The past looks ordered, the future looks messy. That is because we create narratives with things neatly arranged, making the past look almost deterministic.
But lately it has become hard to order the recent past; progress is too fast. To many, artificial intelligence (A.I.) seems to have appeared out of nowhere. Now, we have a new tool that uses past information to craft narratives and perhaps generate new ideas. Over the last year, A.I. has been transforming the landscape of many domains we call creative. We can now ask questions and get answers, text, images, and numbers, and soon we will have systems that seamlessly integrate all three.
The implications of this technology are immense.
This highlights, now even more than before, the importance of asking the right questions. There is little value in finding the right answer to a wrong question. Asking the right question, on the other hand, could lead to creative breakthroughs — and “break-withs” — not possible before.
Creative ideas as combination of the past
Let’s start by asserting that there are two broad classes of creativity and by defining a few terms that will serve to anchor the discussion when we get to specific examples.
We can call the first class combinative creativity — the process of combining existing ideas, concepts, or elements to generate novel solutions. It capitalizes on the diversity and enormous amount of data to create something new.
Many, if not most, innovations, theories, and ideas draw inspiration from what has been done in the past. Many creative people have recognized this fact. Versions of “Originality is undetected plagiarism” and “Originality is nothing but judicious imitation” can be attributed to the French philosopher Voltaire and the poet Lord Byron. Einstein is credited, along with others, with “The secret of creativity is knowing how to hide your sources.” Thus, while we may romantically think we can create ideas and concepts out of thin air, what we are doing is mining and building upon the past. Innovations are mostly based on combinations of existing ideas.
Thus, most creative endeavors are not truly original; they can be thought instances of combinational creativity. Examples could be combining a boat and a plane to create a seaplane, integrating a compass and GPS device to produce a device for runners, using lasers as pointers, using origami to produce less wasteful packaging, and bringing the massive power of computer science into biology to do DNA sequencing.
Another recipe for creativity: combinations of three
In fact, combining two things to create a new third is relatively common. Combining three elements to produce something unique is much less so. The Wright brothers and the invention of the airplane could be imagined as the coming together of knowledge of gliders, use of aluminum, and the wind tunnel. Guglielmo Marconi’s invention of the radio could be considered the merging of the telegraph, telephone, and radio frequency vacuum tube. Henry Ford and the development of the first affordable automobile could be seen as the maturation of the gasoline engine, petroleum refining, and the concept of the assembly line.
This list may even include things like the iPhone. The iPhone is a triumph of design but involves pieces that were already there: the large scale integrated (LSI) circuit, the gallium arsenide (GaAs) radio chip, and the lithium battery. And each of them can be traced much further back. The beauty resides in the artful integration to produce a new whole.
It is also useful to imagine how entire disciplines — not just technologies — may be combinations of three components. Modern medicine could be imagined as the combination of information in the form of medical science and knowledge (all the underpinning science, beginning with germ theory and moving into molecular biology, genomics), new classes of machines (X-rays, MRI, robotics, diagnostics), and new kinds of materials (pharmaceuticals would fall in this bucket, as well as implantable and surgical materials).
The cases listed above suggest the following question: Are there combinations of three things out there waiting to be discovered? None of us can see the future, and AI has access only to the past. But the past may contain vast unexplored and unexploited territories that become visible in their creative possibilities to those who pose the right questions
Still, we must recognize that even if much consideration goes into selecting X, Y, and Z, the question of “What can one get if we put X, Y, and Z together?” may be way too broad to generate any useful answer. Instead, we must think about the right way to ask this question.
Breakthroughs and break-withs
There is also a type of creativity that involves radical shifts in perspective, leading to the creation of entirely new paradigms or frameworks. It is creativity that is more closely linked to disruptive innovation — what many people refer to as “breakthroughs.”
Breakthroughs represent a sudden development and a new insight. The term itself suggests something that pierces the boundary of a domain of knowledge and enlarges the domain.
