If we’re serious about building Artificial Intelligence (AI) that makes choices which are as smart and informed as the ones we humans make, it makes sense to look into an important environmental factor we use to inform our decisions: context.
Essentially AI needs context in order to mimic human levels of intelligence. Context is the information that frames something to give it meaning, after all. For example, a person who says “Get out!” may either be expressing a friendly note of surprise, or angrily demanding that someone leave the room. However, you can’t decide that simply by reading the text alone.
We solve this by using context to figure out what’s important in a situation, and in turn, how to apply those learnings to new situations. For robots to make decisions closer to the way humans do, they will need to rely on something like context as well. After all, without peripheral and related information, AI will require more exhaustive training, more prescriptive rules, and be permanently limited to more specific applications.
The problem is that context has to be discovered. AI scientists try to avoid the problem by building narrow, but powerful, systems that do one thing extremely well. Narrow AI is focused on performing one task very well, such as image recognition, but it can’t scale horizontally and doesn’t offer anything like human-level understanding of complexity – around planning, language comprehension, object recognition, learning, or problem solving.
Connecting Data and Defining Relationships
One way to provide AI applications with context is by extending AI’s power with a graph approach to working with complexity. If you’re not familiar, a graph database is a way of managing data that is quite unlike the traditional relational store approach of Oracle or Microsoft SQL Server. It also differs from NoSQL approaches such as MongoDB. Gartner has identified enterprise interest in the graph database as one of its current top trends and we are also experiencing the ’Year of Graph’.
Graph is applicable to a wide variety of use cases, from Amazon-style shopping recommendations to fraud and money laundering detection. Increasingly, graph technology is being used to power AI and ML (machine learning) initiatives. That’s because its native architecture provides the missing context for AI applications, with early results suggesting outcomes far superior to results from AI systems that don’t incorporate this background. Graph technology connects data and defines relationships, and by enhancing AI with related context graph technology, it offers an effective means to empower the development of sophisticated AI applications.
Consider the case of self-driving cars. Teaching autonomous vehicles to drive in rainy conditions is difficult, because there is so much variability in wet conditions. It would be impossible to train them for all possible situations, but if the AI is supplied with connected contextual information (rain, light, traffic congestion and temperature), it is possible to combine information from multiple contexts, allowing the vehicle to infer the next action to take.
Graphs provide context for AI in at least four areas. The first is knowledge graphs, which are used to improve decision support and ensure the most appropriate answers for a situation are given. The most familiar use case for a context-rich knowledge graph is Google search, but documentation classification and customer support are also common applications. A context-rich knowledge graph works well for organisations that capture a great deal of knowledge in the form of documents. One example is NASA’s Lessons Learned database, which captures 50 years of knowledge about past space projects.
Second, graph accelerated machine learning uses graphs to optimise models and speed up processes. Current machine learning methods often rely on data stored in tables, but teaching a network using such data is resource-intensive. Graphs provide context for improved efficiency because data in this representation is connected. This enables relationships of numerous degrees of separation to be traversed and analysed quickly and at scale.
Third, connected feature extraction analyses data to identify the most predictive elements contained within it. For example, studies show that your larger friend network is a better indicator of how you will vote than even your immediate friendships. A great use case here is how graph algorithms simplify finding anomalies of hidden communities that might be fraud rings or money laundering networks.
Fourth, graphs offer a way to provide transparency about how an AI makes decisions. This ability is crucial for long-term AI adoption, because in many industries, such as healthcare, credit risk scoring and criminal justice, we must be able to explain how and why decisions are made. Context-supported AI helps an AI’s human overseers map and visualise the decision path within the contextual dataset, removing the ‘black box’ aspect of decision-making that can reduce confidence in the conclusions/recommendations offered.
Making AI More Trustworthy
Neo4j is so convinced by the importance of graphs to AI that we have formally submitted a graph and AI proposal to NIST, The US government’s National Institute for Standards and Technology, which is creating a plan for U.S. AI government standards. Our proposal states that AI and related applications of intelligent computing, like machine learning, are more effective, trustworthy, and robust when supported and interpreted by contextual information that only graph software can provide.
AI that doesn’t explicitly include contextual information will result in subpar outcomes, but graph software, developed as a way to represent connected data, can step forward to help. Let’s use the power of graph technology to enrich our datasets to make them more useful and a better basis for the next generation of AI success stories.