What AI can learn from nature

This article by Fiona J McEvoy (YouTheData.com) was originally posted on All Turtles.


In designing his famous flying machine, Leonardo da Vinci took inspiration from bird flight. The inventor’s Codex on the Flight of Birds, details their behaviors and makes proposals for mechanical flight that would influence the development of the first modern airplane hundreds of years later.

Birds aren’t the only animals to influence scientific progress. For many years scientists have sought to unlock the extraordinary qualities of shark skin, which has huge advantages for both increasing speed and repelling germs. Recently, Walmart filed a patent for the creation of a swarm of robotic bees which they hope to use for the autonomous pollination of crop fields. Perhaps unsurprisingly, the humble original is perfectly designed for the task.

This list is certainly not exhaustive. Nature and its sophisticated design have a long history of motivating the creation and development of new technologies. Artificial intelligence is no different. Its creators increasingly look to our environment, and its host of specifically evolved creatures, to inform their own systems and machines.

Dr. Anthony DeConstanzo, a research scientist at Ascent Robotics, believes that biomimicry – designing systems based on biological entities – will continue to influence technological development in a range of ways. He argues that natural organisms are the ultimate general purpose entities, and says that if we wish to increase the number of tasks a technology can competently perform, we will end up at a loss without reference to biology.

Man vs. Machines

The popular media often like to talk about AI as a potentially hostile force. A phenomenon that threatens to outperform, and perhaps ultimately overthrow humans. This anxiety has been fuelled by commentary from high profile figures like Tesla CEO Elon Musk, the late Professor Stephen Hawking, and Apple co-founder Steve Wozniak who famously pondered whether in the future humans might become the “family pet.”

Regardless of these lofty voices, it is still the case that true AI or artificial general intelligence (AGI) is a long way off. Though some impressive systems can already beat humans at Go or make more accurate diagnosesthan doctors, no technology currently has the flexibility to apply its intelligence to a broad and varied spectrum of complicated tasks. Leading academic commentator Professor Luciano Floridi of Oxford University has already come out and dismissed the spectre of superior, ultra-intelligent machines as “utterly implausible”.

Claims about the exponential growth of artificial intelligence, or a forthcoming “intelligence explosion” are hotly contested. But doubt about human-rivaling AI is also compounded when we consider the enormous task at hand. Humans – and other biological creatures – are hugely complex. Even our best engineers aren’t close to creating non-biological systems that can reproduce the full range of physiological and cognitive abilities that our species have evolved over millennia.

Right now we aren’t on the brink of acquiring superintelligent overlords. And some would say that technology still has much to learn from our natural world.

(Deep) learning from nature

Dr. Anthony DeConstanzo understands the interplay between natural and artificial intelligence. His background in experimental and theoretical neuroscience now informs the work he does at Ascent Robotics, a Japanese company working to produce autonomous vehicles. DeConstanzo explains the connection:

“As of now, animals still perceive, navigate, and make decisions much more efficiently than any robot or vehicle that has been designed.  By understanding how animals represent such tasks in their brains we seek to reverse engineer autonomous systems that make efficient use of such representations.  This does not mean, of course, that we plan to recapitulate the entire cellular basis of cognition in silico. It is probably not necessary to do so to achieve very good autonomous agents.  Instead we are using the solutions provided by the brain as a rough guide to give us algorithms that would have been difficult to find with brute force simulation methods.”

It’s not a secret that the very concept of deep learning and neural nets is derived from what we know of the workings the brain, and yet DeConstanzo says that methods like reinforcement learning – inspired by behaviorist psychology – had until recently been dismissed by many as too slow to be useful in computing:

“I always thought that reinforcement learning was an elegant way to express biological learning on a high level, and felt it inevitable that computing would one day deliver powerful AI through the very same mechanism.  I was told by many in machine learning that it was too slow to ever be useful in computing, and the same for evolutionary strategies. Well, it seems they were wrong. I really think we are only at the very beginning of an incredible time in human history.  If you notice, the digital computing architectures we use haven’t really changed much since Von Neumann.  We are going to see some bizarrely organic new computers coming out. They will be lightning fast, very low power, and capable of chillingly animal-like behaviors.”

The word “chillingly” is perhaps not accidental. It is certainly understandable. Most of us are familiar with the “uncanny valley” and the strangeness of seeing typically human behaviors performed by tangibly non-human entities. But could a deeper dive into biomimicry ultimately make humans and animals vulnerable, validating the fears of figures like Musk and Hawking? In teaching machines some of the methods of biology, are we providing a leg-up for our new AI rulers? Should we proceed with caution when it comes to using the methods DeConstanzo describes?

The debate rages on, but Professor Floridi is adamant that more advanced systems do not inevitably lead to superintelligence or an AI threat: “The truth is that climbing on top of a tree is not a small step towards the Moon; it is the end of the journey. What we are going to see are increasingly smart machines able to perform more tasks that we currently perform ourselves.”

Given the current mysteries surrounding minds, brains, and consciousness, it seems unlikely that replication is inevitable. Especially with researchers like DeConstanzo using functions like biological cognition as a “rough guide.” Nevertheless, that the mind is mystery also refuses to rule out that the abstraction of biologically similar machines could one day result in a self-awareness akin to our own.

Whichever way we choose to speculate (and there’s no doubting that speculation is a fun side-effect of the AI boom), we should agree that any panic about biomimicry giving rise to tyrannical AI should take a backseat to other more practical concerns about the technology’s relationship with society – like its implications for privacy and employment.

Moreover, at this juncture, we should privilege our excitement and curiosity over any sci-fi fears and as-yet-baseless trepidation. Who knows what we are yet to learn and utilize from our natural environment, and what solutions to real-world problems these discoveries may yield? In taking our cues from nature, we may be on the frontier of a whole new technological epoch. One which at least partially replicates and complements the world in which we already live.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s