Many organizations face two inherent AI-related problems: the need to automate at least some processes so humans can complete innovative work, and the fact that scores of existing chatbots are insufficient or error-prone.
Hybrid AI isn't a new concept. The most common definition of hybrid AI is technology that combines symbolic AI (human intelligence) with nonsymbolic AI (machine intelligence) to deliver better outcomes.
Usama Fayyad, chairman of technology and strategic consulting firm Open Insights, described machine learning as an iterative improvement of adaptive algorithms based on training data.
"Whether you are building deep learning or higher-level models [such as] probabilistic Bayesian models, you need a way of getting the right training data, which typically comes from humans providing the right classification or interpretation," Fayyad said. "Human interpretation and labeling are essential for learning systems ranging from machine-learned ranking in a core web search engine to autonomous vehicle training."
As for fusing deep learning methods with symbolic methods, Fayyad drew distinctions between procedural knowledge and declarative knowledge. Procedural knowledge means that humans know how to do something without being able to explain it, while declarative knowledge can be verbalized. For example, many speech recognition and vision problems are procedural in nature, as they are difficult for humans to explain; therefore, they are more amenable to black box approaches, or those that lack transparency.
"I see hybrid solutions being very important, both in dealing with procedural tasks as well as addressing current knowledge gaps," Fayyad said. "In my view, the hybrid solutions are the right approach in almost all cases, especially if we want to explain and understand what the AI is doing."
The resources necessary for effective hybrid AI
Successful hybrid AI examples demonstrate both domain knowledge and AI expertise to solve real-world problems. Without domain knowledge, the solution tends to not fit the problem. Without AI expertise, it may be difficult to understand challenges and what to do about them.
"End users who are the intended consumers of some predictions can be given an active role in a hybrid AI system as the final decision-makers on those predictions and may accept, invalidate or modify each prediction according to their own personal and contextual knowledge," said Fabio Pirovano, chief technology officer at AI-based learning suite provider Docebo. "To be effective, though, the AI system must respect a 'contract' with the end user by making its predictions available to expert scrutiny in an acceptably rapid time frame."
Fundamentally, hybrid AI's effectiveness depends on human judgment for training and optimization in most use cases.
The technological support required most is the ability to record the final decisions made by experts, either for offline analysis by the data scientists responsible for the original AI model or for use as supplemental training data that fundamentally improve the models.
While there are a lot of technological building blocks available, building a coherent end-to-end solution tends to be a patchwork endeavor. Pirovano said he considers the most practical hybrid AI example today to be the human-in-the-loop kind because technological tools necessary for leveraging symbolic reasoning and statistical learning are comparatively immature from an enterprise standpoint.
Having the right mindset is also important, and that begins with identifying a business problem and then using the right technology to solve it -- which may or may not include hybrid AI.
"The most important mindset is one where we have a deep understanding not only of the limitations of algorithms but also the deep dependence on data quality, availability and issues," Fayyad said. "Most importantly, an understanding of whatever solution we come up with will need continuous feedback and rebuilding as the data, domain environment and requirements change."
Common benefits and challenges
Today's hybrid AI examples are most effective when humans and machines do what they do best, respectively.
"Humans are good at making judgments, while machines are good at processing," said Adnan Masood, Ph.D., chief architect of AI/ML at digital transformation company UST. "The machine can process 5 million videos in 10 seconds, but I can't. So, let's allow the machine [to] do its job, and if anyone is smoking in those videos, I will be the judge of how that smoking is portrayed."
Fundamentally, hybrid AI's effectiveness depends on human judgment for training and optimization in most use cases. Otherwise, a chatbot could degrade customer experience, for example. Therefore, the first significant challenge is to staff hybrid AI projects with the right technical expertise. The second is to overcome both the lack of industry best practices for how hybrid AI systems should look and the lack of tools and frameworks to implement those best practices.
"The goal must be to understand when and how symbolic AI can be best applied and matched fruitfully with statistical learning models," Docebo's Pirovano said.
