The South by Southwest Conference and Festival in Austin, TX is an internationally-known hipster destination, but it isn’t famous just for amazing music: SXSW also attracts thinkers and executives from a global network of industries to sit on panel discussions as varied as the artists who are performing in the tents outside. One of the topics that caught the attention of both McKinsey & Co. and Forbes Magazine was Artificial Intelligence (AI).
For more on AI:
- Deeper Look: How Artificial Intelligence Affects Business Outlook
- Investing in Artificial Intelligence: How Venture Capitalists Think
- Why Artificial Intelligence Won’t Harm the Employment Picture
Nearly 50 panels addressed AI in its various forms, discussing the impacts of AI on democracy, health care, the workplace, product design, and host of other industries, including music and journalism. As Forbes Magazine noted, “the dialogue has moved from broad sweeping discussions to very specific applications of AI. From agriculture, to healthcare, to insurance, to front-office and back-office business processes, to employee and individual productivity, to personal finance, etc., etc., AI is being inserted everywhere, real world applications offering real value and solutions.”
AI Hype Is Justified
In a session called VCs Reality Check for AI Hype, a group of venture capitalists stated that AI is not hype, even if the idea of a general-purpose AI (like the fictional Skynet of Terminator films) is overblown. Instead, narrow AI, focused on very specific tasks, like automated underwriting, health diagnostics, or autonomous vehicles, is a real product worth investing in. The panel of VCs stated that they look for companies that enable workers to avoid routine tasks and be more effective, more creative, and more customer-oriented.
At present, however, it is the large companies—Google, Baidu, Facebook and the like—that are making the greatest strides with AI. The reason for this is that for AI to be most useful, it needs large data sets, which in turn requires access to highly-skilled data scientists who can analyze it. At the other end of the spectrum, according to some studies, 90% of small companies have not engaged in AI, and 70% of companies overall that have attempted to invest in AI have hit roadblocks, one of which is access to data. (We have already considered the progress, or lack thereof, that companies have made in becoming more “data-driven.”) Investors and corporate purchasers may look for startups whose products that can be used to leverage the big data sets that corporations already have.
Healthy Data and Ethical AI
When companies collect and use data, however, they have to be careful in both the algorithms they design and the data they use. Learning machines will only learn what they are taught: that is, data scientists have to ensure their algorithms don’t produce biased results, and also that their data itself is “healthy”—that it doesn’t contain errors or biases through faulty collection processes.
There are ethical considerations surrounding healthy data use as well. Alejandra Parra-Orlandoni, an attorney for McKinsey company QuantumBlack, cautioned that “microtargeting populations to optimize a healthcare treatment or design a municipal service typically involves data sets that include sensitive information, like gender and age, as well as data that serve as proxies for sensitive categories, like zip codes for race. Any of these … could lead to harmful biases or unfair outcomes when leveraged by an AI model.” As the U.S. Housing and Urban Development’s recent action against Facebook for allowing advertisers to target ads away from certain zip codes, this is not a theoretical consideration.
Parra-Orlandoni further warned that “diversity in the demographics and experiences of the team selecting the data sets can also influence bias.” Unhealthy data that is fed into a system to predict (for example) criminal recidivism or creditworthiness, could produce results that unfairly advantage or disadvantage certain groups.
This warning was echoed by Professor S. Craig Watkins of the University of Texas at Austin, whose talk, Diversifying Artificial Intelligence, urged companies using AI to hire more women and minorities for the simple economic imperative of understanding how their products will impact and be used by broad populations; he cited one example of a facial recognition software that did not work on darker-skinned people because the modelers had never shown the algorithm minority faces.
Businesses should be investing in narrow AI, geared toward automating routine functions so that employees can deliver improved and value-added customer service. One barrier to investment is the lack of data that will make AI projects worthwhile, as well as the expense of hiring data scientists. Finally, as firms collect data and build models to analyze it, they need to be careful that the data and the modeling techniques used produce unbiased results that they then use responsibly.