An AI Crackpot Application

So if you have been following along, you know that I am pretty interested in how AI technologies are developing. There are a lot of powerful new tools with surprising applications. And for every one of those, there are scores of incremental innovations that make their way through the patent system.

But as with any field that starts to gain credibility, the crackpots come out of the woodwork to exploit its newfound cachet. So it is with pleasure that I introduce to you this application for a human brain like decision-making machine:

To set the stage, this application has been filed by an organization called the National Institute for Materials Science in Japan. They appear to be a legitimate research institution, with a long history, so it is not entirely clear to me how this ever got filed. Furthermore, the application was filed internationally, by real law firms. The amount of money it costs to do that puts this effort outside the reach of your average crackpot. So someone out there is taking this seriously.

That someone just isn’t me.

The first paragraph starts off strong, claiming to create an entirely new branch of information theory that they call “Fractal Information Theory.” The concept is that Turing machines and all of the computers that derive from Turing’s work are inadequate to capture true artificial intelligence. And to top it all off, they suggest it is a perpetual motion machine: “It does not require power supply as it harvests noise.”

Rather than using a Turing tape as the basis for their model, they use what they call a “Fractal tape,” where, “The idea is to hack nature and make a computer that can generate most patterns that we see in nature.” ¶2. In the grand tradition of crackpots, the patent application attempts to drown you with nonsense so that you can’t make heads or tails of what it’s actually trying to say. So let me just grab a few choice morsels:

The present invention enables us to create standalone robotic brain that does not require any programming. Using all possible solutions of cavity resonators a prime metric is calculated and the hardware is primarily an assembly of cavity resonators that generates the vibrational frequencies as described in the prime metric.

USPP 2020/0160147, ¶11

Continuous filling up cavity inside until atomic scale enables packing astronomically more clocks. Not just that a time crystal can hold several geometric information to be viewed distinctly from different directions. Hence a single time crystal could hold large number of geometric shapes in a fixed space.

USPP 2020/0160147, ¶12

For this computer, the fractal tape is used as a tape that contains phase as the only variable. As the phase changes 360° one gets time. One interesting aspect of using phase or time is that the inverse of time is frequency and if a frequency structure is made, the entire information architecture could be converted into a material. One to one correspondence between a material structure and its information content is a key feature of the computer hardware.

USPP 2020/0160147, ¶149

The whole thing reads like a computer scientist with graphomania took LSD. Then of course there are the figures. I’m just going to lay these out for you, because I promise my commentary won’t help you decipher them:

When I pick up one of these applications, I like to look at their ongoing prosecution to see how the patent examiner is handling them. As you know from my previous posts, some examiners will simply allow a case that has an especially long claim on the theory that it will never come back to bite them.

But in this case, despite the length and narrowness of the claims, I have reason to believe that the examiner handling it will rightly grind it into paste. I’ve worked with him before, and he has a sharp eye for bullshit. I do not envy the patent attorney who is tasked with ironing this garbage into something sensible.

The application has already received one rejection. The rejection starts off strong by pointing out that the claimed invention doesn’t work and can’t work, and then picks apart each line of the claims to point out that it makes no sense and/or that the specification doesn’t actually say how to do the thing. The patent attorney has made a valiant attempt to pull it out of the fire, and I’m looking forward to the next office action for sure.

Regardless of how the outcome for this patent application, it is a serious embarrassment to the National Institute for Materials Science. Even if it were allowed in its present state, the claims are completely valueless. Hopefully they will give their researchers a bit more scrutiny going forward, before putting their resources behind projects like this.