A.I.: What we can learn from the genius baby

This morning I was listening to a podcast episode of Crazy/Genius titled “Can Artificial Intelligence Be Smarter Than a Human Being?” A question I’m sure many of us have pondered. Artificial Intelligence, and more specifically the subset of Machine Learning, is becoming more and more ubiquitous across all business sectors. In many cases in both public perception and actual business practicality, not implementing A.I. in your startup is seen as a competitive disadvantage. But as we move toward a future in which A.I. might surround almost every aspect of our daily lives, a natural question to raise is “What’s the true purpose of A.I.?”

To backtrack, let’s distinguish Artificial Intelligence from Machine Learning. To simplify, Artificial Intelligence (“AI”) is the broader aspect of machines that can perform tasks that are characteristic of human intelligence. While this is rather general, it includes things like planning, understanding language, recognizing objects and sounds, learning, and problem solving. Machine learning (“ML”) is a subset, or branch, of AI – it is a way of achieving AI by allowing machines to process information and learn on their own. AI can be achieved without ML, but this would require building millions of lines of code with complex rules and decision-trees to perform tasks that would traditionally require human-level intelligence. With ML, you can teach the machine to learn & adjust over time as it processes data.

So where would ML be useful? The answer is virtually every. single. sector. Take the growing field of autonomous vehicles for example which today utilize a field of engineering called “computer vision” to process and analyze visual data on the road. Instead of providing massive amounts of data to these machines to distinguish between objects, we can teach them to learn and reliably adjust over time as they process data. Another example is in the same industry, but in the hardware manufacturing sector where generative design systems were used to create car parts that weighed less with the same flexibility and strength. The AI system mimicked biology to create parts using a honeycomb structure, that when zoomed down to the microscopic level, looks and functions similar to our own bone structure. Expanding even further, you could apply this same generative design process to experimental drugs using a simulator to work through various scenarios of how a drug might react in various conditions without risking a human life.

So AI has the ability to enrich our lives and processes in so many different ways, but can artificial intelligence be smarter than a human? At the moment the answer is yes and no. Let’s take the example of the Deep Blue chess computer developed by IBM that ultimately defeated chess grandmaster Garry Kasparov in 1997. In the dimension of chess-playing, Deep Blue is by definition “smarter” than Garry. However, you can’t interview the computer after the game ends. In this way, machine learning is like a genius baby – it absorbs everything, it does a few narrow things extremely well, but understands almost nothing.

Another example of an ML folly described in the podcast episode was one where researches decided to build a robot that could rearrange itself to cover some distance. The researchers expected the robot to ultimately rearrange itself to have legs, learn the mechanics of walking, and reach its destination by walking over. However the machine decided to stack itself vertically and fall forward. From our narrow human perspective, we would label the result as a “failure” by projecting our own biases onto the machine. But from the objective of the machine, the goal was achieved – it found the most efficient way to cover that distance.

The interesting fact is that biologists discovered that wheat uses this same strategy to propagate. At the end of each season, these tall stalks of wheat fall over and their seeds land slightly farther away from their origin. So what took wheat millions of years of mutation and evolution to learn, this ML system learned rapidly. In a research paper from March 2018 titled The Surprising Creativity of Digital Evolution, the concluding excerpt reads:

“Across a compendium of examples we have reviewed many ways in which digital evolution produces surprising and creative solutions. The diversity and abundance of these examples suggest that surprise in digital evolution is common, rather than a rare exception. For every story we received or heard, there are likely to be many others that have been already forgotten as researchers retire. The ubiquity of these anecdotes also means that creativity is not con?ned to evolution in nature, but appears to be a pervasive feature of evolutionary processes in general.

These anecdotes thus serve as evidence that evolution—whether biological or computational—is inherently creative, and should routinely be expected to surprise, delight, and even outwit us.”

 As AI improves and becomes more ubiquitous, the best way forward in my mind is to augment AI with our lives to better enrich it. But for now, what we can all learn from these “genius babies” is to remove our subjective biases on what the “wrong” or “right” way to achieve a goal is, in order to foster unadulterated creativity. The most creative solutions can often be discovered through unconventional means, as evident through examples in biology and computation. At the moment, these genius babies are learning about the world with pure innocence and wonder, not tainted by the limitations of subjective biases that we tend to thrust on our lives as we grow older. Whether AI will overthrow the human-race like Elon predicts…I’ll save that thought for another time.