Future of work machine learning and middle level jobs 112618
1 like78 views
The essay discusses how AI and machine learning (ML) are poised to transform middle-level jobs by simplifying their deployment through automated ML systems, which lower technical barriers and reduce reliance on data scientists. Companies like LinkedIn and Adidas showcase successful implementations of these systems, leading to improved operational efficiencies, enhanced customer engagement, and strategic decision-making. As automated ML becomes more prevalent, firms may shift from traditional predictive maintenance to more innovative applications involving design and resource management.
Future of work machine learning and middle level jobs 112618
- 1. Future of Work. How the Adoption of AI and Machine Learning may Increase Middle-Level Jobs Robert B. Cohen, November 26, 2018 This essay presents a new framework to analyze the impact of AI and ML on work. Its premise is that AI and ML have already been adopted in many firms. Now, efforts are underway to simplify the next stage of adoption by removing the complex requirement to create well-formulated algorithms. This innovation is automating the deployment of ML ecosystems. Early adopters report substantial gains in new revenues, additional efficiencies in operations and a changed mindset for employees. One example of the latter is LinkedIn’s efforts to establish a “culture of data,” where data serves as the foundation for corporate strategy and data analytics-based operations. This essay contends that by lifting earlier roadblocks to adoption, growth of ML and AI systems will increase, greater attention will be paid to obtaining and structuring data resources, and more ML systems can be applied to evaluating strategic and financial decisions. These automated systems incorporate expert knowledge. This knowledge includes: how to create usable data; how to represent data in a way that identifies the relationships within the data – called a knowledge representation; and incorporate expert understanding of the possible relationships in the data for a specific industry or domain. If AutoML systems succeed, they will lower the cost of deploying and using ML considerably. They will also obviate the need to build large groups of data analysts and data scientists – this is the technical skill barrier many firms must now address before they can deploy ML and AI systems. Many incumbent and startup vendors, including Amazon, Microsoft, Salesforce, SAP, SparkCognition and tazi.ai, have already deployed systems for customers. These automated systems are full-featured ML systems that incorporate industry-specific algorithms. End-users can easily train AutoML tools to model different processes, machines or services. These AutoML systems remove the complexity of using ML. This simplifies the operation and management of ML ecosystems. By making AutoML available as Software- as-a-Service (SaaS), firms that consume it can use middle-level employees to manage and operate these systems. We believe these middle-level jobs can be filled by employees with “domain” expertise, experience in a specific industry, and some technical knowledge. We have spoken with firms, including Bank of American, that have adopted new patterns of training to build the skills in technical expertise by levels of skills. We believe this pattern of up-skilling is more common in companies with a focus on data analysis. These firms hire new employees with basic software skills and train them in new software skills. This gradually improves their technical sophistication. We expect the initial deployments of AutoML systems to focus on software and data analysis tools to create manage and run applications built with containers and microservices, such as Salesforce’s Einstein or Amazon’s SageMaker. Many of the tools will be packages like Splunk to design and code highly scalable, machine learning applications. They will also include data visualization and reporting tools such as Tableau, Hortonworks, and QlikView that convert complex information into formats that are easy to understand. Other tools, like Recurrent Neural Networks (RNNs) or ConvNet will permit the
- 2. AutoML systems to add or access algorithms available through libraries, packages, APIs, to a suitable model-decision tree neural network. As AutoML systems evolve, we expect they will include more sophisticated software and tools to enhance these systems’ data analytic capabilities. This will consist of software and tools to ingest Big Data and to index and analyze data. It is also likely to encompass ML platforms for production models and software to help algorithms run faster, such as Apache Spark, Hadoop and MapReduce. In addition, we expect AutoML systems will incorporate data query and data processing tools, such as SQL. As these tools evolve, they will execute more complex queries, and provide users with easier and faster access to important data. As these systems are implemented more widely, they may also shift the way firms use ML and AI. Rather than focusing on prediction maintenance systems and maintaining the quality of products and services, firms could emphasize using ML to support generative design software that explores new designs for possible changes in configuration, the use of alternative materials or designs, and cost limits. Firms might also focus on strategic concerns, such as making better strategic decisions, controlling inventories better, energy consumption, and the use of costly materials. Evidence of AutoML offerings by Cloud Service Providers and Other Vendors Using Salesforce’s Einstein, Adidas shifted most of its business to focus on its online, or “creator customer”. It developed a premium-like service to connect buyers, get their feedback and use it to reshape the brand. According to an interview with Salesforce, in 2017, this increased online sales by as much as 60% in 2017, helped push Americas sales up 80%, and supported Adidas’ efforts to create 60 new shoe designs a year. With the feedback it receives from customers, Adidas used AI to make its “systems smarter and more integrated” to meet a future of “high density development”. It also used AI to identify drivers of demand, so Adidas could better forecast what products it sends to stores. Adidas believes its use of AI is helping it become more competitive than rivals such as Nike. Other AutoML systems provide similar benefits. SparkCognition’s Darwin automates the complete data science process. For industries such as oil and gas it has “optimized, highly accurate models that provide business insights, and adapt as your organization changes and scales.” Darwin automates the complete data science process. It “optimized, highly accurate models that provide business insights, and adapt as your organization changes and scales.” Darwin addresses “the biggest data science challenges: the hiring and training of personnel, the scalability of data science, and the need to minimize operational costs.” “Darwin-generated models helped process engineers predict workover, rod change, and cleaning operations needs in 12 out of 17 wells across the field with 70-80% accuracy.” This saved millions of dollars in repairs. Darwin’s models also optimized the output from each well. This improved revenues. Darwin also addresses operational costs. One oilfield customer avoided the need to build a large staff of data analysts and data scientists. By optimizing the output from each well, Darwin, improved revenues. Google Cloud AutoML has helped Urban Outfitters develop a sophisticated system of product attributes to provide customers with more chance to choose exactly the item they want. Urban Outfitters has used Cloud AutoML “to automate the product attribution process by recognizing nuanced product
- 3. characteristics like patterns and neckline styles”. This improves Urban Outfitters’ customers’ search, discovery and recommendation efforts. Several cloud service providers offer AutoML software-based services. They include: • Amazon’s SageMaker • Google Cloud AutoML • IBM Watson ML • Microsoft’s Azure ML • Salesforce’s Einstein Auto ML • SAP’s Leonardo • SparkCognition’s Darwin • tazi.ai’s ML Solutions • Seldon.ai’s SeldonCore • Oracle’s Adaptive Intelligence applications • Seebo • Maana Knowledge Platform