Enterprise machine learning adoption has reached a stage at which the approach, utilisation and distinct job titles are being debated and implemented. And as machine learning becomes more widely accepted across industries, we’re only at the tip of the iceberg when it comes to how companies are embracing this technology.
Machine learning is typically used to enable some level of automation. From a rule-based fraud detection system to a complex model that automatically learns from examples. Machine learning for an organisation is only valuable when there are benefits, which could be anything from improving decision making to increasing revenue or engagement among employees or customers.
A new survey from O’Reilly, “The State of Machine Learning Adoption in the Enterprise,” has identified some of the key learnings that derive from deploying machine learning technology and where companies should focus as they begin their journey. The research found that some of the most pressing enterprise machine learning topics include machine learning roles within organisations, bias and fairness, privacy and data protection, machine learning models and metrics for success.
By exploring how these areas and technologies are being implemented by both companies just starting out and those farther along on their machine learning journey, we can better understand where barriers for adoption exist and what’s ahead for this powerful technology.
Machine learning job titles
Many titles that involve machine learning in their job description are fairly new. Most enterprises will use the known title, “data scientist,” as someone who specialises in machine learning and statistics, according to the research. The “data engineer” job title differs in the way that their specific role is specialised in building and maintaining infrastructure and data pipelines. The experienced organisations also tend to use the title of “research scientists” to designate individuals that are tasked with developing complex algorithms. New job titles in the industry include “machine learning,” “deep learning engineers” and "data ops," which are exclusive to engineers with a concentration in building, deploying and managing machine learning models in production.
Overall, job titles specific to machine learning are already being widely used at organisations with extensive machine learning experience. Data scientist is most typically used (81 per cent), followed by machine learning engineer (39 per cent) and finally the newly-dubbed deep learning engineer (20 per cent). The emergence of new machine learning-specific titles indicates that organisations are getting serious about deploying the technology.
Bias and fairness
Companies who adopt machine learning are also factoring in considerations such as privacy, security, fairness and bias. We found that awareness of bias and fairness increases as organisations become more sophisticated with their machine learning activities. More than half (54 per cent) of advanced machine learning adopters who have already deployed the technology reported checking for fairness and bias, whereas only 40 per cent of early adopters check for these elements.
Attentiveness to bias and fairness is an increasing theme as organisations become more sophisticated with their machine learning activities. While there’s no explicit list to systematically address the particular issues pertaining to fairness, transparency and accountability, the good news is that machine learning adopters have begun to take the initial steps to be mindful of these matters.
Privacy & data protection
The European Union’s (EU) General Data Protection Regulation (GDPR) mandate raises critical concerns that organisations deploying machine learning models must keep in mind. There are recent developments in privacy-preserving analytic methods for business intelligence, analytics and machine learning, where an enterprise can incorporate innovative techniques and tools like differential privacy, homomorphic encryption, federated learning, hardware enclaves and more. In order to take a proactive standpoint on privacy and data protection and comply with GDPR, machine learning and analytics practitioners need to closely follow the new regulations.
More than half (53 per cent) of companies with extensive experience in machine learning check for privacy concerns. Fortunately, this number will only increase with GDPR’s “privacy-by-design” directive, which requires that privacy be taken into account throughout the entire engineering process – not just as an afterthought.
Machine learning models
Organisations that are just beginning to explore machine learning tend to rely on external consultants. Those that already have advanced machine learning strategies in place depend more heavily on internal data science teams. Some companies are modifying the processes they’ve used in software development, such as Agile methodology and Kanban, to build data products.
In the early stages of machine learning adoption, nearly a third of organisations surveyed practice no specific methodology and furthermore, organisations that are more sophisticated in machine learning aren’t using Agile as a default approach. There still remains a lot to be studied and discovered about how organisations are building their machine learning models.
Metrics for success
To make machine learning models valuable, they need to be customised to a company’s specific business metrics that are used to measure overall success or failure, which can include customer retention rate, conversion rate, market share or any other metrics to determine how well a business is doing. These metrics for success illustrate how machine learning adoption can introduce challenges for project management teams and leadership that diverge from their regular standard roles and practices.
Early adoption organisations that are just beginning to use machine learning place more weight on product managers or executives to determine particular criteria for project success. In more advanced organisations, leadership shifts toward data science leads. Advanced businesses are more likely to have data science teams in place to ensure priorities are determined and implemented for project success by means of key metrics that include business metrics, machine learning and statistical metrics and measures of bias and fairness.
It’s interesting to see how more sophisticated companies differ from those that are still in the early adoption and exploration phases of machine learning. As the survey results indicate, we can expect to see more specialised machine learning-specific titles emerge in the coming years, as well as more dedicated internal teams to build models and implement machine learning success metrics. Additionally, data privacy and protection will continue to be a high-priority topic of discussion around machine learning, as GDPR is in full swing. It’s clear there’s still a lot to learn when it comes to enterprise machine learning, but it’s encouraging that more organisations are starting to embrace the business value it can provide.
Ben Lorica, Chief Data Scientist, O'Reilly Media
Image Credit: Shutterstock/Mopic