Darwin Solves the Frame Problem

It has been postulated that the “frame problem” presents a theoretically-unsolvable limit to the capabilities of artificial intelligence — a boundary beyond which we cannot cross, even in theory, much less in practice. This paper explains why that is so in any single case, and suggests that Darwinism is, in fact, the only viable solution to that problem.

Originally published 2000

Understanding the Frame Problem

The nut of the frame problem is this: In any complex situation, you have to frame the problem the right way in order to achieve a solution. “Framing” the problem means identifying what is relevant and what is irrelevant, and determining which bits of knowledge in your extensive repository can be brought to bear to arrive at a solution. But to properly frame the problem, you must understand the problem. And that is a chicken-and-egg proposition if there ever was one. If you understood the problem, it wouldn’t be a problem!

As human beings, we routinely solve complex problems. Our success in that process is so frequent and so seemingly effortless, in fact, that it is easy to take the ability for granted. It is only when we begin trying to capture that expertise in automated systems (e.g. robots), and try to give them the intelligence to deal with new situations, that we discover how truly remarkable that capability is.

The root of that capability appears to be the brain’s capacity for reorganizing itself to solve different classes of problems. So, when a neophyte approaches a chess board, they think just as hard as master players (as shown by experiments) but they think in ways that are less productive. (By the same token, it has been shown that grandmasters think differently from masters! The possible moves that are “suggested” to them by a given position may often be radically difficult.)

On the other hand, no master chess player was born that way! They all started as neophytes, and reorganized their thought processes as they learned the game. In life, every single one of us followed that same pattern — learning an reorganizing our “strategies for living” during the 20 years or so that it took to grow up. We started out with very simple stragegies, but over time they became much more complex, as did the awareness of when to apply them!

Similarly, in business it is commonly accepted that “mentoring” is the best (and possibly only) method for learning the requisite organizational and people-management skills. Those skills are hard to describe in the abstract, but possible to pick up by example as one is “coached” through situations by their mentor.

In all of these cases, people are effectively reorganizing their problem-solving heuristics in order to rapidly hit on the “best” problem-solving strategy in any given situation. Currently, of course, our robotic systems tend to be “frozen”, than “self-organizing”. Once programmed, the program tends to operate in exactly the same way as time goes — the situations may change, and the results generated by the program may differ as a result, but the program never changes at all.

So clearly, a program that could “organize itself” would have the ability to address the frame problem. In other words, it could get smarter over time. But that, in itself, does not solve the problem, because any oneattempt to find a solution can still take an inordinately long time. The “Robot Example” illustrates that problem nicely.

The Robot Example

In the article, Cognitive Wheels: The Frame Problem of AI, Daniel Dennett presents a wonderful illustration of the frame problem. It goes something like this:

  • A robot, R1, is created. It’s task is to go into a room and get it’s power supply. It goes into the room, finds its power supply in a wagon, and pulls it out of the room. Unfortunately, there is also a bomb in the wagon. BOOM! The wagon and robot go up in smoke.
  • That’s not good, so a “Robot Decision-maker”, R1D1, is constructed. It is able to to think out the consequences of it’s actions in advance, so it won’t pull out the bomb with the wagon. The new robot goes into the room, and immediately engages it’s deductive engines. It deduces that the floor is under the wagon, and the wagon is under the roof, so therefore the floor is under the roof. Then it starts deducing that…BOOM! The bomb goes off before the program can complete it’s exhaustive set of deductions.
  • That’s not good either, so a “Relevancy-detecting Robot Decision-maker”, R2D1 is produced. This time, the robot won’t spend time on useless deductions, but will focus on what’s important! So it, too, is told to enter the room and retreive it’s power supply. This version sits outside the room without moving for an awfully long time, until finally… BOOM! The power supply is destroyed by the bomb. After debugging the output, it turns out that the program had figured out that the position of the floor wasn’t relevant, and the color of the drapes was relevant, and that was about as far is got in the time available. At least this time the robot was saved, but the “analysis paralysis” induced by the attempt to determine relevancy prevented the robot from taking any sort of action.

The robot we need to solve the problem, of course, is the “R2D2” of Star Wars fame — a Relevancy-detecting Robot Decision-maker that acts Decscively! But the chicken-and-egg problem that is the frame problem suggests that R2D2 is impossible to construct. Impossible, that is, if R2D2 must stand or fail on his own!

