In the video, ‘Why AI Is Incredibly Smart — and Shockingly Stupid,’ computer scientist Yejin Choi discusses the current state of Large Language Model-based (LLM) generative AI systems like ChatGPT. She highlights three key problems with generative AI: their inability to understand context; their lack of common sense; and their inability to reason about causality. (She also goes on to draw lessons from Sun Tzu’s The Art of War, which I’ll come back to later.)
Such is the case for generative AI, what I’ll call the lack of "human/common sense."
To cite Janelle Shane, a research scientist and humorist who experiments with generative AI, who said in her 2019 TED talk ‘The Danger of AI is Weirder Than You Think’, “The danger of AI is not that it will rebel against us, it’s that it’s going to do exactly what we ask it to do.”(You can watch her talk here.) This means that generative AI systems may produce results that are technically correct, but nonsensical or harmful in the real world, because they lack context and an understanding of consequences. They do not have "human/common sense.”
Generative AI systems are designed to generate new content, whether text, images, or music (and soon video), based on existing data. This is very different from industrial AI, the purpose-built application of AI for industries that Canvass AI specializes in.
Industrial AI is applied to everyday business activities, such as optimizing processes or improving decision making. It uses existing time-series data for actionable predictions that guide operations or engineers in practical decision-making, rather than for creating new data or more creative and experimental outcomes.
While generative AI is more suitable for creating new content, purpose-built AI applications are more suitable for analyzing existing data and predicting a future state. They can and will complement each other, but not substitute each other.
This, in part, is why industrials have not moved quickly to adopt generative AI – let alone AI. The points that Yejin Choi makes are equally valid for the use of AI in the industrial sector. There are serious concerns for security and safety bound by AI’s shortcomings, which is why the "human/common sense" which is the SME element must remain paramount.
The hype in the use of generative AI in the enterprise has been around how it can or will replace repeatable jobs. Ironically, when it comes to the industrial sector the reverse is true. "Human/common sense" is not captured in any document or graph. It rests in the minds of those knowledge workers who have worked there for years. Leaving people out of the loop is not an option in the industrial sector. It’s important to note that the risks to safety in an industrial process without human oversight can be very high.
Coming back to Yejin Choi and lessons from The Art of War. In Canvass AI’s case:
Know your enemy – If AI is considered to be the enemy, we strive to make it explainable, so that "human knowledge" can be focused on where it matters.
Choose your battles - The cost of a silly mistake from AI in an enterprise setting today is usually harmless, but it takes a very different turn with the type of risks involved in the industrial sector. With Canvass AI, engineers and operators are empowered to solve the problems they confront on a daily basis with a tool that results in significant improvements to production, profitability and the safety of their industrial facilities.
Innovate your weapons - Yejin Choi speaks about using “Crafted Data” and “Human Judgement” feedbacks as better ways of training AI, which are very expensive. In the industrial world, these two key resources already exist in the organization, and are integral to the process of adoption. This way AI can be made an innovative weapon to enable engineers.
That said, generative AI has a place in every, yes, every industry including industrials.
As I’ve said recently in this CCI interview, wider adoption will require a deeper understanding of the implications of AI.
Generative AI is still in its early stages of development, but it has the potential to revolutionize the industrial sector. As generative AI technology continues to improve, it will become increasingly capable of solving complex problems and automating tasks. This will lead to significant productivity gains and cost savings for businesses.
The future of industrials will run operations based on both generative and non-generative AI. For example, generative AI could be used to design new aircraft by creating a large dataset of images of existing aircraft. The AI could then use this dataset to generate new aircraft designs that are more efficient and effective. Non-generative AI can be used to classify data by analyzing images of aircraft to identify their type and model.
Are there pitfalls to be aware of? Certainly, and here are some key things to keep in mind:
Don't fall for the hype around generative AI and its ability to replace jobs or tasks. Instead, focus on how you can leverage generative and industrial AI to augment your worker’s capabilities and skills to improve decision-making and your business outcomes.
Don't trust generative AI blindly or expect it to have a human-like understanding or reasoning. Instead, always verify and validate its outputs and inputs and apply your own context and knowledge to evaluate the results and implications of what you get.
Don't rely on generative AI alone or isolate it from other sources of information or feedback. Instead, integrate it with other types of data and systems and collaborate with other experts and stakeholders to ensure its accuracy and relevance.