Industry executives and experts share their predictions for 2022.  Read them in this 14th annual VMblog.com series exclusive.

Responsible AI for a Better Future

By Alicia Frame, Director of Product Management for Data Science at Neo4j

As technology continues to evolve, so does our responsibility to use it ethically. At my job, I'm constantly thinking about ways we can leverage our innovations to leap forward into a better future. Although daunting, societal improvement is possible when we think about both the benefits and consequences caused by how we use technology, and this is particularly true for AI.

AI's rapid growth has propelled us into an exciting and still unknown future - and as creators and users of AI, we're responsible for guiding the development and application of the technology in ways that fit our social values, in particular, to increase accountability, fairness and public trust. Heading into the new year, here's what I expect to see as tech leaders begin to consider how to use AI more responsibly.

Industry/technology predictions

Behavioral predictions will only continue to advance as more models are scored in real time, using live data, to interpret images, speech, and text. Even though growth in real time machine learning has paved the way for functionally autonomous vehicles and voice assistants that can deliver intelligent responses, the complicated behavioral context behind these models are lagging. Next year, we'll start to see organizations notice these diminishing returns and seeking to add back in a human element of review and responsibility. From governmental inquiries to new solutions from startups, ethical AI is on everyone's minds. Even as we see greater investment into more explainable AI and ML, it's still an open question as to whether or not this is where big tech wants to invest time and resources.

While we're seeing increased demands for responsible AI, we're also seeing explosive growth in low code and no code solutions for machine learning. Driven by  a skills crunch, these tools help domain experts build best in class solutions without the need for deep data science knowledge. While this  trend is particularly exciting for the democratization of data science, removing data science experts will only further the need for ethical guardrails to be easy to access and implement.

Societal/use case predictions

Saying that responsible AI is important is easy; actually doing AI responsibly is hard. We'll see widespread adoption of more explainable machine learning techniques, where end users can better see the exact data that went into drawing conclusions. Addressing predictions fast enough to use in production is already possible and frequently used, but the framework to understand them is currently missing. Admitting the problem is only half the battle when it comes to technology ethics. The next step is blueprinting the solutions and implementing them.  

One facet of responsible AI that's often forgotten is the energy that goes into training all those predictive models. As complicated deep learning models have gone mainstream, the fact that they require billions of kilowatt hours and the lack of sustainability in this is often overlooked. Rethinking our current performance gains to be more ecologically friendly might look like using slightly less accurate models that are nevertheless cheaper to train; relocating server farms to places that make use of renewable energy, or even just places that require less air conditioning.. Regardless, as the climate crisis worsens, we're going to need to address the fact that a single NLP transformer model produces more CO2 emissions than 20 people going about their daily lives for a year (Strubell et al, 2019).

Despite all odds, the developer community is equipped to overcome the challenges that lay ahead of us in the new year. The incredible work being done by developers in open source lay the groundwork to solving critical real-world problems like tackling climate change and ethically expanding technological growth.

##

ABOUT THE AUTHOR

Alicia Frame is currently the Director of Product Management for Data Science at Neo4j, where she works on building the world’s first enterprise ready data science platform for graph. She earned her Ph.D. in Computational Biology from the University of North Carolina at Chapel Hill and a B.S. in Biology and Mathematics from the College of William and Mary in Virginia and has over 10 years of experience in enterprise data science at BenevolentAI, Dow AgroSciences, and the EPA.