• by Jag Duggal , Columnist, October 18, 2017

For a while now, our industry has understood that AI can make us money. Advertising leaders are embracing the fact that AI, done right, can also improve human creativity. Agencies have been striving to incorporate AI into the fabric of their organizations, seeking to forge more efficient and fruitful partnerships, and to connect people in more productive and resonant ways.

The controversial catalyst

Not everyone views AI as the great enabler of human creativity. Although some industry leaders urge us to choose hope over fear on this topic, tech luminaries Elon Musk and Bill Gates have cast AI as an urgent existential threat. Musk’s crusade has driven him so far as to lobby for that anathema of entrepreneurship — government regulation, as well as to create his own open source development project as a system of checks and balances. Similarly, Stephen Hawking warns AI will soon equal and exceed intelligence in the human brain, and predicts utopia or destruction: Either the curing of disease and eradication of climate change, or the neat disposal of inconvenient humanity.

The student learns from its teacher

Most veteran AI scientists see potential doomsday scenarios as quite a ways off but note AI presents other pressing human welfare challenges. Like, how we can keep human bias and prejudice out of our algorithms. Rachel Urtasun, an AI researcher recently hired by Uber to head a high-profile, advanced technology group, champions this issue. She points out that people train networks to mirror human thought processes. And that, often, we pass on not only our powers of perception but our misconceptions as well. Recent studies show programmers do indeed embed racial and gender biases into network trainings, often in unconscious ways. Empirical evidence — from Siri’s trouble answering many questions about women’s health, to racial profiling in recruiting automation — demonstrates the dramatic susceptibility of “objective” reinforced machine learning to human prejudice and opinion.

Although there’s no turnkey fix to the AI bias problem, experts suggest an array of tactics can help avoid its introduction and counter its effects. Community policing with a feedback loop can help raise network trainers’ awareness of issues and give them guidance on how and where to correct. “Algorithmic auditing” also measures for fairness, legal discrimination, and meaning; network developers can apply these emerging techniques. And the practice of open-sourced AI — the collaborative and shared academic development of Artificial Intelligence (think Elon Musk’s OpenAI initiative) — supports the evolution of best practice standards to eliminate the seeding of bias.

All ‘AI’ is Not Created Equal

Another possible obstacle to the fulfillment of AI’s creative promise? The quality of that AI.

Architects of large scale creative community are definitely barking up the right tree. Adaptive neural networks do have the potential to compare, associate, identify and predict how various team members might best contribute. When properly constructed and directed, machine learning can effectively scale to process massive amounts of data, giving projects the potential for global impact; and provide exceptionally granular insight, drilling down into the capabilities of each community member and placing them in the ideal context to shine.  

But there are some tall order prerequisites for this kind of success: a massive amount of timely data, a massive amount of computer power, and expert supervised learning. Without the data, AI cannot discern patterns and cannot learn. Without the computer power, the system lacks bandwidth to house data and make connections at the high speed needed for effectiveness.

Art is In the Machine?

With recent initiatives, global stakeholders in the advertising space seek to improve human decision-making and artistry through AI-driven collaborative communities. AI has the potential to significantly support this endeavor. But communities are made up of complexly diverse individuals. Architects must build safeguards into their systems so that dynamically goaled, AI-powered initiatives can objectively and holistically “see” each constituent —and assess their potential contribution — not process them through a web of human assumption and bias. They also must have the prerequisite resources in place to build functional AI.  The better industry leaders meet these goals, the more successfully AI can support, and improve, collaborative creativity. By objectively highlighting the diverse strengths and talents of all, machines can move our projects forward in a more productive, and more uniquely human, capacity.