SAN FRANCISCO – AI isn’t going to abolish humanity. In fact, AI can’t work without it. From cleaning up training datasets to augmenting the everyday tasks carried out by human personnel, it is people, not technology, that are the driving force behind AI.
Since 2009, Figure Eight have been building a human-in-the-loop machine learning platform, which they claim significantly boosts both the quality of datasets and the real-world implementation of AI. We sat down with Figure Eight’s Chief Marketing Officer, Randi Barshack, to find out more about the challenges facing enterprises looking to implement effective, practical artificial intelligence solutions in their businesses today.
Offering human-in-the-loop capabilities is one thing, but how can that humans ensure they make the most of an AI platform?
I should clarify what we mean by ‘human-in-the-loop’. One thing that is often misunderstood about AI – even those who purport to be somewhat technical – is that models need to be trained with training datasets that are often created by the human labeling or categorization of raw data. Much as children need to be taught about the facts, feelings, and nuances of the world around them (facts: ‘that is a cat, that is a dog, that is hot’; feelings ‘that was a positive comment, that is a happy person’; nuances: ‘that is hot, that is a good answer’), algorithms need to be trained and humans are required to do this.
Even after a model is in production, these algorithms require constant testing and tuning as the massive amounts of data they are processing changes around them as often as the world around them changes. Humans are required on an ongoing basis to provide the training, input and course-correction required to assure that algorithms achieve the accuracy required to be commercially viable.
How significant is bias as an obstacle to making AI work for all?
It’s not a question of whether when bias- or prejudice- will be an inhibitor to making AI work, but whether it will be an inhibitor to making AI work fairly. To the extent that AI is an extension of the humans that create and train the models at exponential scale, the answer is in the assurance that we can get from knowing that the humans in charge are operating with good intentions and fairly.
Just as with any prior technology, the power can be leveraged by those with evil intentions or even by those who mean no harm, but inadvertently introduce bad.
The answer really boils down to the training data used to train the model. For example, if you have a voice assistant that is trained by a fairly homogeneous set of voices, colloquialisms, accents and utterances, the final product will have difficulty understanding individuals who don’t fall within the demographic of the group/s used to train the model. For commercial or service-oriented applications, this bias will result in lower satisfaction and/or sales.
While that cost is high, the bigger potential cost are in applications that deal with health, live and death decision, legal issues or access.
Why should large enterprises consider AI? How can they start thinking about their own everyday business problems in relation to AI solutions?
I believe that in the long run, this question is going to be akin to a similar question we asked in 1999. Why should companies be on the internet? Now, only 2 decades later, the answer is so obvious; to stay alive. Those who adopted early were able to make strides ahead of their competitors and many who didn’t adapt fell the wayside.
While the widespread adoption of AI is arguably only beginning (I’ve seen estimates that 4% of CIO’s claim to have successfully implemented AI projects), it will become as pervasive as earlier transformational technologies- like the internet.
The best way for businesses to think about AI though, it to think about data. What is the data that drives their business? And not just how do they use data today, but what could they do with the data that is collected if ‘anything were possible’? And even then, many will need to reach a new paradigm of what they even think of as data.
For example, ‘data’ isn’t what we traditionally think of- records of systems or transactions, but data can now be so many other things; a history of every customer service call having ever occured, full-on visibility of an entire supply chain, a video showing every person to ever enter a business etc.
AI is first and foremost about understanding the potential of the available data and what questions to ask of that data. It allows you to process that data at (previously) inconceivable levels of scale, accuracy and speed. Reading millions of medical scans; looking at years worth of traffic patterns for anomalies in minutes; mapping years’ worth of historical buying patterns to a live customer, etc.
How can companies start meeting the challenges of digital transformation in 2019 using AI?
It’s important for companies to know that they don’t need a fully staffed team of data scientists or machine learning experts to get started. While these roles will eventually become as pervasive as developers, there are plenty of off-the-shelf AI technologies available today.
A marketing department could easily get a package using AI to judge the sentiment of social media, an HR department could procure a solution to assess resumes or even judge the ‘fairness’ of how their job descriptions are written.
As with any new technology, companies need to get their feet wet before jumping into the deep end. Finding a quick project and getting to success should be a short term goal of any forward-thinking organization. Use those successful projects to help get others past the emotional hurdle of jumping ahead.
What does competitive advantage look like in the context of AI?
Today, competitive advantage in the context of AI is about being the first (or one of the first) in your industry to go to the next level of scale and automation. But tomorrow, that will no longer be a competitive advantage. That will be table stakes and required to stay alive.
Business leaders must understand the depth and breadth of where AI can add value. Some of my favorite customer stories are in industries like agriculture- places one might normally technology to lag behind. For example, farmers can use robots to spray pesticides in fields; targeting with fraction-of-an-inch precision weeds (vs. crops)., fishermen can detect patterns and disease in fish, satellite imagery can be used to show how and where crop disease is spreading or track migratory patterns.
If these industries are already leveraging AI for competitive advantage, then it’s easy to see how it will touch almost every business or entity imagenable. It’s really no longer a question of competitive advantage in the context of AI, but staying competitive in a world where AI is ubiquitous.
As told to Ciarán Daly
You can learn more about human-in-the-loop AI and Figure Eight’s enterprise solutions at The AI Summit San Francisco, September 19 – 20. Find out more
A consummate storyteller and former filmmaker, Randi is passionate about the left brain/right brain marriage that is tech marketing. With over 25 years’ experience taking companies from scrappy start-up to scale, Randi has helped pioneer and position widespread adoption of application instrumentation, application governance, APIs, and commerce search. She is the co-founder of SAP spin-off TeaLeaf (acquired by IBM) and has built and scaled marketing teams at Mashery (acquired by Intel), xMatters and Mercado (acquired by Omniture).