By Andre Petheram, Strategy Consultant, Oxford Insights.
With so much media, popular and political attention increasingly directed at artificial intelligence (AI) and machine learning (ML), and a growing sense that automated systems will have deep and lasting effects on our lives and societies, it is becoming crucial to teach people about AI’s genuine potential, and its stubborn limits.
It is no longer just programmers in Silicon Valley startups or academics in Oxford or Stanford who need to know about AI.
To get beyond the hype and the fear, and just a little anxious about the array of complex philosophical and statistical concepts that might await, I recently worked with our partner infoNation, who have supported Southampton Data Science Academy in developing and launching a portfolio of data science and AI related training.
Much of The Academy’s content is targeted at helping less technical teams get a clearer understanding of the opportunities, limitations and practical applications for data science and AI.
As a Strategy Consultant at Oxford Insights, helping governments around the world use AI to transform public services, I need to be alert to the ways individuals and teams in large organisations assess the case and need for AI-based transformation.
With that in mind I decided to go deeper into AI and ML and signed up for 60 hours of learning and four assessments with the AI and Machine Learning for Business course. Subtly, this course has changed the way I think about AI.
Machine learning is not magic
Before taking the Machine Learning for Business course, I had a tendency to see machine learning as a solution without a problem. Focusing on how machine learning will change economic, social and power relations often treats (or imagines) it as something that is already up-and-running.
I valued machine learning for its affordances and ingenuity, seeing it as involving a range of deeply clever algorithmic systems with sometimes enormous analytical and predictive powers. However, this obscured the business or operational needs that should justify introducing AI in the first place.
The course showed me that AI will probably only be useful if your organisation has a whole lot of well-ordered, frequently-collected and trustworthy data, if there is a clear and material limit to what humans can already do with this data, and if you can provide evidence for why understanding or presenting your data in new ways will benefit you. Even then, it will probably not be clear exactly which algorithmic system is going to work best for what you want it to do.
Employing Machine Learning just because it is innovative and exciting, or for the purpose of simply knowing more about what is going on in your data, is unlikely to work well. AI is successful when it has a well-defined and bounded role: scanning documents; detecting anomalies in transactions; telling you what other shoppers have bought. Moreover, to understand AI in this way is to make it much less magical or scary.
Machine learning is not necessarily the most efficient option
Southampton Data Science Academy’s course showed me that, even if your organisation is actually in a position where AI will be theoretically useful, it is not necessarily going to be cheap or efficient.
Anxieties about how algorithms might replace human workers are valid, but sometimes skip over the possibility that it could still be cheaper to employ humans than to pay to develop, train and implement a machine learning system. AI still requires high computational power and human ingenuity.
This also affects what kind of algorithmic system is going to be available to your organisation. Naive Bayes, for example, often used for classifying text, is likely to be much less resource-intensive than a deep neural network.
Where Google, with its first-mover advantages and vast data stores, can afford to invest $13 billion in data centres in 2019, smaller organisations - including many governments - may have to be much more humble in their ambitions. If this reassures people about the future of their jobs, this is by no means a bad thing.
Finally, with cost comes energy use. As the climate crisis deepens, organisations will have to confront the desirability of implementing algorithms that might take ‘five times the lifetime emissions of an average car’.
Lessons for governments
With the AI and Machine Learning for Business course inspiring a more realistic, problem-oriented and commercial attitude towards AI than I have previously held, three lessons jump out concerning AI’s role within government, our own area of expertise at Oxford Insights.
Firstly, consider the quality and extent of your data. You might need to systematise data entry and collection practices across and between departments before AI becomes a possibility. Fortunately, this will be beneficial in itself, even if the analysis you then perform is limited by everyday human capacities.
Secondly, proceed cautiously. If you’re interested in AI and ML, examine the specific problems that you face in your everyday work with AI as one solution among many. If, only then, you cannot escape the idea that AI is what will help you, then you will have a strong business case for implementing it. Even then, you will have to experiment with different approaches before getting it right.
Finally, don’t be afraid of AI. My early anxieties were misplaced: you don’t need to have a background in maths, data science or statistics to learn what problems it will help you solve, how to justify your approach and how to oversee a successful implementation. AI is not magic and it is not always cheap, but it retains the potential to accelerate and transform how governments operate. With proper training and education for civil servants and officials, this change will be for the better.
If you want to understand AI’s real potential, and its true value to your organisation, then Southampton Data Science Academy’s Machine Learning for Business course will put you firmly on the right path. I strongly recommend taking it.