The Fuse is Lit for the Next Cambrian Explosion
Back in 2016 in a post entitled “AI: The Next Cambrian Explosion” we speculated about the role open source might play in the acceleration of value created by potential AI solutions. At the time most of the major players were contributing some or all of their AI platforms to the open source community in hopes of spurring adoption and thereby promoting market coalescence for proprietary offerings undergirding their contributions. The question then, and now, was what part of the value generated by existing AI processes could be captured by the platform providers if the engines of those processes were freely available to exploit by any rival within any industry or discipline.
By making the engines of AI, the teachable platforms that produce machine recognition of associable artifacts an open, externalized economic factor of the value creating process (see The Paltry Cost of Priceless Externalities – January 2020), value extraction by the platform players would have to be found either in the resources consumed by those engines (marshaling data and consulting services to feed the engines) or the proprietary insights or intellectual property that resulted from the consumption of those resources. The former being as economically attractive as shoveling sand to make glass and the latter for all practical purposes unobtainable to platform providers owing to the proprietary intellectual capital, data and expertise, required to produce the insight.
Well, in the past few weeks we’ve been given a glimpse of the practical and profound economic implications of just this kind of proposition. And it will likely take several years to sort out the all the ramifications that this one event will produce.
As published in Science and Nature, researchers David Baker and Minkyung Baek at the University of Washington, Seattle, published and open-sourced, both the method and code, a proven, three-track AI model for predicting protein structures. The code has been made available on GitHub, has successfully modeled the structure of over 4000 proteins and has been downloaded worldwide. The program dubbed RoseTTAFold has been hailed as a breakthrough of unimaginable proportions. Tellingly, shortly after this announcement, Goggle’s DeepMind announced that they would also open-source their AlphaFold2 code which received a good deal of attention in the fall of 2020 for similar accomplishments. For life-science and pharmaceutical companies, this portends an unprecedented acceleration of innovation but due to the nature of the data and methods may not necessarily portend an abundance of profit for the ensuing discoveries.
At issue is both the nature of the discoveries, as well as the methods employed to make them.
It has long been held a tenet of intellectual property that phenomenon that occur in nature cannot be subject to patents. The CDC ran into this as recently as 2015 when it attempted to patent a SARS virus and the patent was denied. Proteins are naturally occurring substances and their beneficial interaction is the product of evolution not necessarily drug related research, design and engineering. So the burden of proof for any innovative protein would be to establish beyond reasonable doubt that it is both novel and not predicated on naturally occurring biological circumstances. An artificial protein that possessed such attributes may qualify as patentable intellectual property but since the benefit of most proteins lies in their interaction to other naturally occurring proteins one would have to wonder what the application would be.
Next, one would have to wonder whether any truly unique intellectual property derived from open source methods and code can be considered patentable since the innovation itself could be easily invented by any other random party with the means and expertise to employ the methods, data and code. All invention would be rendered merely latent, prior art.
For instance, problems in nuclear fusion are similar to those facing protein scientists. Recently, it was suggested that there may be up to twelve different, very subtle phases of plasma, the primary means of fusion containment. Understanding and unlocking the behavior of those phases could well be the key to unlocking the promise of unlimited clean energy. Should an open-sourced AI method and code be developed to analyze, model and replicate a successful containment phase of plasma, which is not that big a stretch, would such a discovery be patentable? Recent events suggest that it would not.
Now extrapolate this same scenario for any imponderable problem facing modern science and let the second order effects of that wash over you like a spent wave on a sandy beach tickling your toes.
Capitalism, as we know it, would lose its relevance as would most of the financial and equities markets. Sovereignty of nations and individuals would go with it. The social contracts that bind us would dissolve and it would be incumbent upon those to whom these blessing are bestowed to reinvent civilization.
But we’ll save that for another day.
Cover graphic courtesy of the Alpha Omega Institute