February 23, 2020
By: Lyat Avidor Peleg
How to Combine Human and Artificial Intelligence for Better Business
Sylvain Duranton, management consultant at the Boston Consulting Group, recently gave a TED talk titled “How humans and AI can work together to create better businesses.” Duranton’s central point is that we need to combine humans to work together with AI, in order to save us from the tyranny of machines.
At Precognize, our slogan is “Human enhanced machine learning,” because we use both human intelligence (HI) and artificial intelligence (AI) together, so this talk immediately caught our attention.
The dangers of a “human zero” mindset
Duranton points to a trend of moving from the recent agile, flexible business models to what he terms an “algocracy.” In an algocracy, instead of humans applying their judgement to respond to each situation as it occurs using agile business methodology, the algorithms make the rules and follow them without human judgement.
“If human judgement is not kept in the loop, AI will bring a terrifying form of new bureaucracy, I call it algocracy, where bureaucracy will take more and more decisions without human input.”
Duranton calls this the “human zero mindset.” As he acknowledges, it’s very tempting. Implementing a “human plus AI” model is long, costly, and difficult. It’s understandable that you’d want to save time and effort by sticking to a human zero approach. But it’s also dangerous.
The costs of a human zero mindset
When AI is left to work alone, it can make some very bad decisions. Sometimes, the consequences of these decisions are just silly, like the person who bought a new toilet seat after his current one broke, and then was pestered with online ads for more toilet seats for the next 6 months.
At other times, these decisions are more serious, such as when a US college put AI in charge of accepting or rejecting student applications. AI rejected the application from a bright student who lived in a postal district where no one has graduated from college, because one of its rules was that students from this postal code won’t succeed at college.
AI follows the rules and only the rules. It can’t make an exception for a committed student from the “wrong” postal district, for example. And what’s more, with machine learning AI makes the rules itself. No one can possibly check all the rules that AI uses, because it’s constantly developing new ones based on past data.
Why do we still need AI?
Perhaps this makes AI too dangerous, and we should stop using it? We don’t think so, and Duranton does not suggest it.
AI is an incredibly useful tool. It’s more accurate than people; it doesn’t get tired, and it doesn’t ever fail to spot indicators. It saves us a huge amount of time and it reduces tedious, time-consuming work for humans. In a predictive analytics context, AI plays a vital role. AI can spot anomalies in a haystack of data light-years faster than a human, and pick up on changes in plant data far earlier, saving plants time, money, and sometimes even human lives. Without AI, predictive analytics would be impossible.
We need a way to weave AI and human intelligence together, so that we can take advantage of each of their unique capabilities.
How HI and AI work together to make the world a better place
A “human plus” mindset
Duranton emphasizes that what we need is a “human plus AI” mindset.
We need to stop thinking “tech first,” and instead think “tech and humans first.” That means approaching algorithmic coding by combining the abilities of data scientists and domain experts, which is exactly what Precognize does.
The benefits of “human plus”
To illustrate this mindset of “human plus,” Duranton shares the experience of a famous fashion house. It wanted an AI model that would be more accurate at predicting fashion trends than its (very successful) human buyers. The process was eye-opening.
The first model they used reduced buying errors by 25%. This was excellent, but the fashion house was wise enough to know that their human buyers had insights that can’t be learned by ML models using past data, and they wanted to keep that in the picture.
Their next attempt involved overlaying human intelligence over their new AI model, by having humans go through and correct the model’s results. This was not a success, and errors rose again by 75%.
Finally, the fashion realized that they had to start all over again with a totally different, human-plus approach. They sat down to create a richer model, where AI draws on human domain knowledge. It took them a full year to complete this project, but the results were astonishing. Errors dropped by 50%, twice as much as when only AI was involved. It’s clear that human-plus projects aren’t just about blending AI and human knowledge. It’s also important how you do it.
Precognize’s “human plus” engine
Precognize weaves together human input and AI similarly to what Duranton recommends. We create a rich model of every plant by using human domain knowledge, and then we introduce AI to draw on that knowledge.
AI makes the analytical process faster, spotting anomalies before any human could notice them, but human knowledge is vital in reducing noise so that the AI machine can recognize anomalies against a complex background, and increase the accuracy of AI predictions.
Only HI can actualize the potential of AI
Citizens in developed economies already fear an algocracy. A survey of 7,000 people’s attitudes to AI found that over 75% expressed real concerns on the impact of AI on the workforce, privacy, and the risk of a dehumanized society.
As Duranton points out, a human plus AI mindset is the only way to bring the benefits of AI to the real world, so that we can enjoy the capabilities of AI without suffering from its weaknesses. He emphasizes that we must continue to reward and train human experts.
“Data is said to be the new oil, but believe me, human knowledge will make the difference, because it is the only derrick available to pump the oil hidden in the data.”
Precognize takes this message to heart, which is why our model combines HI domain knowledge and AI predictive models to leverage the true value of both types of intelligence.
What’s your opinion?