Taming the (AI) Beast

By Arohi Jain, Head of Strategy, The Future Society

Artificial Intelligence (AI), broadly defined as the ability of machines to ‘think’ in ways comparable to that of humans, is hailed the most revolutionary technological phenomenon of our time. With massive implications that cut across society, industry domains and geographies, the likes of Sundar Pichai, Google CEO, have gone as far as claiming that AI will be “more profound” in its impact than was electricity or fire.

With the promise of improved productivity and efficiency for companies, AI has the potential to unlock vast economic upside globally. From improving supply chain processes, such as inventory management and demand prediction, to accurate forecasting of when machines and equipment might fail, firms can better manage their operations to optimize profit. McKinsey & Company estimates that such improvements from AI embedded in supply chains and manufacturing can tap into $1.3 – $2.0 trillion of value. For context, in 2016, the U.S. GDP totalled $18.6 trillion, indicating at least a 7 percent expansion in value from applying AI to one specific domain.

Moreover, it is not only fields that appear to be susceptible to automation that will be impacted by AI, but also areas where currently a “human touch” is considered imperative. For instance, businesses operating customer help services through call centres and email queries could gain 20 percent more return on investment through greater efficiency with the help of AI chatbots, better managed employees through AI generated feedback, and specific but automated email responses through enhanced Natural Language Processing capabilities. Although it is difficult to envision the delivery of such services without any humans in the process, AI is certain to act as a complementary comrade to humans.

In the October 2017 forum discussion hosted by The Future Society at the Harvard Kennedy School, Jason Furman, former Chairman of President Obama Council of Economic Advisors, claimed that “People are working more but not producing very much, which is exactly the opposite of what you’d expect from Artificial Intelligence. So maybe the robots aren’t coming fast enough.” There remains much value to be gained from the advent of the AI revolution, with the ultimate promise of greater human prosperity.

But the grass isn’t so green. While it is clear that AI has significant economic potential, the untapped value claims must be broken down to better understand the implications; who will gain from this value; who are the winners and losers; what are the negative externalities caused by automation, algorithmic bias and scaled use of black box technologies across society?

A real measure of economic value must be a balanced one, which takes into account the costs generated by embedding AI in our society. One area, which has garnered critical attention, is the impact of AI automation on the human workforce. If we were to observe the consequence of using AI in supply chains or customer services on unemployment, the net impact on the economy looks very different. Unemployment reduces household disposable income, diminishing consumption, a key component of GDP. Moreover, long-term unemployment reduces morale and can trigger negative consequences including depression and substance abuse. While several Economists have attempted to model the impact of AI-enabled automation on unemployment, the net effect on the economy is complex, difficult to measure and subject to unproven assumptions, such as rate of technological uptake and the relative costs of labour and capital. Although difficult to measure, these effects must be accounted for when assessing the real value that AI will generate and eradicate.

Additionally, claims of the billions and trillions forecasted to be created through AI are often ambiguous as to who will ultimately reap this value. It remains quantitatively unclear how different geographies and demographics will be impacted by AI, and whether the emergence of AI superpowers such as US and China will create an irrevocable global economic divide. One possible consequence is heightened inequality in our society, as the value generated by AI is guarded by technocratic elite.

Finally, wide scale application of AI black box algorithms presents a critical challenge in assessing the potential negative externalities, including economic losses. Training data, which AI algorithms rely upon to learn and replicate the output at scale, is based on human actions that are inherently biased.  As a consequence, AI algorithms may disproportionately disadvantage specific sections of our societies. Without explainability and transparency into how AI algorithms arrive at a given output, which is then used for decisions in justice systems, hiring, healthcare diagnosis and access to financing, the inertia of AI could lead to intense prejudice and inequality. When a certain ethnicity, gender or age group is repeatedly experiencing disadvantage in access to vital services such as employment, medical insurance, home loans or fair trial, the economic contribution of such groups will degrade significantly. This in turn lowers consumption and output in the global economy.

To fully understand the economic implications of scaled use of AI, we need models that simulate the broader impact on society, with a keen eye on the potential downside consequences. AI holds vast hope for humanity, but, collectively, we need to tame the runaway hype of value it will generate through a wholesome and cautious approach.