18 Oct Slow down! Is regulation putting the brakes on the hottest technology?
A few weeks ago I was asked to participate in a panel discussion in front of an audience of data experts, software vendors, agencies and customer organisations. The subject was artificial intelligence and cognitive technology. It was a lively discussion, ranging from the changes in working conditions brought about by the Industrial Revolution to audience views on the best ways to convince companies to adopt new technologies.
One particularly thorny question was posed; how, given increasing data privacy regulation such as GDPR, should we work with cognitive technologies which use sophisticated and often byzantine algorithms to make decisions for our customers? GDPR has itself been the subject of this column and should, rightly, be consuming the thoughts of organisations with customers and employees. However, when cognitive technologies are employed to assist a customer’s journey how will your organisation respond if it is asked to explain a decision?
One of the specific provisions of GDPR is a “right to explanation.” However, machine learning solutions produce results, in part, by ascribing more weight to certain factors and then making calculations across large datasets using mathematics that even the technology’s vendor will struggle to explain.
GDPR also calls for organisations to prevent any form of discrimination based on personal characteristics like race, gender or health history. We may assume that cognitive software won’t be delivered with discriminatory factors baked in – but can you be sure that familiar or unconscious biases won’t emerge in the technology? After all, these are learning technologies that require guidance from human teachers.
Spotting these outcomes behind an arcane technology solution will be very hard but I suspect that will not be a well-received defence.
Essentially what we are seeing articulated here are some of the reasons that these technologies are not pervasive yet – in spite of the ambitions and considerable marketing budgets of the technology developers. Companies have legitimate concerns about the use of these new technologies and the veil of mathematical impartiality does not sit comfortably with organisations seeking to enhance their customer experience.
Our good old friend, Governance, is critical in this arena. Understanding what data you have and what decisions are being made using that data will be key to ensuring that the undoubted benefits of cognitive technologies do not create more trouble than they are worth. It might be less sexy than the promise of the technology itself but the responsibility to do the right thing by your customers – or at least explain your decision – is paramount.
More than just semantics!
At the same panel event another interesting question was posed;
How should we best sell the benefits of AI and cognitive solutions to our own companies? My response was a cautionary tale but, I hope, offers some insight on how to propose the adoption of these technologies.
An early encounter with the CIO of a household name company left a lasting impression on me as we attempted to position a cognitive and AI solution. It turned out that the CIO had a PhD in the very subject! We spent 80% of the initial meeting discussing his strongly held view that cognition means thinking and that, unless I was proposing the single most significant development in the history of mankind, my solution would not be thinking. The remainder of the meeting was used to explore his contention that, similarly, current AI is not in fact intelligence in any meaningful way but that he was willing to accept that brute force computation might produce some useful insights. Moreover, machine learning does not mean, apparently, learning in any traditional sense.
The second meeting was with senior marketing professionals and we had certainly learnt our lesson. That discussion was far more fruitful as we discussed business challenges and how the technology solution’s capabilities might help overcome them. The label still said cognitive but the tenor of the discussion was a more traditional one!
Perhaps the lessons are very old: know your audience; focus on your organisation’s challenges and the value of the solution. Either way, please talk to us and let’s share ideas.
Written by Dominic Bridgman
get in touch to find out more