Solution: A Herd of Robots

The attempt to determine relevancy is not something that is best accomplished in real time! It takes a while to figure out what is relevant. And that is a problem in a situation where time requirements work against you, because you are virtually assured of many false starts. As a general rule, then, the search for relevancy will take sn excessively long time, because of course the question of “what is relevant” can only be answered after you have arrived at a solution.

The solution to that problem was identified by Darwin. If you have many individual problem solvers, and each is going about the problem in a different way, then odds are good that if a solution exists, then it (or possibly several) solutions will be found. And if there is a way to communicate that solution (via DNA, for example) then future generations which have that solution will prosper. So the answer appears to be: employ a herd of robots.

The need to find solutions

Suppose, for example, that there were a long of robots, each with a room to rescue their power supply from, and that each approached the problem in somewhat different ways. Some might rush in and do the first thing that occured to them. Others might sit back and think about what to do. (In design work, we do the same kind of thing, trying out different designs in our “mental space”, and eliminating those which are obvious failures.)

Most of the former group would probably be destroyed when the bomb went off. And many of the latter group would probably still be sitting outside when it happened. But somewhere in that line, there would most likely be a robot that came up with a solution.

The trick, of course, is that it is the first robot to solve the problem who wins the big prize. That’s how it works when it comes to research prizes, and that’s how it works in an entreprenurial economy. But the interesting observation is that there is no way to know whether, for any particular situation, the solution lies in being one of the faster acting robots, or in being one of the slower thinking robots. Somewhere between the extremes is right, but there is no way to know a priori, where that point is! It could be very near the front, or very near the back.

The need to share solutions

So it seems clear that you need a lot of robots to find a solution. But you also need the ability to share those solutions. Communication is the key, in fact. Whether communicated through language, by example, or by transmitting the solution in DNA, the ability to inform future generations is paramount! (Otherwise, the Darwinian solution is no solution all, as only a miniscule percentage of the population ever successfully solves the problem.)

With respect to the frame problem, the Darwinian solution tells others that these are the relevant facts, that this is the way to solve the problem. If the others have self-organizing capacity, then they add a new “situation-identifier” to their “master list” of situations, and then call up the appropriate problem-solving strategy whenever they are in that situation.

Here are a few other observations on the subject:

  • My attempts at constructing an artifical intellence program that played a board game were eminently successful. And yet the entire experience was unsatisfactory because the program didn’t make me smarter. My original goal had been to construct a program which figured out how to win, so it could teach me how to do it. Using alpha-beta pruning and a brute-force lookahed, the program was eventually able to succeed in very stages of computer competition. (And none of the people who played were ever able to beat it!) But to play that well, it had to evaluate thousands of possible moves and the resulting positions. That was nice, but it certainly didn’t inform me!

    Oddly enough, I did get better at the game, as a result of my human capacity for pattern recognition. I saw the kinds of moves in made when I was testing the program so often, that eventually began to predict where it would move. When someone asked me how I knew what move was coming, I had to reply that I didn’t know! That move was just “suggested” to me the position!

  • Similarly, I would rate the work with neural networks to be of limited value, until and unless some way is found to share the results with other programs! (One way to do that is by identifying the most efficient “presentation sequence” that produces the correct heuristics. Coaches do that when they train atheletes, for example. By putting them in situations that require particular heuristics for success, they assist the atheletes’ self-organizing capacities to set the appropriate heuristics and give them a high priority in the situation-matching chain.
  • Finally, it should be noted that humanity as a whole values those who find solutions, that we share those solutions with others, and that we frequently reward those who do the sharing.

The ability to delegate, as well as learn

Note that in any such society, delegation also becomes an option. Need a latrine? Call Joe. He knows how dig them so they work. We do the same thing in our professional lives: “Get Ned. He knows how to solve that problem.”

The ability to delegate depends only on the ability to recall who knows what we need. It still requires a bit of reorganizing to find the appropriate person quickly, but much less effort is expended than is required to learn everything Ned knows!

However, while that capability is useful in many work efforts, the alternative is equally prevalent. Frequently, we cannot depend on Ned doing the job for us, so we have to find out what Ned knows. If you’re writing a program or playing chess, for example, Ned won’t do the job for us. So you’ll call on Ned when you have a specific problem you need to solve, find out how to do it. Later on, you’ll be able to solve similar problems by yourself (and you will do so, if only as a point of pride).

