9 out of 10 execs say Artificial Intelligence looms as opportunity, but why, and how?
A recent Boston Consulting Group and MIT report found that, based on nearly 2,500 executives, 9 out of 10 of the respondents agreed that AI represents a business opportunity, but what exactly is AI? And what makes it something capable of conferring advantage so many industries?
When Hugh Bradlow, Chief Scientist from Telstra gave a substantive address on the future of information and its impact on technology, as reported in the Digest in May 2015, he predicted the future. He said “We will see abundant computing, abundant data, new analytics and artificial intelligence, through cloud services. Today, we take the data mainly from human sources. But now, we can expect more and more data from devices, with streaming analytics and data lakes.”
He was right. That time has come.
What is it?
AI gives computers adaptive intelligence analogous to that which humans have so that computers are able to learn and reprogram themselves based on the available data. As a general framework, and particularly over the last two years, AI has dramatically shifted the landscape in everything from marketing to advanced material design, and it is going to revolutionize nearly every business landscape unlike any other advancement we have seen before.
Who is using it?
For one, organizations such as Sandia National Labs, Google, ABB, and the software supporting your phone .Material scientist’s are using it to expedite advanced material discovery. Grid managers are turning to AI to maximize the optimization of a distributed renewable energy grid.
How does it work?
AI will have different fundamental approaches to problems depending on the task, similar to the way people have different approaches to thinking numerically versus thinking creatively. Each approach has its own unique set of advantages, but ultimately the type of program depends on the question at hand. For example, Google’s DeepMind program, which was responsible for reducing the energy used in their data centers by 40%, was ultimately looking to optimize something. This a fundamentally different question than the types being posed by material scientists leveraging AI, where their model involves the “careful explanation of the underlying physics between the representation and the properties of interest.” The most relevant problem types in the industry are classification, continuous estimation, and clustering. Google’s DeepMind, was a program based on reinforcement learning, which is essentially learning by trial and error. It is often the case that the program is not clear what goal it is trying to accomplish, and through a process of positive and negative reinforcements it is able to optimize a system of interest.
The AI and Machine Learning back story
Use cases abound all over the sector.
We reported earlier this month that the DOE, along with the Hydrogen Materials Advanced Research Consortium and Sandia National Lab, have begun collaborating to develop more advanced storage materials in order to meet federal energy density requirements set by the DOE for hydrogen fuel-cell vehicles. The group is leveraging machine learning to rapidly identify the physical properties of these storage materials that correlate to the performance necessary to reach the federal targets. Their approach allows them to understand how the computer identified its predictions.
Last December, we reported that researchers at Carnegie Mellon University, with support from the DOE and the National Science Foundation, are developing a machine learning algorithm that can play into the discovery of new catalysts. Through the development and implementation of novel machine learning algorithms, the rate at which researchers can discover new, effective catalysts will increase exponentially.
In March, we reported that Lygos partnered with the Center for University of Massachusetts-Industry Research on Polymers (CUMIRP) to develop novel high-value applications for the company’s Bio-Malonic Acid product family,. This research collaboration will characterize polymer microstructures using novel artificial intelligence and machine learning-based approaches to elucidate structure-property relationships for disordered crosslinked systems. Despite an abundance of data on crosslinked polymer systems, the knowledge of how to achieve the ideal balance of chain extension and crosslinking for optimizing performance in specific applications is still lacking.
In July 2018 we reported that the University of East Anglia was searching for a first level researcher to apply machine learning approaches to investigate cyanobacteria metabolism and identify optimal pathways and metabolic fluxes for compound production
In May 2018 we reported that scientists from the Department of Energy’s Lawrence Berkeley National Laboratory (Berkeley Lab) have developed a way to use machine learning to dramatically accelerate the design of microbes that produce biofuel.
We reported in 2017 that John Deere announced they are buying Blue River Technology, a Sunnyvale, California company that specializes and is amazingly smart in artificial intelligence for agriculture. It isn’t coming cheap at $305 million, but Deere is focused on expanding their automation technologies and artificial intelligence product offerings.
And we reported in July 2018 that Growlink, a leading agriculture technology company, announced a beta program for its new AI solution for greenhouse and indoor farming. Growlink Plant Health AI uses cloud machine learning, computer vision, sensor data, and AI to track and predict plant health.
AI in the broader energy sector
In the renewable energy sector, AI overall is set to have an incredible impact in the design and manufacture of advanced materials, as well as optimizing the demand response for renewable energy to minimize curtailment, or a reduction in the power output. A promising area of study in the AI and material science intersection is the development of an AI system that is capable of receiving desired properties and outputting a material with those very properties: also known as property prediction.
The digital technologies supplier, ABB, manages their solar and wind supply/demand mismatch with AI. Using their facility’s power data, they are able to pinpoint opportunities for productivity improvements and energy cost savings. To successfully identify as many improvements as possible, their AI platform— developed by Verdigris Technologies— forecasts the demand and weather 24 hours in advance. This enables them to take timely action to reduce unplanned consumption spikes and switching off non-critical loads– taking advantage of the time of use tariffs.
When it comes to distributed energy networks, the sheer volume of the system can prevent its organization from being manageable by any single person, therefore the most cost-effective method for managing and optimizing every factor at play for distributed systems is using an AI network. The graph below shows the Power Usage Effectiveness on a time scale, and it shows rather undeniably that AI algorithms can have a major impact on the overall efficiency of a system. On a large scale, this untapped savings potential could represent billions of dollars worth of savings. That is probably why in late October, National Grid invested in companies specializing in AI grid management.
Property prediction would allow for the rapid screening of materials and could revolutionize catalysis research. AI is being used to determine better catalysts for renewable energy production of chemicals and plastics through the utilization of biomass and CO2. The advancement in both the energy capacity of advanced materials and the utilization of carbon dioxide could be a major force behind a knockout punch to the unsustainable habits of the petroleum industry.
Reducing trial and error
Although there is still a need for human intuition in experimental synthesis from the newly discovered materials, one can reduce unnecessary trial and error for exploring undesirable or implausible chemical outcomes by utilizing ML models. Stanev et al proved this during their study of superconductive materials, where they applied a “random forest method” to classify critical temperatures to ID superconductive materials, and the algorithm identified 35 promising materials for the researchers as a starting point. They showed how AI is narrowing the gap between theory and experiments. the ability to discover game-changing materials with superior efficiency and performance has the capability to fundamentally shift the state of the renewable energy industry. However, a materials revolution is slow-moving, a more tangible application of AI is how it is changing our approach to managing the electrical grid.
Bring lots of data
In order for any industry to capture the benefits AI has to offer, data must not merely be available, but also be at an appropriate volume, velocity, and variety for it to be leveraged with positive results. For this reason, even among those who are pioneers in their application of AI to their business, 30% have yet to see the business value of AI materialize. The room for error is also rather small, as a small unidentified bias in data could cascade into a catastrophic outcome. Ultimately, those looking to integrate AI should begin collecting high-quality data as soon as possible, because in an industry that moves as rapidly as artificial intelligence, not being prepared for an opportunity when hits could mean the difference between pioneering leadership and getting left in the dust.
More on the story
For more information on how to avoid being left in the dust, take a look at the IRMA report here, which provides a general overview of AI’s impact on renewable energy.
Category: Top Story