But there is another term that goes beyond breakthrough: break-with. A step above a break-through, a break-with is conceptual advance that breaks with the ideas that were at the very center of old way of thinking. In extreme cases, new domains form and a new order sets in. The old way of thinking becomes useless or outmoded when the new way kicks in.
Somewhat ironically, A.I. itself had its own break-with. Today’s machine learning-based A.I. is itself a paradigm shift, for this is not how the A.I. community in mid-1970s imagined A.I. The goal then was to understand how humans thought. A.I. began as a “top-down” enterprise in which symbols were used to denote concepts as designated by humans. Researchers designed complex representations from these symbols, along with a range of mechanisms to reason about the ideas represented.
In parallel, a very different approach was taking hold. Machine learning — a form of statistical learning — does not rely on the same kind of symbolic representation. Rather than assert from the top down how concepts are related, associations of features are learned “bottom up.” In recent years, the development of “deep learning” has led to systems that, in effect, discover or create features at an intermediate level of abstraction that, in many cases, are not understandable by people.
Could A.I. have created break-withs? Some thought experiments.
To understand A.I.’s ability to generate break-withs, let us consider a few thought experiments. It is 1970, and A.I. as we know it is already with us. You are an architect based in Los Angeles designing hotels with big atriums. You want ideas, and you ask A.I. for help. A.I. knows everything that has happened with buildings in LA, the US, and the world. And let’s say this knowledge covers the last 200 years and you want 100 ideas.
Could any of those 100 ideas anticipate what John Portman did in the mid-1970s, making elevators visible in atriums or putting them outside the building? Most likely no, since no pre-Portman building had used this idea. Perhaps A.I. could have come up with it if you instructed it to “put the elevators anywhere.” But then question already contains the answer…
Another example: Origami, the Japanese art of paper folding, has been around for at least 500 years. But at some point, in the last several decades, it has become of interest to mathematicians and engineers. So much has happened in this space that now we have an entire domain called origami engineering. Applications range from packaging to architecture, robotics, medicine, and aerospace. Being able to transport something small — the critical issue in transporting things to space — and having it unfold to produce a solar panel half the size of a tennis court is something that would have been unthinkable in 1960. Now it looks easy.
What kinds of questions could have generated this merger of ideas? Questions such as “Could origami inspire questions in math?” or “Can you find applications of origami that solve issues in engineering” already contain the answer. How can we ask questions to find unexploited territories?
Tensegrity and its connection to art, architecture, engineering, and biology
Tensegrity is a concept that is rather tricky to explain in words but easy to see when a tensegrity structure is in front of us. The idea emanates from mechanics and is a structural principle where members of a unit consisting of rods and cables are either pure compression (rods) or pure tension (cables). The term was coined by Buckminster Fuller in the 1960s, but precedents go back to a Latvian artist in 1920 or even earlier, to James Clerk Maxwell, the father of electromagnetism, and his work in structures.
Once the tensegrity concept was understood, myriad applications and discoveries emerged. Sculptures were the first, and architecture followed, beginning in the 1960s, with Olympic Domes and bridges. Aerospace exploited the concept as well. NASA’s Super Ball Bot is an early prototype designed to land on another planet without an airbag. Since the early 2000s, tensegrities have also attracted the interest of roboticists due to their potential to design lightweight and resilient robots.
And then there is what one can see what was already there, but we did not see because we did not have the right lens. It happens that biology has been using tensegrity ideas since the beginning of time. Not surprising this realization has a name: biotensegrity. Bones, fascia, ligaments and tendons, rigid and elastic cell membranes, all have been using tensegrity in one way or another. More recently it has even been used to describe numerous phenomena observed in molecular biology.
Tensegrity was out there, ready to be deployed, and it took less time than origami. But how could we have discovered that it was there and that it could have so many applications?
