tion on the human mind regarding the speed, accuracy, and ability to correlate and intercorrelate multiple variables with all possible outcomes and treatment consequences." There is a limitation on the computer, too. Practically speaking, there are many limitations. But in purely mathematical capability, the human mind is much inferior to the computer in multiple-discriminant analysis.

This is a function vital to diagnosis. It refers to the ability to consider a large body of facts, and on the basis of those facts to assign a patient to one diagnostic category or another on the basis of probability. Consider a simple set of categories: appendicitis versus no-appendicitis. (This is a simplification of what is, practically speaking, a larger problem in diagnostic categories, but it will serve to explain the principle.) Let us assume that a surgeon seeing a patient with pain on the right side must make only this decision. How does he make it? No single piece of information will tell him the answer (except, perhaps, the fact of a previous appendectomy). Certainly such routine data as sex, age, white count, degree of fever, duration of pain in hours will not tell him. But considered all together, they permit him to arrive at a decision.

This is all very familiar. But the point is that it is not very precise. A discriminant function can be produced that weighs each variable-age, sex, white count-on the basis of how important each variable has been in the past. Thus the discriminant function has two uses. First, it can make a diagnosis and act as a consultant to the surgeon.* Second, it can determine which variables are most significant, which account for the greatest proportion of the variance, in making a diagnosis. This can sometimes be surprising. For example, one study of women undergoing gynecologic surgery indicated that the age of the patient had great diagnostic importance, and that notation of last menstrual period, biopsy of smear, pre-operative hemoglobin, and urine-sediment study were all much less important to making a diagnosis.

These applications of the computer are designed to be diagnostic aids to the doctor, rather than a replacement. One hears, for example, that "even though computers can be programmed to make diagnoses, it takes the physician to perform the physical examination, and it takes the personal contact of the doctor-patient relationship to gather the information needed for the identification process."

For better or worse, this is not quite true. History-taking, long the province of the physician and a keystone to his "art," is partially amenable to automation. The Logan station utilizes a simple program, which is not so much a conversation with the computer as a quiz. But more sophisticated programs exist, and still better ones are being planned.

*Several discriminant functions have been made and shown to be as accurate as the surgeons, though interpretation is made more complex by the fact that these functions are generated from data on previously diagnosed patients.

Some years ago, Joseph Weizenbaum of MIT designed a program to conduct psychiatric interviews. Although in some senses the program can be regarded as a "trick," it can nonetheless achieve impressive results. Dr. Andrew T. Weil has reported one interaction between the computer and a human psychiatrist as follows (the computer writes in upper-case letters; mere humans, in lower-case).

