Hybrid AI through data, space, time, and industrial applications: Beyond Limits scores $113M Series C to scale up

For a hitherto relative unknown, scoring a $113 million Series C at this time is bound to get some attention. The amount of attention is bound to grow upon learning that the company is backed by, and works with, the likes of Bp, its AI technology is based on IP from NASA and Caltech, and it looks like the closest thing to the vision for AI in the real world today.

Beyond Limits, an industrial and enterprise-grade AI technology company active in energy, utilities, and healthcare, today announced a milestone Series C funding round with $113 million closed and another approximately $20 million committed. This round is led by Group 42, a prominent AI and cloud computing company, and Bp ventures, an existing two-time investor and customer of the company.

ZDNet caught up with Beyond Limits CEO and Founder AJ Abdallat to discuss business, technology, and applications.

From NASA to Bp

Beyond Limits (BL) was founded in 2014, and states that its mission is to create automated solutions with human-like powers of reasoning that amplify the talents and capabilities of people. The mission statement goes on to add that BL specializes in complex challenges in extreme environments.

This is a direct reference to the company’s origins in NASA’s JPL AI Reasoning Laboratory. As Abdallat pointed out Caltech operates JPL, and as part of Caltech’s technology sharing and commercialization program, the technology used to drive autonomous decision-making beyond the reach of human experts in space is also used to address similar needs on Earth.

BL aims to support “mission-critical systems with acute situational awareness in real-time, predictive analytics, domain expertise at the edge, and instantaneous human-like reasoning to make informed decisions and take meaningful action.” Abdallat noted that some of the key people who built such capabilities and technologies at Caltech JPL are now part of BL.

BL lists sectors such as energy, power plants, refineries, oil and gas solutions, power and natural resources, manufacturing and industrial IoT, as well as healthcare, finance, automotive, and logistics as its field of operation. These are high-impact, high-valuation sectors, while many of them also seem close in nature to the original field of application for BL’s technology.

Abdallat noted that, while the technology is agnostic to applications or sectors, it was really built for the most complex and demanding applications in space. Those capabilities are now being applied to the energy utility power sector and other applications in the industrial space, in which Abdallat said he is a big believer.

The fact that BL works with Bp also plays a role here. Abdallat cited a statistic mentioning that most refineries and manufacturing facilities operate about at 45% efficiency, going on to add that using AI can enhance and increase that significantly, therefore creating a tremendous opportunity.

This is a line of reasoning we have also seen in Amazon’s partnership with Bp. Others, like Google, plan to stop making AI tools for oil and gas firms. A related point that we saw raised is the aging personnel in the oil industry: EY reported that the Society of Petroleum Engineers found that nearly 54% of its members are over 55 years old.

As these seasoned experts with decades of experience retire, Abdallat noted, the energy industry will be left with a knowledge gap. Cognitive technologies help close this gap by encoding domain expert knowledge into the system and democratizing that knowledge, he went on to add. But there are some counter-arguments to this as well.

First, knowledge owned by employees is over time transferred to the company. That may be a good thing for the company, but not necessarily for employees. Second, in the same way that habitual use of navigation technology such as GPS erodes human navigation skills, it’s conceivable that habitual use of AI could erode human reasoning skills. Abdallat begs to differ, as he believes that codifying human knowledge will preserve it:

“I know we talk a lot about machine learning and deep learning and data and the abundance of data and data-driven models which are a phenomenal capability, but we cannot undermine human knowledge and human expertise.

The approach that we’re advocating here is preserving that knowledge and then not just stating that, but evolving that, allowing the younger generation, the younger scientists and engineers to actually benefit from that but also enhance it and improve on it.”

Hybrid AI: machine learning, reasoning, and physics

Apart from the different points of view, this also touches upon the core technology itself. Abdallat mentioned that one of the capabilities BL’s AI model has is the ability to reconcile contradicting opinions. Through continuous enhancement, the system will figure out what was the best suggestion, Abdallat said.

