On 3 June 1997, during a warm-up match between Brazil and France, Roberto Carlos stunned the world with one of the most spectacular free kicks in football history. It was a 30 yard kick. Carlos took a long run-up of 20 yards to take that kick. The ball was kicked at a speed of 136 kmph. The ball went around the entire French defence. Goalkeeper Fabian Barthez thought the ball was going off the pitch into the crowd. That’s why he didn’t move. But the ball swiftly and magically curled back into the goal. Football experts and even physicists have studied every aspect of this kick in great detail. But years later, in an interview with ESPN Brasil, Roberto Carlos said, “To be honest, to this day I don’t know how I did it.”
Explainability is the ability to explain why a person or system reached a particular decision. As businesses increasingly rely on artificial intelligence (AI) systems for faster decision making, there is an increasing emphasis that AI systems must be interpretable. What data do they use? How do these models draw their conclusions? Are they free from all prejudices? A McKinsey study found that organizations that establish digital trust among consumers through practices such as making AI interpretable are more likely to see double-digit growth in annual revenue and earnings. So is AI interpretability a desired goal? But the more important question is whether humans, who insist on the interpretation of AI systems, can explain themselves.
For a long time, it was believed that human behavior was the result of conscious, rational processes. So any explanation humans gave for why they behaved the way they did was almost always considered a sufficient explanation for that behavior. But with the advent of neuroscience, this belief began to change. It is now clear that 99.99% of the 11 million-bit processing capacity of the human brain operates at an unconscious level. That’s why most human behavior is done below the threshold of consciousness. It is quite clear that humans are not capable of explaining why they did what they did. Think of Roberto Carlos’ kick.
The earliest stages of any human learning involve conscious processes. As learning progresses and as a person becomes an expert, this person no longer needs to think consciously to act as an expert. This can be done unconsciously. But what happens if a specialist tries to think consciously about his performance? No doubt, this would improve its interpretability. But the moment man tries to bring his unconscious knowledge to the conscious level, he tends to ‘suffocate’. By insisting on bringing better explainability to AI algorithms, will we ‘choke’ its efficiency?
There is no doubt that technologies enabling AI-system interpretability can more quickly reveal errors or areas for improvement. This will make it easier for machine-learning operations teams monitoring AI systems to efficiently monitor and maintain those systems. It is believed that interpretability helps organizations reduce risks. AI systems that violate ethical norms, even if unintentionally, can ignite intense public, media and regulatory scrutiny. If the algorithm can be interpreted, the legal and risk teams can use the explanation provided by the technical team to establish that the system complies with applicable laws and regulations.
Explainability helps improve risk mitigation. Data from the black boxes of crashed aircraft has helped to better understand the causes of those accidents and prevent similar accidents. But should greater interpretability be at the expense of better efficiency?
Despite recent advances in neuroscience, we are a long way from explaining the ‘why’ of any human action. Humankind has progressed not by improving human interpretability, but by taking greater accountability for actions. Can we focus on making AI systems more accountable for the quality of their outputs than on making their algorithms more interpretable?
A study by academics from Harvard University, the Massachusetts Institute of Technology and the Polytechnic University of Milan suggests that over-interpretation of AI systems can lead to some unique problems. Tapestry’s portfolio of luxury brands employees were given access to an AI-based forecasting model. Employees were more likely to rule out models they could understand because they were mistakenly sure of their own intuitions.
It will come as no surprise to those cognizant of the intricacies of human behavior that interpretability influences the adoption of AI systems. The sense of ignorance among the masses that an elite Latin language created in the West and Sanskrit in India went a long way in increasing popular acceptance of the superiority held by a priestly class in these societies. Feelings of ignorance and expertise are closely linked. If so, are we compromising better adoption by improving explainability of AI systems?
There is a part of human behavior that is considered explainable. Behavior that is carried out in strict accordance with instructions, for example, where the instructions themselves provide a clear explanation of what has been done. The generation of strict instruction-based behavior in human history is called slavery. Does today’s growing clamor for the interpretation of AI systems reflect a desire for AI machines to be enslaved by humans? Will the human-imposed shackles of interpretability curtail the freedom of AI systems to freely express their innovative capabilities?
catch all business News, market news, today’s fresh news events and Breaking News Update on Live Mint. download mint news app To get daily market updates.