Episode Summary:
A French research team has proposed a radical innovation in artificial intelligence: electromechanical artificial neurons. The team has developed technology that mimics the function of mammalian neurons closely, using graphene to trap water molecules in a similar way to the internal flow of ions within nerve cells. Simulations suggest that the water molecules will assemble themselves under the influence of electric fields in ways that show a memristor effect, very promising for both storage and information processing applications.
In a different yet similarly significant AI development, the US Patent and Trademark Office has rejected an inventor’s application that a neural network be named as inventor on a patent application. That rejection was upheld by a federal court, but the appeals process is ongoing. The situation in Australia is exactly the opposite, with lower courts ruling that AI can be named inventor, with that nation’s patent authorities appealing that decision. The outcomes may have a profound effect on how we define invention and innovation in the 21st century.
Access all episodes of This Week in Engineering on engineering.com TV along with all of our other series.
Transcript of this week's show:
To see any graphs, charts, graphics, images, and/or videos to which the transcript may be referring, watch the above video.
Segment 1: Artificial intelligence has traditionally been about algorithms. The goal has been to take conventional digital logic running on semiconductor based very high density circuits and use software to mimic the thought processes of the human brain. This technology is literally simulation: it is not an analog to the biological processes that generate human intelligence. But what if it was possible to engineer systems that physically mimic neurons? Human neurons are essentially electromechanical signal processing units. Ions travel in channels and are shunted into and out of nanoscale channels depending on external electrical stimuli. Could it be possible to engineer something similar from nonbiological materials?
A team from the ENS Laboratoire de Physique (CNRS/ENS-PSL/Sorbonne Université/Université de Paris) in France thinks so and has proposed an artificial neuron composed of extremely thin graphene slits which trap a single layer of water molecules. The team has demonstrated that by applying an electric field, ions arrange themselves into clusters and more importantly show a memristor effect. The molecular clusters store some of the charge from field stimulation in the past. This nano fluidic system closely resembles the natural neurons ion flow behaviour.
With computer simulation, the scientists have demonstrated that these artificial clusters can be assembled to produce the equivalent of a natural neuron’s action potential, the key to information processing in the human brain. The French team, now collaborating with scientists from the University of Manchester, hope to build hardware to demonstrate the process with simple learning algorithms. If it works, the result may be a major paradigm shift in artificial intelligence research: away from electronics and into nano-fluidics. And mechanical engineers may have an entirely new role in the computer industry.
Segment 2: In other AI news, a little reported but very significant ruling reported by the register.com was handed down last week by a US federal judge: AI systems cannot be granted patents and will not be legally recognized as inventors in US law. The decision upholds a previous ruling by the US Patent and Trademark Office. This important test case was brought by Stephen Thaler, founder of Missouri-based Imagination Engines, a firm that developed a food container based on fractal geometry using a neural network that Thaler calls DABUS.
According to Thaler, the neural network is the inventor of the container, but his 2019 patent application was rejected on the basis that only natural persons are allowed to be named as inventors in the application paperwork. Thaler sued the director of the patent office, Andrei Iancu, in federal court in eastern Virginia in an effort to reverse the decision. According to Thaler’s lawyers, there is no specific statute or precedent in case law to prevent an AI created invention from earning a patent or that prevents an AI from being listed as inventor.
But federal court Judge Leonie Brinkema rejected that argument, agreeing with current patent office director Drew Hirshfeld, and ruled that since the law requires “individuals” to take an oath to swear they are the inventor on a patent application, artificial intelligence systems are excluded, since only natural persons can swear an oath.
Thaler has had more success in Australia, where a federal court in that country agreed that an AI could be granted a patent. That ruling however is under appeal, says Australia's Commissioner of Patents Paula Adamson. Australia’s intellectual property protection agency is staying neutral on the issue, but Thaler feels that machines should be protected from human theft of their ideas. Do AI systems have rights? This little-known case may drive the first coherent legal decisions on an issue that sure to become very important in the future.