Doctor, I have terrible news

YOU SAY YOU HAVE TERRIBLE NEWS

yes, there is a computer program at m. i. t. that'11 rui n us 

WHY DO YOU SAY THAT

i have seen it working

YOU SAY YOU HAVE SEEN IT WORKING

it's frightening - you must see it

YOU THINK I MUST SEE IT

you sound depressed to me

IN YOUR FANTASY PERHAPS WE SOUND DEPRESSED TO EACH OTHER

yes - pe rhaps

I UNDERSTAND

i do too

YOU SAY YOU DO TOO

we are both in the same boat with respect to the computer

LET'S CONTINUE TO TALK ABOUT YOU - SHALL WE

you are a comfort

WHAT MAKES YOU THINK I AH A COMFORT

at a time like this you want to talk about me

WHY DO YOU THINK I WANT TO TALK ABOUT YOU

because you share my feelings about that damned computer

WHAT DOES THAT SUGGEST TO YOU

that, of course, you hate computers too

YOU THINK I HATE COMPUTERS TOO

why not, they're hardly human, are they

PERHAPS THAT'S WHAT WE ARE HERE TO FIND OUT

no computer could converse as you can, for example

OH

i can't understand how anyone could be fooled by a mere machine

CAN'T YOU REALLY

no, after all, there is nothing creative about a computer program

YOU ARE BEING NEGATIVE

i am convinced i am correct

ARE YOU WORRIED TO HAVE BEEN CONVINCED YOU ARE CORRECT

no, not in the least, not at all. not even a little

YOU ARE BEING NEGATIVE

you said you agreed with me, didn't you

IS THAT WHAT YOU HEARD ME SAY

why don't you try to dissuade me - computers diddle with numbers that 'sail

YOU WISH ME TO TRY TO DISSUADE YOU

you can't

YOU THINK I CAN'T

for example, could a computer in conversation like this raise a new topi c

YOU SEEM TO HAVE AVOIDED SPEAKING OF YOUR PARENTS ALTOGETHER

precisely as you just did YOU THINK I JUST DID 

yes, and that's what shows the difference between a man and a computer

I SEE

This kind of interaction is suggestive, at least, of the ways computers can be utilized in the future. And as the hospital reaches out into the community, these may be increasingly important. According to Jerome Grossman of the MGH Computer Sciences Laboratory, "A large part of history-taking is directed toward finding out how sick a person really is. The first question is always whether a patient needs to see a doctor at all. That's the biggest decision doctors make over the phone now-talking to the patients, trying to decide whether they're sick enough to be seen now, or whether it can wait. The patients want to know the same thing, so they spend all night or all weekend trying to get hold of the doctor, who's off duty, or out of town, or something…

"In the near future, when the home computer and television set is practical, you're going to be able to plug right into the hospital computer without ever leaving your home. The computer will flash questions on the screen, like 'Do you have a cough?' and you answer by touching the screen with your finger at the appropriate place. We've just developed a screen like this. It doesn't require any special gadgets or light pens or anything, just your finger. Touch the screen, and the information is recorded. Eventually, the computer will flash back some directions, like 'Come to the hospital immediately' or 'Call your doctor in the morning' or 'Have a check-up within six weeks,' or 'Someone will come on the screen, if further classification is necessary.' So there you have it. That first big decision-who needs to be seen-is settled by the computer, without ever having required the doctor's presence."

The idea is interesting not because it is an imminent practical development-it is not [What is imminent is the use of computer stations to take a portion of routine history and to advise the doctor on further tests. Such consoles are already in use experimentally in the MGH medical clinics and in certain private doctors' offices] -but rather that it represents a further extension of the hospital into the community-not only into clinics via TV, but into the homes of many individuals, via computer. One can argue, in fact, that those who predict the hospital's role as "primary physician" or "first-contact physician" is declining are wrong. It will, ultimately, increase with the use of computers.

Automated diagnosis is one thing; automated therapy, quite another. It is probably fair to say it is feared equally by both patients and physicians. It is also important to state firmly that the following discussion is largely speculative; automated diagnosis is in its infancy, but automated therapy has hardly been conceived. Its modern forerunners are the monitoring systems that check vital signs and the electrocardiogram. These monitors are not computers at all, in any real sense; they are just mechanical watchdogs, about as sophisticated as a burglar alarm.

At the present time, there are serious problems facing anyone who wishes to automate the therapy of even a circumscribed class or category of patient. To automate the therapy of all patients, with the full spectrum of disease, would be an enormous undertaking. Whether or not it is done will depend largely upon the demand for it, which in turn depends upon the availability of physicians. In assuming that it will be done, at least to some extent, I have also assumed that the shortage of physicians in this country will increase in the foreseeable future, necessitating a practical change in the doctor's functions.

Partially automated therapy is already desirable. The reasons are twofold. First, modern therapy makes necessary an enormous amount of paperwork; one hospital study concluded that 25 per cent of the hospital budget was devoted to information processing. The usual hospital systems for collecting, filing, and retrieving information consume great quantities of time for nearly everyone working in the hospital, from the physician who must spend time thumbing through the chart, to the nurses who must record routine data, to the personnel who work full time in the chart-record storage rooms. One consequence of the present methods, aside from the expense, is the number of

errors that occur at various points along the line. And the possible advantage of putting all data through computers is the ability to check errors. For instance, if medications are ordered by the physician through a computer, that computer can tirelessly review orders for drug incompatibilities, inappropriate dosages, and so on.

The second reason comes from experience with present monitors in intensive-care units. These monitors "watch" the patient more carefully than any group of physicians could; the patient's condition is sampled continuously, rather than just during rounds. Such monitoring has already changed many ideas about the nature of disease processes [One example: the incidence of cardiac arrhythmia following myocardial infarction is now suspected to be virtually 100 per cent; it is thus an almost certain consequence of heart attack- this is useful information since the arrhythmia are the most common cause of sudden early death from heart attack.] and it has renewed consideration of therapy at intervals. For example, most drugs are now given every six hours, or every four hours, or on some other schedule. But why not continuously, in an appropriate dose? And in that case, why not have a machine that can correct therapy on the basis of changes in the patient's condition?