Communications of the ACM
AI Accelerates Innovation
The deep learning algorithms of artificial intelligence can identify patterns that help inventors think laterally, make connections between nonobvious ideas, pinpoint hidden invention features, and exploit new science and technology-based opportunities.
Credit: ipCapital Group
"To invent, you need a good imagination and a pile of junk." So said Thomas Edison, America's most prolific inventor. Yet the march of technology is now changing the great man's inventive equation: powerful algorithmic advisory systems are now giving inventors far more fertile imaginations, even if they don't have very much of one themselves.
After being fed vast datasets of information on a field of inventive endeavor, deep learning algorithms identify patterns that help inventors think laterally, make connections between nonobvious ideas, pinpoint hidden invention features that rivals have missed, and exploit new science and technology-based opportunities from, say, patents and journals. In other words, artificial intelligence (AI) provides highly relevant additions to what Edison called his "pile of junk."
This far-broader range of insights is not only helping people to invent, it is doing so at great speed, as a meeting of inventors and patent attorneys heard at a London conference on AI-Generated Innovation held by the International Association for the Protection of Intellectual Property (AIPPI) in mid-March.
Such growth in patent filings is something IP experts are seeing across the board, says Fiona Stevens, a patent attorney with London-based legal practice Gill, Jennings and Every. "What is most amazing is the sheer speed with which AI is now letting pharmaceutical and healthcare companies innovate. Its use has quickly become mainstream," she says.
This speed boost was to be expected, says Julian Nolan, CEO of Iprova, a company based in Lausanne, Switzerland, that provides algorithmic invention acceleration services to blue-chip firms. "The rate of patent output is greater as the technology is now giving us the right stimulus to explore more inventive permutations," says Nolan, adding, "Our agile teams invent for our clients just two weeks after starting a project."
Inhuman, obvious, and invalidated?
While AI-accelerated invention might sound a great idea, the conference also heard some strong notes of caution over where inventions borne of artificial imaginations could ultimately lead us. If AI ever becomes smart enough to invent without human input, critics told the AIPPI conference, it could undermine the patent system, which generally holds that named human inventors are assigned a patent for developing a nonobvious invention.
Yet if inventing were to become an everyday function of algorithms alone, their output could be deemed an obvious product of data processing, giving examiners a reason not to grant a patent, and corporate litigators ammunition for the outright invalidation of a rival's patents. "If an artificial intelligence can be an inventor, it could call into question the very basis of intellectual property law," said Jonathan Moss, a London patent attorney and barrister.
Accelerating serendipity
It was not meant to be this way: the idea of algorithmically assisted inventing is to let people get more out of the patent system, not destroy it, by lessening the degree to which industrial R&D labs depend on unpredictable eureka moments (that is, plain old serendipity) for their success. "We live in a data-driven world, so why is invention still left to analog chance?" asks Nolan.
It no longer is, provided you can afford a decent GPU stack and the kind of algorithmic code that Iprova and Healx run. By using machine learning algorithms to sort and spot patterns in datasets, to make predictions that lead to logical connections, or to scour the literature, AI-based software products are able to make inventive decisions faster, better, and cheaper, says David Brown, chief strategy officer at Healx, a firm developing medications for the treatment of rare diseases in Cambridge, England.
"We regard machine learning training data as the fuel for the machine," says Nolan. "The machine can be as fast as you like, but without that fuel, it's not going anywhere. It's collecting really good training data that improves the data we are able to present to our invention developers."
So what patentable inventions have they come up with alongside the AI? One Iprova client wanted a way to tell if mouth-and-nose-covering facemasks used in healthcare were fitting snugly, so the patient is not breathing in pollutant particles in the air around an ill-fitting mask. The algorithms made a connection to a biocompatible chemical that, when exposed to humid air, changes color. By coating the mask in this substance, moist air breathed out by the patient to the exterior of the mask leaves a visible trace on the mask, so users can look in a mirror, or a phone camera, to see if the mask is fitted properly.