In such cases, the result of the interaction is learning — an activity which might well be defined as “identifying the relevant features of the problem, and memorizing or recording the steps in the solution”.

A Meta-Framing Design

What drives humanity? The giggles we get when we solve somthing — research, crossword puzzles, adult education programs, a child’s insatiable desire to learn — all of these exemplify the spirit which drives us. We are motivated to the point of being driven to identify problems, frame them constructively, and solve them. Reasearchers and writers know well the “aha” experience — the surge of joy that occurs when you bring a new discovery or solution into the world.

One of humanity’s distinguishing characteristics then, is that we are a race of problem-solvers. In addition, we delight in sharing our knowledge with others. In general, we are equally happy when teaching or learning, and somewhat miserable if we are doing neither!

In our jobs, a job that isn’t teaching us new things isn’t “stretching” us — so we can be successful, yet unhappy. On the other hand, if we’re not succeeding at all then it’s clear we haven’t learned enough to succeed, and that makes us unhappy as well. So we need to find that middle ground, where we are succeeding but also growing — then we are truly happy. In our hobbies, we also like to learn new things. Whether music, dance, sports, or games, we delight in finding new ways to accomplish our objectives.

That fact that we are “wired” to enjoy discovery and to enjoy the communication process is the meta-principle that solved the survival problem so well for us that we now threaten to overrun the planet. As life forms randomly mutated in efforts to increase their chances for survival, the one life form (us) that was motivated by the thrill of discovery and by the joy of communicating the findings, turned out to be such an extreme competitor that there is nothing left which stands in the way of its survival, save itself.

The result, of course, is that we may have been too successful for our own good — a fact which has produced a whole new problem we need to solve. And if there is any cause for concern, it is the fact that as a planetary body, we represent an experiment of one. As a result, the frame problem may very well defeat us. In other words, we may not be able to find the answers we need in the time left before we destroy the planet and our lose our civilization for lack of critical resources like energy.

The Role of Government

With all that said, is it any wonder that America is the home of innovation? I claim that is our civil and economic liberties which account for the unprecedented degree of invention that has occurred on this contintent — a level of innovation and productivity hitherto unknown on this planet.

It is not difficult to see the correlation. Civil liberty lets people define the life styles that make them happy and productive. True, many mistakes are made. But there are many successes, as well. (A personal example: Throughout childhood, I had to get up early to go to school. And I was only an indifferent student. It took a couple of years of college before I got wise enough to ensure that none of my classes started before noon. Result: My academic achievements skyrocketed!)

A society unencumbered by religious dogma and the dicates of the majority is one in which individuals are free to find social organizations that work for them. The “separation of church and state” and the principles of civil liberty, therefore, lie at the very heart of our ability to innovate. When you add to that mix the ability of people to choose their employers, and to start their own businesses, then we add motivation and incentive to the mix, which virtually assures the result.

Of course, the capacity of unregulated businesses to inflict egregious harm on people for the sake of profit has been amply demonstrated in both the tobacco and the food-processing industries. But that fact actually reinforces the main point — people who are free are so inventive that government regulations are necessary to define the “rules of a game” in a way that ensures the results are truly beneficial, rather than merely seeming so!


The “frame problem” states that no single individual is consistently capable of identifying the relevant features of a novel, complex situation in a favorable time frame. The critical dimension in that equation is time. If time is an issue, then the organism must of necessity be “pre-programmed” to peform the task at hand.

The “law of averages” suggests that in any sufficiently large group will (unless they are marching in lock step) produce an individual who finds a solution. If that solution is then shared among the group (and the individuals in the group have self-organizing capability), then the members of the group can “pre-program” themselves for success.

Finally, it is noted any group which values both discovery and sharing (as “meta principles”) will succeed wildly, far eclipsing its nearest competition (all other considerations being equal).


  • Crockett, Larry J. The Turing Test and the Frame Problem.
  • Dennett, Daniel C. Consciousness Explained.
  • Dennett, Daniel C. Cognitive Wheels: The Frame Problem of AI.
    (A great article with the clearest exposition yet of the frame problem.)
  • Ford, Kenneth M. The Robot’s Dilemma Revisited.
  • Minsky, Marvin. Society of Mind.

Copyright © 2000-2017, TreeLight PenWorks

Please share!

Add your thoughts...

This site uses Akismet to reduce spam. Learn how your comment data is processed.