Breaking fundamental assumptions and giving rise to a whole new discipline
Physics, prior to 1900 could explain a lot. But there were some pesky experimental results, the so-called black-body radiation problem, that resisted explanation. Max Planck’s solution in 1900 and Albert Einstein’s 1905 explanation of the photoelectric gave rise, in what it looks in science-timescales remarkably quickly, to the birth of a new physics: quantum mechanics, which along with the near simultaneous development of relativity theory, led to a relabeling of all previous physics as classical physics. The new physics required abandoning central concepts at the very heart of classical physics. Physics became classical physics plus new physics.
It is doubtful that a machine analysis of the entire history of physics prior to 1900 would have been able to come up with quantum mechanics or relativity. Only in hindsight does it become clear that quantum mechanics, or something like it, was necessary because classical physics could not explain particle physics. But few physicists at the time believed that we needed to break the assumptions underlying classical mechanics. Why? Because classical physics was so successful. This new way of thinking violated the very cornerstones of all prior knowledge. Energy being discontinuous? Space and time being equivalent? Could any of those radical concepts have emerged by looking backwards? What kinds of questions could have led to quantum mechanics?
What about art?
Whether AI could have predicted break-withs becomes even murkier when it comes to art.
Science is about unveiling, revealing what may already be there. If Issac Newton had not come up with the theory of gravitation, someone else would have done so, though possibly in a different form. In this sense science is inevitable; technology much less so. Art, especially modern and contemporary art, even less. One could argue that the history of art — especially modern and contemporary art — is a succession of breaking assumptions. This, in particular, seems hard to codify. Could AI, if fed with the entire body of art between 1800 and 1900, have predicted cubism?
Prediction and understanding: The difference between A.I. and mathematical theories
In 1960, the physicist and Nobel Prize winner Eugene Wigner wrote an influential paper, aptly titled “The Unreasonable Effectiveness of Mathematics in the Natural Sciences,” where he marveled at the ability of mathematical theories to explain much of nature. He said, “The miracle of the appropriateness of the language of mathematics for the formulation of the laws of physics is a wonderful gift which we neither understand nor deserve.” Feed math into the right theory and, voilà, predictions! This was a bit of a stretch, and the paper generated lots of commentary, but on the whole, critics and supporters alike would have agreed that prediction depends on understanding or, more forcefully, that understanding is a necessary condition for prediction. To put it the other way around, “there cannot be any prediction without understanding.”
AI, of course, has changed this. One can now predict things without understanding them. This is liberating, but questions still rule.
Asking the right questions
How our AI co-pilots and other assistant tools can help with questions remains an open question. There is no theoretical reason we can think of to believe that creativity cannot be modeled. Generative A.I. excels at combinatorial creativity. In fact, it is better than humans in many cases, because it can mine a vast amount of input data to find patterns that are beyond the field of view for humans.
But left to its own devices it creates pastiche. Maybe even great pastiche, but in the end, it is comingling and recombination of past creations. The very thing that makes AI so effective is also its limitation.
Transformational creativity seems, as of now, beyond the reach of generative A.I. Generative A.I. is entirely based on the past, so it cannot break from the past to create fundamental innovations. Transformational creativity is rare and infrequent, so it is tempting to equate with the essence of human genius.
Perhaps generative A.I. will get there. Either way, the path to using it as our creative assistant lies in learning to ask better questions to unlock creative potential. In 1968, in what it what it was a different era, especially in computer-years, Picasso said :“Computers are useless, they can only give you answers.” A bit extreme, but he had a point. As of now, there is no algorithm to ask questions, at least not the kind of questions represented by our examples above. Questions generating those types of rich connections rarely emerge as epiphanies, fully formed prescriptions on how to develop an entire new area.
But you can learn how to ask better questions in general. Or how to be prepared to ask better questions. Being exposed to new ideas, being constantly curious, and even developing a sense of awe, seem like preconditions to uncover new types of connections. And we must remind ourselves that the really good questions already carry the answer.
