This is a guest blogpost by Neo4j’s Amy Hodler. Imagine the possibilities when AI can handle ambiguity, she says.
Business and governments are turning to Artificial Intelligence (AI) to automate and improve their decision-making and uncover multiple opportunities. The problem is that AI has been effective in powerful, but narrow, contexts, on applications where it can do one thing extremely well. But AI systems don’t readily flex to new situations at the moment and certainly don’t offer a nuanced understanding of complexity.
An increasingly promising approach for teaching AI systems to be more intelligent is by extending their power with graph technology. Why? Because graphs help us better understand and work with complexity, as it’s a technology uniquely suited to managing connections.`
Lack of context equals poorer understanding
Context is the information that frames something to give it meaning. We humans deal with ambiguity by using context to figure out what’s important in a situation, then extend that learning to understanding new situations. Consider autonomous cars – teaching them how to drive in rainy conditions is difficult because there is so much variability. If the autonomous vehicle’s AI needs to see every possible combination of light and weather conditions, it’s a huge challenge to train it for all possible situations. If the AI is supplied with connected, contextual information (rain and night plus night and temperature plus temperature and bridge, etc.), however, it is possible to combine information from multiple contexts and infer the next action to take.
Graph software’s ability to uncover context is being used to make AI and ML (Machine Learning) applications more robust. This means outcomes that are far superior to results from AI systems that rely on disparate data points. That’s part of why between 2010 and 2018, AI research that mentions graphs has risen over threefold, from less than 1,000 to over 3,750.
One example where graph enhanced AI can have a high-value impact today is fraud. According to Stratistics MRC, the global fraud detection and prevention market was valued at $17.5 billion in 2017, and is expected to grow to $120 billion by 2026. We can use graphs today to find predictive elements in data (feature engineering) that are highly indicative of fraud and then use those features to increase our machine learning accuracy.
In another area, knowledge graphs are being used to help AI systems make smarter decisions by dynamically adding rich, connected data. For example, with the eBay App on Google Assistant, a knowledge graph holds the probabilistic models that aid their AI in understanding conversational shopping scenarios.
Finally, an issue in AI is avoiding the danger that we will automate human flaws and biases, creating systems that efficiently discriminate against certain groups.
Context-supported AI could also help accountable humans better map and visualise the AI decision paths. This helps reduce the ‘black box’ aspect of decision-making that can reduce confidence in why AI systems reached a particular conclusions/recommendations.
It is our belief at Neo4j that context should be incorporated into AI to ensure we apply these technologies in ways that do not violate societal and economic principles. We know that context can help guide AI systems – and we’re so convinced of this, that we have submitted a graph and AI proposal to NIST, the US government’s National Institute for Standards and Technology, which is creating a plan for the next wave of US AI government standards.
Could graph technology help us become the beneficiaries of a more accurate, insightful, and responsible technology of the future? For more and more of us, the answer is, yes.