would have been drafted on the second round instead of having been ignored.
So the computers aren't exactly stupid, are they?