In another leap of thought, Iprova's algorithms helped invention developers realize that the time-of-flight depth camera in a TV set-top, Kinect-style gesture-recognition widget does not need its own light source, as the LEDs in the TV screen can provide the light. So Iprova and researchers at École polytechnique fédérale de Lausanne (EPFL), the Swiss Federal Institute of Technology in Lausanne, patented this cheaper way to perform gesture recognition. Showing the patent's worth in the IP firmament, says Nolan, is the fact that it has been cited as prior art in patents from tech firms like Texas Instruments, Intel, Google, and Amazon.
Standing up new therapies
Healx, meanwhile, is initially using AI assistance to help its drug discoverers in the repurposing of existing drugs, rather than finding totally new drugs (having already passed human safety trials, repurposed drugs can be gotten to market far quicker than new ones).
Brown knows all about repurposing: he was a member of the Pfizer team in the U.K. that developed the erectile dysfunction therapy Viagra in the 1990s (a drug initially developed to treat high blood pressure and angina), before a group of former coal miners in a test cohort discovered its hidden talents during a clinical trial.
"I'm so impressed with the output," Brown says of the invention-assistance algorithms, "but it doesn't just produce the answers. It takes a drug hunter one to two weeks to analyze the algorithm's output and to try to make sense of it. But what we are finding is that about three drugs out of seven that it finds are what we would call 'active'; in other words, they look promising to consider for possible clinical trials."
The medical conditions that Healx is currently seeking candidate drug molecules to treat include two childhood cancers, as well as Fragile-X Syndrome, a heritable disorder which causes cognitive impairment and severe learning disability, and Amyotrophic Lateral Sclerosis (ALS, also known as Lou Gehrig's disease or motor neuron disease, the illness from which late cosmologist Stephen Hawking suffered).
What if AI stakes a claim to invention?
Despite such firms pressing ahead seemingly successfully with algorithm-assisted inventing, worries persist over what will happen in the long term if an ever-smarter AI starts to challenge humans for the position of named inventor. "Lawyers in the U.S. are licking their lips at the prospect of litigating a patent where the patented insights came from an AI platform, as American law requires the inventor to be the person who conceives and holds the claimed invention," explains Peter Finnie, also a patent attorney at Gill, Jennings, and Every.
"We have to ask if our [patent] system is fit for purpose," says Moss. To do just that, the leading global IP lobbying outfits (the American IP Law Association (AIPLA), the International Federation of IP Attorneys (FICPI), and the aforementioned AIPPI) are attempting to thrash out policies on which the world's patent offices might one day need to agree.
In a joint primer document, the trio pose this intriguing question: "In most jurisdictions (China, Japan, South Korea, and the U.S.), an 'inventor' is defined as either an individual, human or person—or is undefined entirely, as is the case in Europe. Might this be a case where, for the good of all, a common definition is laid down for an inventor—such that an AI entity can be considered a co-inventor? Then the next step might be to figure out what happens if the AI is the sole inventor."
Nolan says such concerns are overblown, as the AI Iprova uses is nowhere near capable of working alone, nor is it likely to become so. "We are not making invention easier. In fact, in many ways we are making it more difficult. It's not simply a matter of making an investment in machine learning technology. It's not magic. The people we have creating the inventions are very, very highly skilled and typically have a Ph.D. in physics or electronics engineering. We are not using unskilled people to create inventions."
Healx's Brown agrees. "With machine learning, there is no intelligence at all; it is just data being analyzed by software. In the end, it is still humans writing the algorithms, and driving them, putting the right data in and getting the output. All it is doing is speeding that process absolutely massively."
Speeding it, in fact, much like Edison did with his industrial-scale invention factories. "There are many examples of initiatives to improve the efficiency of invention creation over time, and Edison's labs are one such example," Nolan says.
"What we are seeing now is simply the digital transformation of Edison's long-established trend."
Paul Marks is a technology journalist, writer, and editor based in London, U.K.
No entries found