Speaking of contradictions brought up another interesting topic: COVID-19. This is a fluid, evolving topic, with a constant influx of data, models, expert theories, and opinions. BL recently announced a Coronavirus Dynamic Predictive Model. The forecasting model was built to help healthcare and governmental leaders predict the impact of COVID-19 on people, medical facilities, and regional recovery plans.

Abdallat noted they felt an obligation to embark on this, teaming with some world-renowned medical professionals. The idea was to predict the impact of COVID-19 on resources, whether that was in hospital capacities, ventilators, or equipment. Abdallat added that while the situation was evolving on daily basis, the system was delivered to hospitals in South Florida in about two weeks. It was also made available to policymakers.

One of the evolving assumptions Abdallat mentioned for the COVID-19 model was the infection rate of the disease. While it was initially assumed this will remain static, this was revised to account for factors such as social distancing and mobility. At any rate, the system was purely based on machine learning and statistical analysis.

But this is not the case for all of BL’s systems. Abdallat confirmed that BL is using a hybrid approach to AI. Part of it is based on machine learning, and statistical, numerical approaches. As machine learning does not always generate explainable results, and BL wants to provide explainability for its AI models, it takes a popular approach in this field.

It translates “black box” machine learning models to decision trees. Decision trees are a type of machine learning model whose decisions can be traced, thus lending itself to explainable AI. And then there is also a rule-based, knowledge-oriented, symbolic AI element in the system as well.

This is used to codify knowledge, as alluded to earlier, and provide human-like reasoning, which numeric approaches lack. This is a very interesting mix, to which Abdallat added something on top — physics:

“It’s not just rule-based system. It’s also implementing, deploying a physics based model — mathematics, equations. Experience, observation and expertise that basically is a representation of human knowledge and human expertise.”

To the best of our knowledge, this is not something that has been widely deployed in the industry elsewhere. Abdallat noted this is an approach championed by the AI team at the Jet Propulsion Laboratory, to be able to deal with phenomena that were not previously encountered. BL, picking up on Caltech’s legacy as one of the leading producers of IP in the US, is in the process of filing patents for applications.

From hybrid AI to artificial general intelligence?

BL’s approach to building hybrid AI systems seemed reminiscent of the one we have seen championed by Gary Marcus — even before getting the tip on the elements of physics incorporated in it. With this, the correspondence becomes almost 1-1. Although that is not to suggest that the 2 parties are working together, Abdallat was aware of Marcus’ work and concurred in this assessment.

Marcus’ main point is that systems such as GPT-3, great as they may be, inherently lack reasoning capabilities. They may be trained on tons of data, and have very elaborate models, but without the insertion of some “bias,” such as reasoning, spacial and temporal cognition rules, they are not, and cannot be, truly intelligent.

While Abdallat seems to share this view, he also expressed something we have not seen Marcus express: The conviction that Artificial General Intelligence (AGI) would likely be a reality by 2040 or sooner.

When discussing what is it that he bases this conviction upon, Abdallat said that for someone who’s a big fan of science fiction movies, it would be nice, but it’s not needed. Abdallat went on to add that today we already have a collection of independent AI system that are experts in their own field, and increasingly outperform humans:

“As you start connecting them together, you have a network of intelligence society, something that another great mind in AI, Marvin Minsky, was advocating. Now you’re beginning to look at insights and correlation — a purely independent agent would have not been able to achieve that.

But as you’re connecting these agents together, there’s definitely more insight more knowledge and more capability that can assist humans. And if you put an executive AI agent on top of that, it’s something that can yield tremendous value. I think if we put our energy in such systems, we’re going to make strides in a lot of sectors and a lot of applications.”

More often than not, talk about AI is highly speculative and hype-driven. This one is one of the times we have the feeling it may have some grounding. BL’s $133 million funding will be used to expand business both in the US and abroad, including the launch of Beyond Limits Asia, with further expansion across Europe, the Middle East, and Africa. The funding will also accelerate Beyond Limits’ Cognitive AI application development and SaaS product portfolio and fuel the Beyond Labs R&D program.