New perspectives on artificial intelligence and machine learning
Data is one of the most important assets a business has, and organizations around the world are increasingly taking advantage of the data they hold. However, how organizations understand data is changing, with an increasing focus on artificial intelligence and machine learning to help deliver new business insights from big data that humans cannot obtain themselves.
In modern BI, there is value in data being interpreted by both humans and computer algorithms, and both are essential for organizations to utilize experience-led and data-driven insights. While BI allows you to answer what has happened, AI allows you to predict what will happen. Both are superpowers, and when coupled with a human’s cognitive ability, a step function increase in productivity occurs.
A human’s cognitive ability and skillset are fundamentally different from a computer’s. Both are better at certain things: AI can outperform humans on memory and pattern recognition, but a human can outperform a computer on virtually every other cognitive function, including reasoning, association, concept formation, action, and mental imagery. When AI’s capacity for pattern recognition is combined with a human’s reasoning, human decision- making is greatly augmented.
Mike Bugembe, author of Cracking The Data Code, believes there has been too great a shift to machine learning that organizations are forgetting to include humans in their decision-making processes. “My fears are, with all the excitement around the machine and what data promises, we’re shifting the responsibility onto the machine, because we think of it as the ultimate authority,” he says, in a conversation with Donald Farmer, principal at TreeHive Strategy, as part of the ‘Data Curiosity: Do Data Differently’ podcast series.
“Whilst we increase our sophistication in the machine and the algorithms that can help us [with decision-making], I still think that [humans] are the better big data machine. There are aspects that we can capture that the machine can’t. The machine can only work with the data it has been given, and not everything we encounter and experience has been coded into a database to enable an algorithm to be trained. Therefore it is missing data, while we have more available. There has to be a fusion between man and machine,” Bugembe adds.
The issue of machines only being able to work with the data they have been given is critical to recognizing the limitations of AI. The cognitive ability of machine learning algorithms is extremely limited today. Machine learning can surpass human- level performance on many tasks relating to pattern recognition because a computer’s perception and memory are superior to that of a human. However, AI systems have yet to develop the ability to reason. Because machines have strengths that people do not and people have strengths that machines do not, we can achieve better performance together as an augmented system than on our own. In this way, we could have a system with a computer’s pattern recognition and memory, and a human’s ability to reason.
Ryan Schrupp, SVP of Enterprise Research at US Bank, highlights how the limited cognitive ability of machine learning algorithms can be problematic in the banking sector. “Negative decisions [algorithms make] are going to dramatically impact our relationship with clients, and we need those relationships over the long haul. No machine can tell me that. I have to rely on [human] managers, who are having one-to-one relationships with these clients, to do the right thing, in addition to what our data is already telling them.”
Where the algorithm’s strength lies is in its ability to enable organizations to scale. Farmer says: “In the old days when you wanted to get a decision from a bank, you typically met with the bank manager, and [their decision about what credit to provide] was very much [based on] your relationship with the manager. But that scenario just doesn’t scale. You can’t do that 1,000 times a day, but an algorithm can. And the great advantage of the algorithm is that the bank can scale its risk, and therefore even out the risk, and therefore potentially give out more credit and more financial flexibility to more people.”
Watch out for algorithm bias
Machine learning depends on the quality, objectivity, and size of training data used to teach it, and, unfortunately, bias can creep into algorithms; via the individuals who design and/or train the machine learning systems, or through misleading “proxy data,”, which is when algorithms use scattered bits of data to characterize people. It is against the law, for example, to make hiring decisions based on race, gender, age, or sexual orientation, but there are proxies for these attributes in big datasets.
“One of the challenges of [having big, historical datasets] is handling the inevitable bias that has grown up in that dataset,” says Farmer. “[Biased data produces AI models] that are by their nature biased… and we’re not looking for an unbiased [model], but we do need to be smarter and more precise about the biases we want to support, and the biases that are unacceptable to us.”
It is difficult for companies to mitigate algorithm bias because it is impossible for a human to pick up on the arbitrary bias that a machine learning model uses to make its decisions. Instead, businesses should prioritize the use of interpretable AI models, which provide a detailed view of how they make decisions across all data, and correct any unethical bias.
Domo’s software helps organizations achieve pragmatic value in light of the complications machine learning and AI can pose. Explainable AI can help businesses understand and interpret predictions made by their machine learning models. They can debug and improve model performance, and help others understand their models’ behavior. Find out more here.
Humans and machines working together
Both artificial intelligence and machine learning turn data into action by analyzing datasets and information to either perform an action or provide suggestions to humans to act on something.
Making intelligent decisions based on data requires collaborative interaction between machines and humans across multiple disciplines. Domo’s flexible app-building framework offers organizations the opportunity to leverage their BI and data investments to create intelligent apps that drive action from their data. Find out more here.